Back in 2008 Fujitsu has one of the best performing 10Gbps Switches. We were building 40 Gbps packet sniffers at Google (4x 10 Gbps NICs) and needed switches that could do things like mirror traffic across ports at line rate. Fujitsu was way ahead of the pack. I always wondered what held them back from building a meaningful networking business in the US.
> I always wondered what held them back from building a meaningful networking business in the US.
It was just timing. Their networking gear was built on some very impressive ASICs which became commodity chips a few years later before they could grow. Arista, Juniper, et al ate their lunch using other vendors’ IP. IIRC it was Broadcom and Fulcrum that released the chips that killed them.
>We were building 40 Gbps packet sniffers at Google (4x 10 Gbps NICs) and needed switches that could do things like mirror traffic across ports at line rate
I'm sure there were good reaasons for this, but man this just sounds bad. Like that's the spec I think NSA must give to their contractors outfitting 611 Folsom Street's Room 641A
These tools are very common in non-nefarious ways for troubleshooting networking, and while vendors set up to serve this space solely as troubleshooting tools, I've also built my own that can sit on a network link and monitor for problems.
In my case it was for mobile wireless signaling traffic (all the coordination for creating a mobile internet connection, handing the connection off between towers, etc), and I'd credit it as one of the reasons you're mobile internet connection is so stable. When LTE first came out, myself and many others solved all sorts of bugs in the equipment and protocols by using or building these sorts of tools.
Port mirroring is the network engineer's equivalent of using breakpoints in a debugger for a programmer. It allows you to inspect what happens on the wire to get a clue for what's wrong.
It does sound bad the way OP described it, but packet or HTTP mirroring is a standard feature of A/B testing and blue-green deployments of a very critical code. Mirror traffic between version 1 and 2, then compare response body, response time and return response A to the end users.
I used to operate several dozen 10G/25G/40G/100G taps which fed a series of tools in the DC at a regional healthcare back in the day.
80% of the relevant outcome was typically to get a true-to-the-wire sniffer capture (switch mirrors won't always mirror 100% of packets for various reasons) for troubleshooting performance of whatever the complaint of the day from the server team was.
19% was for feeding a security monitoring systems which looked for abnormal flow patterns to let us know a server was compromised.
1% was for the call recording system compliance requirement for the emergency department.
0% was because I was a cool superspy tasked by the government to siphon info to them or trying to sell medical records on the black market or something. I mean, you can try to something nefarious with such tools... but one could say the same about a generic server, SAN, application, etc as well. People are just used to understanding what those would typically be used for so they don't assume it must be for the scary thing they've heard about.
That said, it doesn't rule it out either. But again, the concern shouldn't be sourcing from their usage of normal infrastructure tools it should be sourcing from... well, all of the user analytics Google very publicly does directly in the server.
It's super common and very useful for security systems to work like this. Instead of putting something in-line that might choke on a burst of traffic you put it off to the side and send it a firehose of packets it may or may not be able to handle. If it falls over, no problem.
There are companies that mirror their traffic to storage for a few days to be able to look back and troubleshoot issues. ALL their traffic, proactively.
For US federal government, "Office of Management and Budget" says to store 72 hours of packet capture, since 2021 (partially in response to SolarWinds attack). (M-21-31)
The XG series switches? Force10 sold them until Dell bought Force10 and continued to sell them (although the firmware changed a lot along the way). They were, AFAIK, pretty popular in the US under the Dell brand.
My home network is XG2000 10G core and a few XG0224 1G/10G dist. Very nice switches.
Ubiquity proved you can bust into the market from the bottom end if you want. Nothing was stopping them from doing the same but like many companies they prefer to focus on how much money they can siphon from the Fortune 1000 so I can't say I feel sorry for them.
I'm not sure why that's relevent, particularly since Ubiquity's approach wouldn't have remotely competed with Fujitsus custom ASICs if they were actual contemporaries, but you clearly have an agenda you're in love with, so...ok, sure.
AFAIK those sold relatively well, considering the TCO of both sides (NICs were very expensive and there were like a dozen companies vying for position once upon a time). They were also interesting because they were cut-through. CX4 was noticeably lower latency than SFP+. So a very fast, but feature limited architecture which is probably why other players like Cisco and then Arista (aka Broadcom) took over.
We have alliances with Japan and S Korea which predispose us to believing we might be able to get a clean supply chain from them if a new Cold War starts. Out other options don’t even pass a giggle test and that’s the state of the world until some manufacturing is brought back onshore for strategic reasons.
The same way Congress has been subsidizing Boeing and its competitors since WW2 to make sure we can build things that fly if another war ever starts. Or with drones, that we can build ones that will be flown BY US instead of defecting the moment they get in range of the other side.
> The same way Congress has been subsidizing Boeing and its competitors
Funny you should mention that. Boeing deliberately buys 35% of a 787's airframe from Japan (along with other equipment for the 787 and other models).
That's a deliberate play: we buy a lot of parts from Japanese companies, your airlines buy mostly Boeing airplanes. Even today despite some diversification Boeing has over 70% marketshare in Japan.
If you ever wondered why Airbus setup factories in the USA that's a similar deal. The US Gov and Mil want to ensure factories and workforce on home soil should the need arise. In some unrecorded meetings somewhere strong hints were no doubt dropped that it would be in Airbus' best interest to play ball. They obliged.
That said the USA is the largest supplier of parts for Airbus airliners so there's also a benefit to having aerospace manufacturing and R&D here.
For a single CPU:
844 GB/s memory bandwidth (12 channels of DDR5 RDIMM, 8800MT/s). 4.3-6 TFLOPS.
Basically comparable to a modern (although not top-end) GPU, so good for HPC & AI workloads. (though only 2 CPUs per node, with GPUs 4-8 is more common)
For anyone not familiar with Fujitsu's CPUs, worth noting their overall long history with HPC, and more recently, ARM. FugakuNEXT is "NEXT" because the current Fugaku (#1 on TOP500 for a time, and still in the top 10) also used an ARM CPU from Fujitsu.
Benchmarks will be interesting to see but I doubt the cores alone will be able to reach beyond ~500GB/s of BW, and even if they match that figure it will be great.
Man this looks amazing, more competition is always welcome! The issue for Ai inference is always firmware/software support. Up until recently, that rocm hardware became usable, let alone optimized. Nvidia had a big start, now everyone is trying to catch up
Press releases are sadly like that. Fujitsu's main page has it in the first line:
> FUJITSU-MONAKA is a Next-Gen Arm-based processor, set for release in 2027, designed to address the challenges of next-generation data centers with its unmatched performance and power efficiency.
Fun fact, Fujitsu used to run their own fabs in Japan until they were sold to UMC and now they're another UMC fab although they are stuck at like the 20nm node. So this is likely made at JASM.
>The Phase 2 factory was under construction as of January 2025, next to the Phase 1 facility, and is expected to be completed by 2027.[3] Initially planned for manufacturing semiconductors using a 6 nm process, 3 nm was incorporated into production plans for 2028 in 2026
> Achieving world-class AI inference performance through Japan-developed 2nm 3D-stacked CPU and server integrated, developed, and manufactured in Japan
The compute part must have been made in Taiwan because it uses TSMC N2. TSMC N2 is only available in Taiwan right now due to a law restricting the latest node to Taiwan for 1-2 years.
All the component chips are made at TSMC, in various processes.
"Developed in Japan" means the same thing like "developed in USA" means for NVIDIA or AMD or Intel. You can design chips anywhere on the planet, as long as TSMC is willing to provide you the required documentation and EDA libraries (which it does only if you already are or they believe that you will be a big customer).
Doesn't really make sense to me. The whole "sovereign" thing is to basically say you won't be screwed if you go to war with a country tomorrow because you are dependent on their supply chain. Nothing about this changes that IMO. They're still getting the chips fabbed somewhere else and run the same risks.
True, but besides wars there are other dangers, e.g. USA does not declare war when they apply "sanctions" randomly, under the legal theory that they can dictate how anything that includes some piece of US origin may be used by its owner.
Since that theory began to be used, anyone outside US has become anxious to become "sovereign", i.e. to stop using anything about which a foreign state can claim to have rights.
It's a nuclear weapons thing. Basically you want your entire nuclear weapons supply chain infrastructure to be in your own country. These days that also includes super computers to do simulations on what happens with your aging asrenal (just look at how many Top 500 supercomputers are at national labs like Oak Ridge, LLNL, or Sandia). Japan doesn't technically have nuclear weapons, but they're rumored to be "a screwdriver's turn away", or practically within a few weeks of having a weapon. That posture itself requires even more simulation just to be sure you can make good on that.
The AI stuff is basically "well sure, these are really good at distributed matrix math, why not?" hoping to get better volumes and make the program cheaper internally.
And they're working on getting chips fabbed in Japan too now that IBM sold their 2nm process to Rapidus, but that's probably successor to these. Maybe 2028? In the meantime there's still more nuclear simulations to run than there is compute to run them.
« nuclear weapons » — They're science fiction, smoke blown around huge quantities of explosives still being produced at WW2 levels.
Michael Palmer, in his book Hiroshima revisited, doesn't go as far as to rule out “nukes” entirely, but damage assessment, medical evidence, witness reports and measurements show that no such thing as claimed by the US and Japanese governments happened in Hiroshima and Nagasaki. Instead, there were other ingredients to the atrocities, like the usual napalm, so-called “pumpkin bombs” (the iconic Nagasaki model), flash bombs and coloured smoke (“special effects”), and mustard gas (to generate “radiation sickness”), which is a gel that can linger on and emit poison for weeks (depending on the weather).
Napalm and mustard gas are 100 % consistent with the medical record, which the author is professionally qualified to assess, citing academic literature (which is scarse in case of napalm, probably on account of its heavy use by the US against East Asian peoples in the “defense of democracy”).
The damage pattern is extensive rather than intensive (that is the military expert assessment), just as seen in other napalmed cities in Japan. Miraculous survivors at the alleged “ground zero” (no such thing can really be derived from the damage pattern) preclude any generation of heat drastically higher than what is attainable by conventional means (“victims were vaporized”). So do slender objects like antennas and trees (which blossomed again next spring, then interpreted as a “miracle”). The few stone buildings there were can be seen standing out among the remains of flimsy huts that were blown and burnt.
And of course there was no disruption in settlement, the cities were built up and inhabited again.
« simulations on what happens with your aging asrenal » — Swarms of red herrings to keep the myth alive. Surely other uses of supercomputers are conceivable, even for non-disclosed purposes.
Strangely, while the non-existence of such weapons is undoubtedly good news, there are many “nuke fanboys” (no girls, though – this is a male thing) who will get upset and even angry at this, as if the “atomic bomb” were some kind of god to venerate, when it really only is lies and propaganda that we've been fed when we were kids.
To be clear, this is not to dispute nuclear energy, just the claim that it may be used to obtain explosions, and even extraordinary ones.
> Doesn't really make sense to me. The whole "sovereign" thing is to basically say you won't be screwed if you go to war with a country tomorrow because you are dependent on their supply chain.
I don't think it is that basic. If anything it's probably more about having protection from trade and sanction threats, and insulation from shortages, supply chain problems.
> Nothing about this changes that IMO. They're still getting the chips fabbed somewhere else and run the same risks.
It certainly changes it. Chip design is of enormous importance and the technology leaders there are concentrated in the USA.
Silicon manufacturing is a different thing. Japan is also trying to bring up sovereign leading(ish) edge manufacturing. Obviously you keep peeling the onion on everything though - to do digital logic they need EDA tools and quite possibly IP blocks for PCI and DDR and so on. To do manufacturing they need scanners from ASML and a process design from IBM and so on... It doesn't mean that because they can't do everything from the ground up that it's pointless. At some point they need to import coal and iron ore and bauxite.
Indeed, it is extremely clear: by not saying "manufactured in Japan", the words "developed in Japan" tell you that they drew up the designs for the CPU in Japan, and then sent those designs off to a fab outside Japan for actual manufacturing. Then they bring those CPUs back to Japan, so that the server can be manufactured in Japan even if the server components are not.
That manufacturing step may change in the future, but right now that CPU is not "Made in Japan", it's only "Designed in Japan".
They did a presentation at Hot Chips, with all these details.
The core die is made with TSMC-N2P, while the SRAM die and the peripheral die are made with TSMC-N5.
Their CPU is designed in Japan, like the AMD CPUs or the Intel CPUs are designed in USA (the latest Panther Lake Intel CPUs have returned for fabrication to USA in 2026, but they also cost double than the corresponding Arrow Lake models made at TSMC in 2025, and their GPUs are still made at TSMC).
The US administration created at least 90% of what we now call technology sovereignty concerns. Technology sovereignty is front and center in this announcement. Before the president started making wildly inappropriate remarks about grabbing territory from allies and other previously unthinkable utterances, almost everyone was happy to buy technology from US hardware and software vendors. Complacency about the state of democracy and rule of law in the US was the norm.
Why does this matter? Markets outside the US account for about half of many US technology vendors' revenue. Who is going to speak up first?
While others have made good points, your view is also valid. The messaging nowadays is more toward sovereignty even if the core direction hasn't fundamentally changed significantly.
I do find it funny how the harder some folks push for an outcome, the more they bring about the opposite effect.
I mean the US had this sense of confidence that they were at the top of the power structure, it was widely accepted and overall that position and deal seemed to benefit many others even if not to the same scale. For example with Europe, the US got favorable trade terms in exchange for defense.
Then the Trump administration came in and started pushing the narrative of MAGA, to almost be professional victims. Winner of the world lottery and yet still not happy. And so instead of making America Great again, they are on the fast track of being cut out of favorable global trade. They over played their hand. And once others end up being self sufficient again, it is very difficult to get back in the good books.
In trying to avoid being pushed around, they are now starting to see they could be pushed around.
« For example with Europe, the US got favorable trade terms in exchange for defense. » — It is occupation under the guise of defense, always has been. Keeping Germany divided for 35 years (when it could have become neutral and free in the early 50s), eager not to release control but to expand east instead, exploiting disunity, instigating the Yugoslavia wars of the 90s, more recently the Ukraine war (which goes back more than a decade in preparation).
That's what you get when you put a clown in power. I am not american so I don't care but the whole world noticed the bullying and acted accordingly. For example, US learned the hard way they can't bring down Huawei, tariff china or slow their frontier AI. Now Europe with Mistral and Japan with Moneka, both pointing in the same direction. Big middle finger if I've seen one
there are many things to blame POTUS for. This isn't one of them... most major fabs were already outside the US to begin with and the US chipmaking industry (well, really Intel) was really struggling even years ago.
I don't see the connection. No matter where US chips are manufactured, it's not the location of the fab that causes technology sovereignty concerns. Those concerns come from questions about the stability and rationality of our leadership: can my jet fighter get nerfed if I wanted to defend Greenland from an invasion? Are my spies being spied on by US technology?
Those questions didn't used to be significant factors.
I'd argue that the consolidation of the chip industry in Taiwan has had the most impact here. If/When China does move to take over Taiwan the global chip supply will be devastated and countries are well in their right to want to diversify from that. If anything, POTUS statements just also solidified the point that like Taiwan, they should not rely on USA or any other country for that matter. Hence why so many countries are pushing for some level of self sustainment.
Chip fabs are a red herring. They haven't been controlled by the US, and give the US no leverage over nations that used to be unequivocal allies. Those places are where technology sovereignty is a hot button issue. They can't count on the US, or US technology companies. You say Taiwan shouldn't rely on the US. But of all places where technology sovereignty is at issue, Taiwan has the least risk of being traded away in a corrupt real estate deal.
Yes the lack of transparency on ARM architecture and the obfuscation of the RAM capacity with just the slot information is annoying. Not sure if it is deliberate to sell this at a premium to "mainframe" customers or whether its just their marketing folk not being on the ball. Fascinated by the fact Japan has a 2nm process node but I suspect that too is marketing. Perhaps the 2nm portion of the chip is done at TSMC and the rest of the package done in Japan or something like that. It's not like there's a plethora of 2nm Fabs around the planet to choose from.
It looks to me like it is an AI optimized HPC CPU and while HPC CPUs are actually quite fast, they are not particularly well suited for transformer based neural networks.
It's weird that they don't give out any of the interesting numbers like number of memory channels, how much SRAM they have (CPUs tend to have more of it) or what their expected performance is going to be.
They actually do not in that article. I had to look elsewhere to find that out. Weird, one would expect that a big announcement like this would probably need to be vetted against licensing agreements with Arm that would probably want that fact to be prominently highlighted in press releases.
Yeah, searched the page for common architectures (x86,arm and Sparc since it was Fujitsu) since it'd seem silly to use a totally novel architecture but didn't find anything until noticing SVE2.
Fujitsu has a long history of designing their own micro but using a standard ISA for HPC. Used to be SPARC, now it is ARM. IIRC Fujitsu were the main architects behind the SVE ARM extension.
No it is not at all Neoverse V2, which is obsolete and has much lower performance.
It is a custom Armv9.3-A design (same ISA like the Arm C1 CPUs from the flagship smartphones of 2026), but it has double-width execution units for SVE2 (i.e. 256-bit width, vs. 128-bit for the other Arm CPUs) and it has some ISA extensions for AI/ML, e.g. instructions for inference with FP8 (BF16 is already supported by the standard Arm ISA).
Despite the fact that Armv9.3-A may include SME (scalable matrix extension, like in the Apple CPUs and in the Arm C1 CPUs), Fujitsu did not mention SME, so I assume that they did not implement it and they rely on their enhanced SVE2 (which is not surprising, while the origin of Arm SME is at Apple, the origin of Arm SVE is at Fujitsu).
Thanks kinda how Japan works. Inventing something new is hard, but taking something that's a cool idea and making it actually usable is well within their wheelhouse.
I'm sure there's plenty of other stuff invented in Japan, but that's really beside the point.
The industry is now mature enough that nobody really wants the headache of a new architecture at this point in time, Risc-V got a pass for being a grassroots movement growing out off FPGA's since the IP situation was more or less clear or even predatory with all existing architectures, Power, Sparc,SuperH,68k,etc are buried for good reasons.
PPC could have any number of merits, the 2 nail in the coffin imho is that the community/implementations seems to have held on to a big-endian tradition whilst the mass of users/developers firmly shifted to little-endian when Apple went to x86 and that IBM didn't push it properly open in the 2005-2008 timeframe when PS3/Xbox360 were still relevant (OpenPower seems to have been created in 2013, after the creation of Risc-V and PS4/XboxOne going to x86).
Dunno, I think Risc-V's simplicity gave it the initial popularity rush rather than being from a certain institution (the institution perhaps gave it visibility, but it does have own merits and no idea if other archs from the time had those?), don't remembere exactly but iirc it was possible to implement with a ridiculously small amount of FPGA gates. Conversly heard some rumblings that it had parts t hat were unsuitable for scaling up but it seems to have gained enough momentum for people to engineer past stuff like that.
I had this idea that Fujitsu was one of those Japanese conglomerates that had operations in all kinds of markets, from electronics to heavy earth moving equipment to air conditioners to locomotives. But that doesn't appear to be the case so I'm wondering which company I'm confusing them with.
Fujitsu was a telco equipment spinoff from Fuji Electric, so Fujitsu do/did make a lot of things that would be in a phone exchange building over its history, from air conditioners to terminals to actual mainframes. Just not as everything as Mitsubishi, which was a do-all procurement company for the new government when the Tokugawa stepped down.
About 25 years back here in Australia they were advertising Fujitsu air-conditioners, they had the promise of some like "They will get the air down to 17c in all conditions."
About a month later there was a summer heat wave and I went to a friends house who had just bought one. It was wild walking in from a 45c day into, as advertised, a 17c house. Had to give it to them, they lived up to the claim.
That they come from the telecom/server space, it makes sense they had a good eye for aircon engineering.
Sun invented SPARC, Fujitsu just happened to be the last producer since they built mainframes on it. Reading it seems like they plan to stop producing Sparc processors and machines in a near timeframe so that makes this announcement less surprising.
For many years the Fujitsu SPARC implementations were much faster than the Sun SPARC implementations.
During the decade 1995-2005, until the 64-bit AMD Opteron servers with Linux offered an alternative that crushed any kind of SPARC CPU with Solaris by a much higher performance combined with a much lower price, the SPARC Sun or Fujitsu servers dominated the market for servers used to host the CAD/EDA design tools used in electronics engineering, for the design of integrated circuits or electronic equipment. This means that all the EDA software from vendors like Mentor, Cadence, Synopsis etc. was available only for Solaris. That kind of software could not be run on Windows, due to the 32-bit memory limit. During that time, whoever got Fujitsu servers instead of the sluggish Sun servers, was very lucky.
It was an issue of fundamentally differing directions.
In the mid-2000's, Sun decided to take SPARC towards designs with many small SMT cores. In the era of single-core processors, the UltraSPARC T1 had 8 cores x 4 threads per core. This was at the same time Intel released the Pentium 4 with hyper-threading, so it was an industry trend.
This of course works great for very specific applications, particularly considering efficiency, but is awful for others. Scientific computation was especially bad because the T1 had only one FPU for 8 cores.
Fujitsu's SPARC64 didn't go in this direction, and stayed with a conventional design (2 way SMT at most). Sun realized this and started to also sell the Fujitsu SPARC64 for customers who couldn't use the thread level parallelism, an arrangement that lasted until the end.
The idea of lots of slow cores is still a thing today: Intel's Sierra Forest Xeon is 144 E-cores.
1 FPU for 8 cores is wild, but then again they were selling webservers.
Still, iirc FPU's were silicon heavy back in those days and it'd be interesting to know how far ahead the foundries Sun and Fujitsu were, maybe it was simply a factor of being too far behind in the foundry race that left Sun with few options.
Intel's E-Cores still are functionally complete for most parts though (excl Avx512?)? Todays limits seems to be memory bandwidth and power and I guess many of the 144 core customers are in it for virtualization and servers?
Likely none? I believe they just don't give graphics out to non-x86 servers. If you mean GPU as in SIMD processor, I believe this is supposed to have it in forms of massive SIMD capabilities on CPU itself.
x86 servers are like that. PC/AT memory map have the range for VRAM, but RS232C is optional, so you can't just have the RS232C and not VGA on an x86 as its debug/recovery interface.
Platform designers can make up their own memory maps, so non-x86 ones can just have UART I/O buffer and hardware interface wired up in whatever address you want and forgo VGA.
I have a 32bit SPARC machine on a shelf I got from a junk hardware shop, it has a DB25 for console. It's ok.
Not mentioning the architecture is rather strange. Being Fujitsu I'd expect SPARC, but apparently not. Maybe we're reaching a point where some circles either don't care or just defaults to ARM?
Obviously they can just ignore the license in the future and continue development out of a local branch, but it's also a bit disingenuous to speak of "sovereign infrastructure" and then use licensed processor design.
It think we are now heading towards the final long term ISA setup. x86 and ARM at the top, ARM in the middle and RISC-V for smaller embedded machines that don't need the ARM toolchain.
According to a Feb press release, they're fabbed by TSMC:
> For FUJITSU-MONAKA, the 2nm
semiconductors will tape out next year. Due to this, Rapidus would be unable to meet the
deadline, so TSMC is handling the manufacturing.
Their CPUs are faster at inference than many GPUs. These are server CPUs with 144 cores and wide memory interface.
Fujitsu has added some ISA extensions for AI inference to the standard Arm ISA, e.g. instructions for inference with FP8.
Even their previous CPU generation was for some time the champion in energy efficiency, with better performance per watt than the NVIDIA GPUs, until a newer generation of NVIDIA GPUs has leapfrogged them.
There are good chances that this new Fujitsu CPU might be again for some time the CPU with the best energy efficiency, but it remains to be seen how it compares with the recent GPUs.
At the very top of the article:
"Achieving world-class AI inference performance through Japan-developed 2nm 3D-stacked CPU and server integrated, developed, and manufactured in Japan"
"Japan-developed CPU..." plus the server itself (parts, power supplies, case, etc.) integrated, developed and manufactured in Japan. It sounds like they're weasel-wording it a bit, but I don't think the CPU itself is manufactured in Japan, only "developed."
Could someone knowledgeable please explain why an AI-focused CPU is superior to a GPU-based solution with an ARM CPU driving the high-level operations, please, especially when the RAM isn't on-chip?
I know that for GPU's the model weights have to be transferred over the bus initially, but that only has to occur once for inference use cases, so is the Fujitsu system more about training scenarios? Or is the focus more about efficiency, as these are ARM-based cores with AI additions?
Honestly I think this is an HPC CPU that has been AI-washed. Then you might ask why not use GPUs for HPC and I think the simple answer is that Fujitsu just doesn't want to take on the effort of building a GPU.
I fail to see how they will take a significant market share or even break even on this venture.
It's an overcrowded market. Far better would be to focus on semiconductor supply chain, which Japan already supplies some elements, to sell to fabs.
Because the bottleneck is the fabs. If a new 2 nm fab came online today, it would immediately sell all its capacity to 2030 no problem, without Fujitsu trying this gambit.
It's not 'sovereign', the architecture is not designed in Japan, the silicon is not fabbed in Japan. This whole thing is sideways.
Previous generation of this was a monstrosity based on a custom SPARC with ARM front end grafted on top, supporting the version of SPARC extensions they created backported to ARM as a new extension.
So there's a good chance that it's still an optimized half-SPARC inside.
Every developed country is going to push for their own homemade chips. Japan used to make Sparc CPU's back in the day. Probably many others I'm not aware of.
Every developed country except Australia, perhaps. This government is atrocious. The record for highest number of bureaucrats per capita in the world is owned by Australia. All they know how to do is tax people. All the wealthy people have left, or are leaving.
> Oh god you're one of those. People who complain about taxes have historically been on the wrong side of arguments.
Oh you're one of those. The arrogance is astounding. You never been a country which overly taxes it's citizens and squanders the gains on bureaucracy and monetary black holes? Grow up, your ideological blind spots are making you look like a fool.
> CapeTheory on Dec 14, 2024
> Fujitsu seems to be weirdly stubborn about global commercialisation of their CPUs. I know of at least a couple of HPC customers who wanted to build systems based on A64FX and the answer was basically "...nah we're good".
This attitude is seen in a lot of Japanese tech companies and their products over the past multiple decades, and it really irritates me. This, or the root cause of this, has to be one of major reasons why Japanese economy had stagnated, if not the key reason. There had been just so many things created, launched, and ... vanished in the wind.
Yes, AI is that important and every big actor should develop their own, avoiding any strongarming to slowdown and comply to "save the world" by certain bully we all know
Fujitsu MONAKA Server will be
made broadly available from
November 2026 to data center
operators, enterprises, and
the academic and HPC sectors
in Japan and Europe, as well
as to the defense sector,
contributing to national
security.
So, not in America. A sign of the geopolitical times? If so, a very unusual one to come out of Japan.
One problem we’re going to have with AI hardware is coming up with a standard set of specifications that are comparable. I don’t really care about the CPU GHz and the memory bandwidth, at least not directly. What I really want to know is how many tokens per second this will deliver, but that also depends on the model. We need a standard metric for that. Perhaps we agree on a specific open weight model (e.g. GLM 5.3 Flash or Qwen vWhatever) and then measure TPS on the hardware of interest.
I always heard of GB/s as most important number ... AI told me their 8800 MT/s on 8 Byte, 12 DDR5-Channels means 845 GB/s. A Nvidia RTX 4090 has 1008 GB/s. Nvidia B200 has 8000 GB/s.
Is this the right way of looking at it?
You have to load the model weights into VRAM over PCI-E (from RAM). So the (PCI-E) bandwidth strongly affects time to first token.
You have to run inference on the GPU by reading and writing to VRAM. So TFLOPS of the compute matters, and bandwidth to the VRAM (Always integrated with the GPU, rarely a bottleneck), and this strongly affects tokens/s
If you're doing training workloads or offloading to system RAM, it gets more complicated. (And mostly bound up trying to feed compute on time)
> So the (PCI-E) bandwidth strongly affects time to first token
On dedicated inference hardware I'd expect model weights to never leave the RAM, and you'd probably load them on startup before even starting to serve requests
The two made-in-Japan, premium grade, Fujitsu laptops I had to work with were by far the worse pieces of electronic I have ever used. Both exhibited identical defects: fans running continuously at maximum speed, and the batteries would completely die within hours of the devices being powered off. Windows or Linux. And don't talk about that tiny unresponsive trackpad to me. Never again.
The Fujitsu Lifebooks were legendary for their robustness back in the day, I'm genuinely sad you had that experience.
The only fujitsu laptops I used was back in 2011- it was a budget version for on-call and it was fine..
Sounds like the fans running 100% probably contributed significantly to the issue here. If I had to guess it was likely that the C-States were disabled in BIOS somehow so the CPU clock was running at full tilt the whole time, maybe combined with a bad thermal paste job.
I'm not here to defend Fujitsu but I've had really bad experiences with basically every major laptop brand (Dell: majorly bad coil whine and especially faulty soldered RAM, HP: keys vanishing from the keyboard and very weak hinges that break all the time, Apple: The GPU unsoldering itself and the butterfly keyboard shenanigans).
I don't really know any brand with a flawless track record sadly.
Before anyone brings up Thinkpads, they're trading on a reputation that hasn't been true for over a decade. If you have fond memories of a thinkpad it's most likely that you had it from before 2016, or it's "fine", but certainly not great.
I do recall that Fujitsu had a very... unique approach to exposing the Setup option in their firmware. It was just a UEFI boot entry. If that entry was removed, you couldn't get back into the firmware setup; you had to find the name of the setup EFI file and then run that from the EFI console.
Someone high up at Fujitsu is obsessed with making excessively miniaturized laptops. I once had an online fight on whether it's safe to run laptops closed, and realized that only Fujitsu manuals explicitly states that the machine uses keyboard top surfaces as a heatsink and running closed for extended periods is neither safe nor covered by warranty. IIRC Panasonic ones did say similar, but mostly towards noting that it ain't a server. Theirs were also one of the few that didn't seem to have been tested for Jack Bauer subset of MIL-STD-810 criteria, like thrown into a black SUV and flown somewhere in Middle East on a C-5 then C-130 then UH-60 and all the way back to the local soccer field.
Reaching further back, the Sylistic line was mostly quite good/durable, and I still use my Stylstic ST-4110 when I need to use an old scanner or control a CNC machine on my back deck when cutting tropical hardwoods.
Really miss the transflective display and wish that there were newer devices with such technology
I just have Fujitsu Lifebooks at Home since 2007. I think they are good, came with a bloat free Windows. I bought even one for my Mom over 10 years ago. I just changed once the HDD with a SSD. Now all my Lifebook run with Linux just flawless.
What year? The ones I had were the best. Very expensive, but service at the office & at home, very sturdy, very long battery life. That was somewhere early 2000s.
I'm kind of a Fujitsu fanboy. Since i came across the fact that a pretty modern xeon ecc ram nvme Workstation boards can be optimized to only draw ~10 Watt idle is just impressive, that is less than a gaming router. Even older ones like D3417-B12 were crazy effizient.
I also liked their Primergy Servers, but never got one for a reasonable price to test it out.
Another thing was the Futro series (e.g. S930) that could be used as opnsense firewall or low budget proxmox host.
Unfortunately it was always hard to obtain the high quality stuff as a consumer and years ago they sold their mainboard section to kontron. So no new Fujitsu Mainboards for now... What a pity.
I've been calling for high-multicore CPUs (at least 100 cores) with local memories for a quarter century now. No winners so far.
A Pentium 4 hit 3.8 GHz longer than that ago in 2004, so that's not special. And the estimated price will be $7,000-10,000, which isn't special either since that's about 10x more than it should be.
Hot take: GPUs disrupted the CPU industry to such a degree that CPUs never recovered, and like the k-shaped economy, the current status quo only serves a small fraction of customers. We can and should do better, but sadly we won't. Still, it's good that Fujitsu did this, for the competition if nothing else.
Unless they are used for multi-processing in classical UNIX fashion, or proper microkernels, most applications will hardly take advantage of them.
Managed languages runtimes are probably the ones that would be better equipped to take advantage of them, for distributed JIT, GC and asynchronous code.
The Connection Machine style with StarLisp.
Very few devs can write optimal multi-threaded code that explores the single digit count of cores on their laptops or phones already.
Ya functional programming is key, as it's equivalent to a spreadsheet or digital circuit. In other words, it removes the element of time and often dynamic state, which imperative programming generally fails to do.
IMHO most concurrency problems don't actually need async behavior, they need higher-order methods and models which behave deterministically by encapsulating async behavior internally.
Another maturing solution is to use SAT solvers to prove that all exceptional behavior and failure modes are handled. I'm hopeful that AI will help with that and allow us to exercise programs fully, rather than rely on unit tests and fuzzing.
Personally I think that the UNIX model of orchestrating small async programs that do one thing well is the only proven mainstream solution. Erlang and Go come really close, but unfortunately we need a hybrid of the two, which doesn't currently exist. The pattern for that is functional core, imperative shell. Which mimics the real world where business logic can be formerly proven correct or constrained by types and categories, then we wire up programs cookie cutter style. That avoids the use of monads (promises/futures in imperative languages), which are the main footguns. The closest language that does that is ClojureScript, whose runtime is analogous to suspending and resuming a coroutine or green thread that makes Lisp calls (although that's a poor fit and I'm sure I'm wrong about it).
But you're right that we don't currently have a language that can recruit multicore CPUs. I used MATLAB/GNU Octave in the past, but think that Julia probably has a brighter future since it can already run on GPU mostly unmodified. I would not try to do it with a mainstream language like C# or Python, or even PHP for that matter. Although some of PHP's multiprocessing metaphors are pretty solid, since they intentionally use processes instead of threads like in Ruby. There the problem is latency introduced by poor process models used by Microsoft and Apple, not something fundamental with spawning processes. Real-time Linux attempts to unify the kernel under one process model to provide deterministic timing, which is hard enough for regular Linux and probably out of reach of the big OS companies, because money can't buy everything.
I think what's most notable about this for me is the reduced environmental impact. The specifications list liquid-cooling as optional and it highlights the CPU as the main processor instead of a power hungry GPU. My takeaway is that the next generation of AI HPC's will focus more heavily on sustainable operations.
Their software is so poor that it caused probably the biggest miscarriage of justice in history. Well that and Fujitsu staff colluding with post office executives to put innocent people in prison rather than admit there was a problem
The 'Fujitsu' that made the shitty software was a British consulting company that was acquired by Fujitsu [1] to farm public contracts. This is made by the Japanese parent company.
In a very specific legal sense, it isn't, we are talking about two entirely separate legal entities (and ultimately groups of people) which have a relation of ownership, not directorship.
You're right, but the scandal is so upsetting that I personally cannot ever look at Fujitsu the same way even. Also, I think you're neglecting that a large part of why it became so bad was that Fujitsu claimed there was no issue with their software. Even if it were built by a company they acquired, the response to the developing situation was something that the parent company should have stepped in to oversee.
Conglomerates in general, and especially Japanese ones, are heterogeneous enough that those are basically two different companies. The British "Fujitsu" was an acquisition that I'm certain they now deeply, deeply regret.
Probably not, Japanese firms excel at hardware and generally build extremely poor software. This a hardware product, so consequently it is probably fairly good.
Japanese software don't get much recognition. Nintendo Switch(1 and 2) isn't running Linux, and it supports NVIDIA graphics stack. That shouldn't be possible unless Nintendo is actually capable of building a graphical operating system, which it is. Hardly anyone realizes this. There are also tons of (trashy)software embedded in Japanese hardware and none of those somehow mentally register.
Japan is also extremely weary of getting in the way of US. Software is one of few remaining consumer high tech domains that US leads. Challenging that without a good reason isn't a great idea.
I think Ruby would be a counterpoint - although it's the only one I know of.
I worked for a Japanese company although in a local office with local management. Through all those layers of protection I got the impression that they suffer from what all corporates suffer from - no software engineers that have made it to the top. Software companies run by people who have no clue.
In the US, for example, I think some people who have a clue or a partial clue have made it up the ladder in some places and therefore although everyone there has their horror stories of idiotic behavior it is probably somewhat better.
This is just my hypothesis, but I believe the language barrier is an issue. English is effectively a necessity for programming because real programming languages with ecoystems, error diagnostics, documentation, and useful articles/blogs/books are written in English. Japanese software developers know enough to get by, but they're still by-and-large insulated from the Anglosphere learning culture, where ideas and best practices rapidly propagate in the open. I'm not involved in hardware, but I'm guessing this is less of a barrier for hardware because hardware/documentation tends to be proprietary rather than open anyways, so whereas the Anglosphere gains a cumulative learning advantage from an open source + technical blogging culture in their fluent language, this does not manifest so much in hardware.
Honestly, no idea. In my career I have worked with a lot of Japanese firms and have always found their engineers to be excellent. I used to think it was a lack of consideration for UX, but that obviously can't be true when you see excellent examples of UX in hardware (e.g. Switch controls), even though its exceedingly rare on the software side.
If anyone on HN knows (there are several other commenters more deeply steeped in Japanese engineering culture), I would love to know why as well.
Japanese culture and work culture is very hierarchical/top down. They excel at waterfall but it's basically impossible to do agile in a Japanese office environment, and waterfall is suitable for hardware but terrible for building software.
＞Japanese culture and work culture is very hierarchical/top down.
This, and the 'rise or climb to the top' is more about seniority and social standing than anything else, especially at large Japanese conglomerates. The result us that there are almost none of the software competencies in senior management, and even if there were those would be software competencies from the 1970' and 1980's who are still leading while being 80 years old. There is also a 'filter' put in place at most organizations which are sufficiently large or diverse, such that middle-management is moved across multiple divisions every two or three years during their career. This is presented as a cross-training/cross-competency, but has an effect of preventing those with the most direct contact to the production staff from building enough localized social capital and systemic capability which would enable changes to the workflow not directly from the top. The middle stays in the middle, and the top gets to keep their seat well past any reasonable capability to adapt to the changing world of business.
This leads to the situation, where if leadership were changing with incoming capacity and points of view, there would be a motive and opportunity to admit to company errors; such as acquiring an international subsidiary which produced a faulty product that directly harmed people, because the newer leadership could say: "Sorry, that was a series of bad decisions, and as new leadership my role is to correct the course of business operations."
However, when the leadership is still the same clique that made the errors in either acquisition, oversight, response to the negative outcomes, then that "Sorry" statement would be a career ending move and result in massive social penalties beyond simply a forced retirement which would have lasting socio-economic impacts on the entire clique's family in every facet of business and later prospects. Combined with the strict social responsibility of a large enough scale of harm could lead to actual, not-figurative, suicide.
So, no, there's not going to be any 'mea culpa' for the direct actions of a foreign subsidiary, because those negative outcomes of those poor decisions are not in the direct social awareness in Japan of those surrounding the entrenched leadership and thus do not impact the daily social standing of that clique to the extent that a formal 'taking responsibility' would.
Instead, there will be very public announcements of improved capability which has the indirect and implied meaning of: "Engineering a way to make a better product will keep that from happening again" without ever directly addressing the existence of the faults. Social and economic uplift to counter any negative pressure in the local social awareness equals a success, not a failure.
In order for Japan to actually become successful at software in the near and medium time-frames, there will need to be a large shift in the work-life balance and internal social structure of new startups. This is beginning to happen in small scale, and to the extent that those newer 'brands' can avoid becoming subsidiaries of the larger established conglomerations will determine the slope to that success.
The groups which can empower and embrace a 'star' network topology inside and among organizations instead of 'waterfall' hierarchy will both; struggle to reach the social equity available through corporate-government-financial integration that is strongly entrenched in the established systems, and will outperform those entrenched systems in terms of real productivity through reliable delivery of improved capacity.
The socio-economic inertia of the giant keiretsu-gaisha are immense, but the geriatric cliff is already here and the nation and it's business cultures are at an 'adapt or cease operations' threshold. It's questionable whether any of the globally well know names will remain operational in their prior fields of production over the next 50 to 100 years without a significant shift in developing a way to build direct connections to capacity and applied skills at every tier of each company.
The 25 to 30 year old employees group in every main workforce is barely a fraction of what it used to be, and handling that small group as if it needs a 'filter' to keep the authority concentrated at the top is a significant risk for these organizations.
Those workers at the start of their career who are looking hard at the work-life balance aspect of employment have an inertia which is at odds with that of the organizational behemoths. Many of them are in a position to be drawn out of the crowded city, embrace a rural lifestyle and remote work without the centralized office and it's toxic overwork for underproduction culture. The rural municipalities are developing stronger incentives to attract this mobility, including relocation assistance and direct support with childcare and family building resources. Some prefectures and town offer up to several years of basic food and housing subsidies, low to zero cost fibre internet; next-generation FLET'S Hikari Cross or NTT's eqivalent 10Gbps in a HOUSE instead of a tiny apartment. The major central large conglomerates are going to find that the local in-office work pool will stagnate as the most capable and best informed no longer flow in directly from university and fill every available desk from a waiting list.
The "bigs" have insulated the ledger books with massive foreign investments, so they have the capability to stay 'in-the-game' for quite some time, so long as those investments continue to be productive, but the hollowing out of the main offices in Tokyo is pretty much guaranteed in the next two generations as rural Japan's liveability and economics for raising a family continue to grow, because the 80+ hour workweek labor staff in the city has no incentive to pair up and have kids, and the kids who are graduating in this and the next generation's rural school graduates are less and less likely to stay in the city after university if they want a family.
It's pretty much Japan's only way out of the population crisis that is still worsening. So, either they will adjust and become a productive, blended high tech semi-rural powerhouse without the high stress office culture that developed in postwar recovery, or they risk becoming a history lesson about over-concentration of inflexible management culture and loss of productivity.
The third option: succeed AND keep hyper-concentrated power is dependent on the types of moves that Fujitsu et al. are making today, find a way to automate more and more capacity via Machine Learning and robotics in the hopes of not actually needing an influx of skilled workforce for production.
If this 'hail Mary' play actually were to succeed it only cements a productive income for the shrinking population without resolving any of the causes of that shrinkage, i.e. the population still works 80+ hours a week in a socially disconnected state that in antithetic for raising healthy youthful progeny.
Taken to an 'in extremis' view: Within another 60 to 80 years, this might result in a well funded country of the elderly being waited on by robot butlers with until there is no one left to be waited on or fed the produce grown in automated fields.
I think that he most likely path will be a mixture of all three, in varying measure and with an array of outcomes. We live in interesting times indeed.
Back in 2008 Fujitsu has one of the best performing 10Gbps Switches. We were building 40 Gbps packet sniffers at Google (4x 10 Gbps NICs) and needed switches that could do things like mirror traffic across ports at line rate. Fujitsu was way ahead of the pack. I always wondered what held them back from building a meaningful networking business in the US.
> I always wondered what held them back from building a meaningful networking business in the US.
It was just timing. Their networking gear was built on some very impressive ASICs which became commodity chips a few years later before they could grow. Arista, Juniper, et al ate their lunch using other vendors’ IP. IIRC it was Broadcom and Fulcrum that released the chips that killed them.
Another broadcom related tragedy
>We were building 40 Gbps packet sniffers at Google (4x 10 Gbps NICs) and needed switches that could do things like mirror traffic across ports at line rate
I'm sure there were good reaasons for this, but man this just sounds bad. Like that's the spec I think NSA must give to their contractors outfitting 611 Folsom Street's Room 641A
These tools are very common in non-nefarious ways for troubleshooting networking, and while vendors set up to serve this space solely as troubleshooting tools, I've also built my own that can sit on a network link and monitor for problems.
In my case it was for mobile wireless signaling traffic (all the coordination for creating a mobile internet connection, handing the connection off between towers, etc), and I'd credit it as one of the reasons you're mobile internet connection is so stable. When LTE first came out, myself and many others solved all sorts of bugs in the equipment and protocols by using or building these sorts of tools.
Port mirroring is the network engineer's equivalent of using breakpoints in a debugger for a programmer. It allows you to inspect what happens on the wire to get a clue for what's wrong.
It does sound bad the way OP described it, but packet or HTTP mirroring is a standard feature of A/B testing and blue-green deployments of a very critical code. Mirror traffic between version 1 and 2, then compare response body, response time and return response A to the end users.
But the comment says ”we were building packet sniffers” … and needed mirroring as part of that.
I used to operate several dozen 10G/25G/40G/100G taps which fed a series of tools in the DC at a regional healthcare back in the day.
80% of the relevant outcome was typically to get a true-to-the-wire sniffer capture (switch mirrors won't always mirror 100% of packets for various reasons) for troubleshooting performance of whatever the complaint of the day from the server team was.
19% was for feeding a security monitoring systems which looked for abnormal flow patterns to let us know a server was compromised.
1% was for the call recording system compliance requirement for the emergency department.
0% was because I was a cool superspy tasked by the government to siphon info to them or trying to sell medical records on the black market or something. I mean, you can try to something nefarious with such tools... but one could say the same about a generic server, SAN, application, etc as well. People are just used to understanding what those would typically be used for so they don't assume it must be for the scary thing they've heard about.
That said, it doesn't rule it out either. But again, the concern shouldn't be sourcing from their usage of normal infrastructure tools it should be sourcing from... well, all of the user analytics Google very publicly does directly in the server.
I was under the impression that the superspy stuff usually happens more at the undersea-cable terminal or at satellite links anyway.
It's super common and very useful for security systems to work like this. Instead of putting something in-line that might choke on a burst of traffic you put it off to the side and send it a firehose of packets it may or may not be able to handle. If it falls over, no problem.
Common tools for any network engineer
Nah, packet mirroring is a standard networking feature and ideally every feature is line rate.
Indeed. Juniper had this over 20 years ago.
It came for free on hubs :)
the cost there was performance
There are companies that mirror their traffic to storage for a few days to be able to look back and troubleshoot issues. ALL their traffic, proactively.
For US federal government, "Office of Management and Budget" says to store 72 hours of packet capture, since 2021 (partially in response to SolarWinds attack). (M-21-31)
https://www.cisa.gov/sites/default/files/2023-02/TLP%20CLEAR...
I do that with very, very low traffic industrial automation systems.
The XG series switches? Force10 sold them until Dell bought Force10 and continued to sell them (although the firmware changed a lot along the way). They were, AFAIK, pretty popular in the US under the Dell brand.
My home network is XG2000 10G core and a few XG0224 1G/10G dist. Very nice switches.
Ubiquity proved you can bust into the market from the bottom end if you want. Nothing was stopping them from doing the same but like many companies they prefer to focus on how much money they can siphon from the Fortune 1000 so I can't say I feel sorry for them.
I'm not sure why that's relevent, particularly since Ubiquity's approach wouldn't have remotely competed with Fujitsus custom ASICs if they were actual contemporaries, but you clearly have an agenda you're in love with, so...ok, sure.
AFAIK those sold relatively well, considering the TCO of both sides (NICs were very expensive and there were like a dozen companies vying for position once upon a time). They were also interesting because they were cut-through. CX4 was noticeably lower latency than SFP+. So a very fast, but feature limited architecture which is probably why other players like Cisco and then Arista (aka Broadcom) took over.
The likely just didn't want another round of tantrums and troubles: https://en.wikipedia.org/wiki/1986_U.S.%E2%80%93Japan_Semico...
Having a us based vendor. Fujitsu is huge in Asia for ISPs and so is Mitsubishi and Sumitomo (yes they made or used to make network equipment too)
We have alliances with Japan and S Korea which predispose us to believing we might be able to get a clean supply chain from them if a new Cold War starts. Out other options don’t even pass a giggle test and that’s the state of the world until some manufacturing is brought back onshore for strategic reasons.
The same way Congress has been subsidizing Boeing and its competitors since WW2 to make sure we can build things that fly if another war ever starts. Or with drones, that we can build ones that will be flown BY US instead of defecting the moment they get in range of the other side.
> The same way Congress has been subsidizing Boeing and its competitors
Funny you should mention that. Boeing deliberately buys 35% of a 787's airframe from Japan (along with other equipment for the 787 and other models).
That's a deliberate play: we buy a lot of parts from Japanese companies, your airlines buy mostly Boeing airplanes. Even today despite some diversification Boeing has over 70% marketshare in Japan.
If you ever wondered why Airbus setup factories in the USA that's a similar deal. The US Gov and Mil want to ensure factories and workforce on home soil should the need arise. In some unrecorded meetings somewhere strong hints were no doubt dropped that it would be in Airbus' best interest to play ball. They obliged.
That said the USA is the largest supplier of parts for Airbus airliners so there's also a benefit to having aerospace manufacturing and R&D here.
Their servers were gorgeous (on the inside) as well. Clean designs, no clutter.
Seems like a story mirrored across Japanese companies.
More detailed presentations about the Monaka CPU:
wccftech summary: https://wccftech.com/fujitsus-monaka-chip-3d-stacks-2nm-cpu-...
2026: https://global.fujitsu/-/media/Project/Fujitsu/Fujitsu-HQ/te...
2023: https://global.fujitsu/-/media/Project/Fujitsu/Fujitsu-HQ/te...
FugakuNEXT, the supercomputer it will be used in, also some details about the next-gen Monaka-X: https://global.fujitsu/-/media/Project/Fujitsu/Fujitsu-HQ/te...
For a single CPU: 844 GB/s memory bandwidth (12 channels of DDR5 RDIMM, 8800MT/s). 4.3-6 TFLOPS.
Basically comparable to a modern (although not top-end) GPU, so good for HPC & AI workloads. (though only 2 CPUs per node, with GPUs 4-8 is more common)
For anyone not familiar with Fujitsu's CPUs, worth noting their overall long history with HPC, and more recently, ARM. FugakuNEXT is "NEXT" because the current Fugaku (#1 on TOP500 for a time, and still in the top 10) also used an ARM CPU from Fujitsu.
Benchmarks will be interesting to see but I doubt the cores alone will be able to reach beyond ~500GB/s of BW, and even if they match that figure it will be great.
Man this looks amazing, more competition is always welcome! The issue for Ai inference is always firmware/software support. Up until recently, that rocm hardware became usable, let alone optimized. Nvidia had a big start, now everyone is trying to catch up
Thanks for clarifying why this makes sense for inference.
A reminder of the great minds working tirelessly at this mega-corp: https://en.wikipedia.org/wiki/British_Post_Office_scandal
Fujitsu is such a large company that it is unlikely that there is any overlap between these two groups.
If anyone is curious and want to skip all the PR talk:
>Combined with SVE2 vector operations and software optimization,
It’s ARMv9.
It was funny looking over this just waiting to see what ISA it was. No surprise that it was ARM but you never know.
Nowadays if you have the source code the ISA is practically irrelevant.
I hoped it would be SPARC.
I was hoping risc-v. One of these days
I hoped it would be Alpha.
Why not 68000
I can't believe its not butter
I wonder how many cores they plan to cram in there, at least like 256 right?
144 per the article
Why did they bury the lede so much on what the ISA is?
Press releases are sadly like that. Fujitsu's main page has it in the first line:
> FUJITSU-MONAKA is a Next-Gen Arm-based processor, set for release in 2027, designed to address the challenges of next-generation data centers with its unmatched performance and power efficiency.
https://global.fujitsu/en-global/technology/research/fujitsu...
It is no secret it is Arm; MONAKA was announced in late 2024.
https://www.techpowerup.com/329761/fujitsu-previews-monaka-1...
Funny first thing I did was to search 'arm' and I got 1 result at the bottom of the page.
Hipe they'll be affordable
Fun fact, Fujitsu used to run their own fabs in Japan until they were sold to UMC and now they're another UMC fab although they are stuck at like the 20nm node. So this is likely made at JASM.
I think they'd say if the CPU was fabbed in Japan, instead there's some very "careful" wording:
>next-generation CPU, FUJITSU-MONAKA [1], designed and developed in Japan
>Fujitsu MONAKA Server, powered by the FUJITSU-MONAKA CPU, enhances sovereign capabilities through domestic manufacturing,
>FUJITSU-MONAKA utilizes a 3D-stacked architecture with a state-of-the-art 2nm process for the core and a 5nm process for the cache and I/O sections.
https://en.wikipedia.org/wiki/Japan_Advanced_Semiconductor_M...
>The Phase 2 factory was under construction as of January 2025, next to the Phase 1 facility, and is expected to be completed by 2027.[3] Initially planned for manufacturing semiconductors using a 6 nm process, 3 nm was incorporated into production plans for 2028 in 2026
Right below the headline it says:
> Achieving world-class AI inference performance through Japan-developed 2nm 3D-stacked CPU and server integrated, developed, and manufactured in Japan
Sounds pretty clear to me
The compute part must have been made in Taiwan because it uses TSMC N2. TSMC N2 is only available in Taiwan right now due to a law restricting the latest node to Taiwan for 1-2 years.
That's Taiwan's silicon shield which the taiwanese see as important. That's why japan has rapidus and the us has Intel
So, server manufactured/integrated in Japan; actual chip manufactured elsewhere (but "developed" in Japan, whatever that means)?
They made a detailed presentation at Hot Chips.
All the component chips are made at TSMC, in various processes.
"Developed in Japan" means the same thing like "developed in USA" means for NVIDIA or AMD or Intel. You can design chips anywhere on the planet, as long as TSMC is willing to provide you the required documentation and EDA libraries (which it does only if you already are or they believe that you will be a big customer).
Doesn't really make sense to me. The whole "sovereign" thing is to basically say you won't be screwed if you go to war with a country tomorrow because you are dependent on their supply chain. Nothing about this changes that IMO. They're still getting the chips fabbed somewhere else and run the same risks.
True, but besides wars there are other dangers, e.g. USA does not declare war when they apply "sanctions" randomly, under the legal theory that they can dictate how anything that includes some piece of US origin may be used by its owner.
Since that theory began to be used, anyone outside US has become anxious to become "sovereign", i.e. to stop using anything about which a foreign state can claim to have rights.
It's a nuclear weapons thing. Basically you want your entire nuclear weapons supply chain infrastructure to be in your own country. These days that also includes super computers to do simulations on what happens with your aging asrenal (just look at how many Top 500 supercomputers are at national labs like Oak Ridge, LLNL, or Sandia). Japan doesn't technically have nuclear weapons, but they're rumored to be "a screwdriver's turn away", or practically within a few weeks of having a weapon. That posture itself requires even more simulation just to be sure you can make good on that.
The AI stuff is basically "well sure, these are really good at distributed matrix math, why not?" hoping to get better volumes and make the program cheaper internally.
And they're working on getting chips fabbed in Japan too now that IBM sold their 2nm process to Rapidus, but that's probably successor to these. Maybe 2028? In the meantime there's still more nuclear simulations to run than there is compute to run them.
« nuclear weapons » — They're science fiction, smoke blown around huge quantities of explosives still being produced at WW2 levels.
Michael Palmer, in his book Hiroshima revisited, doesn't go as far as to rule out “nukes” entirely, but damage assessment, medical evidence, witness reports and measurements show that no such thing as claimed by the US and Japanese governments happened in Hiroshima and Nagasaki. Instead, there were other ingredients to the atrocities, like the usual napalm, so-called “pumpkin bombs” (the iconic Nagasaki model), flash bombs and coloured smoke (“special effects”), and mustard gas (to generate “radiation sickness”), which is a gel that can linger on and emit poison for weeks (depending on the weather).
Napalm and mustard gas are 100 % consistent with the medical record, which the author is professionally qualified to assess, citing academic literature (which is scarse in case of napalm, probably on account of its heavy use by the US against East Asian peoples in the “defense of democracy”).
The damage pattern is extensive rather than intensive (that is the military expert assessment), just as seen in other napalmed cities in Japan. Miraculous survivors at the alleged “ground zero” (no such thing can really be derived from the damage pattern) preclude any generation of heat drastically higher than what is attainable by conventional means (“victims were vaporized”). So do slender objects like antennas and trees (which blossomed again next spring, then interpreted as a “miracle”). The few stone buildings there were can be seen standing out among the remains of flimsy huts that were blown and burnt.
And of course there was no disruption in settlement, the cities were built up and inhabited again.
« simulations on what happens with your aging asrenal » — Swarms of red herrings to keep the myth alive. Surely other uses of supercomputers are conceivable, even for non-disclosed purposes.
Strangely, while the non-existence of such weapons is undoubtedly good news, there are many “nuke fanboys” (no girls, though – this is a male thing) who will get upset and even angry at this, as if the “atomic bomb” were some kind of god to venerate, when it really only is lies and propaganda that we've been fed when we were kids.
To be clear, this is not to dispute nuclear energy, just the claim that it may be used to obtain explosions, and even extraordinary ones.
It means it's not beating B300 in FP4 TOPS.
Isn't the supply chain on advanced semiconductors ludicrously extended?
Wafers rare Earth's, specialized equipment, etc?
> Doesn't really make sense to me. The whole "sovereign" thing is to basically say you won't be screwed if you go to war with a country tomorrow because you are dependent on their supply chain.
I don't think it is that basic. If anything it's probably more about having protection from trade and sanction threats, and insulation from shortages, supply chain problems.
> Nothing about this changes that IMO. They're still getting the chips fabbed somewhere else and run the same risks.
It certainly changes it. Chip design is of enormous importance and the technology leaders there are concentrated in the USA.
Silicon manufacturing is a different thing. Japan is also trying to bring up sovereign leading(ish) edge manufacturing. Obviously you keep peeling the onion on everything though - to do digital logic they need EDA tools and quite possibly IP blocks for PCI and DDR and so on. To do manufacturing they need scanners from ASML and a process design from IBM and so on... It doesn't mean that because they can't do everything from the ground up that it's pointless. At some point they need to import coal and iron ore and bauxite.
depends on whether you read it as
((2nm 3d stacked CPU) and server) (integrated, developed, and manufactured in Japan)
or
(2nm 3d stacked CPU) and (server integrated, developed, and manufactured in Japan)
Indeed, it is extremely clear: by not saying "manufactured in Japan", the words "developed in Japan" tell you that they drew up the designs for the CPU in Japan, and then sent those designs off to a fab outside Japan for actual manufacturing. Then they bring those CPUs back to Japan, so that the server can be manufactured in Japan even if the server components are not.
That manufacturing step may change in the future, but right now that CPU is not "Made in Japan", it's only "Designed in Japan".
They did a presentation at Hot Chips, with all these details.
The core die is made with TSMC-N2P, while the SRAM die and the peripheral die are made with TSMC-N5.
Their CPU is designed in Japan, like the AMD CPUs or the Intel CPUs are designed in USA (the latest Panther Lake Intel CPUs have returned for fabrication to USA in 2026, but they also cost double than the corresponding Arrow Lake models made at TSMC in 2025, and their GPUs are still made at TSMC).
When will they be using Rapidus?
Yeah I had a slight hope that they'd be a one of the first real Rapidus customers. but apparently not (for this one at least).
It's 2nm, so it has to be manufactured by TSMC.
I would assume they want to switch to rapidus asap once it comes online
The US administration created at least 90% of what we now call technology sovereignty concerns. Technology sovereignty is front and center in this announcement. Before the president started making wildly inappropriate remarks about grabbing territory from allies and other previously unthinkable utterances, almost everyone was happy to buy technology from US hardware and software vendors. Complacency about the state of democracy and rule of law in the US was the norm.
Why does this matter? Markets outside the US account for about half of many US technology vendors' revenue. Who is going to speak up first?
What? Fujitsu has been making supercomputers in Japan for decades.
Its ARM predecessor started in 2014: https://en.wikipedia.org/wiki/Fugaku_(supercomputer)
And before that they were the using SPARC64: https://en.wikipedia.org/wiki/SPARC64_V
While others have made good points, your view is also valid. The messaging nowadays is more toward sovereignty even if the core direction hasn't fundamentally changed significantly.
I do find it funny how the harder some folks push for an outcome, the more they bring about the opposite effect.
I mean the US had this sense of confidence that they were at the top of the power structure, it was widely accepted and overall that position and deal seemed to benefit many others even if not to the same scale. For example with Europe, the US got favorable trade terms in exchange for defense.
Then the Trump administration came in and started pushing the narrative of MAGA, to almost be professional victims. Winner of the world lottery and yet still not happy. And so instead of making America Great again, they are on the fast track of being cut out of favorable global trade. They over played their hand. And once others end up being self sufficient again, it is very difficult to get back in the good books.
In trying to avoid being pushed around, they are now starting to see they could be pushed around.
« For example with Europe, the US got favorable trade terms in exchange for defense. » — It is occupation under the guise of defense, always has been. Keeping Germany divided for 35 years (when it could have become neutral and free in the early 50s), eager not to release control but to expand east instead, exploiting disunity, instigating the Yugoslavia wars of the 90s, more recently the Ukraine war (which goes back more than a decade in preparation).
That's what you get when you put a clown in power. I am not american so I don't care but the whole world noticed the bullying and acted accordingly. For example, US learned the hard way they can't bring down Huawei, tariff china or slow their frontier AI. Now Europe with Mistral and Japan with Moneka, both pointing in the same direction. Big middle finger if I've seen one
there are many things to blame POTUS for. This isn't one of them... most major fabs were already outside the US to begin with and the US chipmaking industry (well, really Intel) was really struggling even years ago.
I don't see the connection. No matter where US chips are manufactured, it's not the location of the fab that causes technology sovereignty concerns. Those concerns come from questions about the stability and rationality of our leadership: can my jet fighter get nerfed if I wanted to defend Greenland from an invasion? Are my spies being spied on by US technology?
Those questions didn't used to be significant factors.
I'd argue that the consolidation of the chip industry in Taiwan has had the most impact here. If/When China does move to take over Taiwan the global chip supply will be devastated and countries are well in their right to want to diversify from that. If anything, POTUS statements just also solidified the point that like Taiwan, they should not rely on USA or any other country for that matter. Hence why so many countries are pushing for some level of self sustainment.
Chip fabs are a red herring. They haven't been controlled by the US, and give the US no leverage over nations that used to be unequivocal allies. Those places are where technology sovereignty is a hot button issue. They can't count on the US, or US technology companies. You say Taiwan shouldn't rely on the US. But of all places where technology sovereignty is at issue, Taiwan has the least risk of being traded away in a corrupt real estate deal.
It's Arm based, although they barely mention that.
Yes the lack of transparency on ARM architecture and the obfuscation of the RAM capacity with just the slot information is annoying. Not sure if it is deliberate to sell this at a premium to "mainframe" customers or whether its just their marketing folk not being on the ball. Fascinated by the fact Japan has a 2nm process node but I suspect that too is marketing. Perhaps the 2nm portion of the chip is done at TSMC and the rest of the package done in Japan or something like that. It's not like there's a plethora of 2nm Fabs around the planet to choose from.
Reading of JASM (TSMC/Sony subsidiary) it seems they're planning on producing at least 3nm chips there in their expansion plans.
It looks to me like it is an AI optimized HPC CPU and while HPC CPUs are actually quite fast, they are not particularly well suited for transformer based neural networks.
It's weird that they don't give out any of the interesting numbers like number of memory channels, how much SRAM they have (CPUs tend to have more of it) or what their expected performance is going to be.
They did a detailed presentation at Hot Chips.
See e.g.:
https://chipsandcheese.com/p/hot-chips-2026-fujitsus-monaka-...
They have provided far more details than companies like Apple or Qualcomm.
Looks good for HPC but the AI performance is only good against other CPUs. Even Intel GPUs can beat the quoted numbers within that power envelope.
They actually do not in that article. I had to look elsewhere to find that out. Weird, one would expect that a big announcement like this would probably need to be vetted against licensing agreements with Arm that would probably want that fact to be prominently highlighted in press releases.
It is in the article - they cite "SVE2" support
Yeah, searched the page for common architectures (x86,arm and Sparc since it was Fujitsu) since it'd seem silly to use a totally novel architecture but didn't find anything until noticing SVE2.
Fujitsu has a long history of designing their own micro but using a standard ISA for HPC. Used to be SPARC, now it is ARM. IIRC Fujitsu were the main architects behind the SVE ARM extension.
If they didn't mention SVE2, I wouldn't have known either... that would make it arm9 at the earliest...
Quick Google result is it’s basically a Neoverse V2 system, so yeah, armv9.
Edit: it’s funny because this article from 2021 mentioned Fujisu precisely https://siliconangle.com/2021/03/30/arm-unveils-armv9-archit...
No it is not at all Neoverse V2, which is obsolete and has much lower performance.
It is a custom Armv9.3-A design (same ISA like the Arm C1 CPUs from the flagship smartphones of 2026), but it has double-width execution units for SVE2 (i.e. 256-bit width, vs. 128-bit for the other Arm CPUs) and it has some ISA extensions for AI/ML, e.g. instructions for inference with FP8 (BF16 is already supported by the standard Arm ISA).
Despite the fact that Armv9.3-A may include SME (scalable matrix extension, like in the Apple CPUs and in the Arm C1 CPUs), Fujitsu did not mention SME, so I assume that they did not implement it and they rely on their enhanced SVE2 (which is not surprising, while the origin of Arm SME is at Apple, the origin of Arm SVE is at Fujitsu).
Monaka-X will include SME2 but it didn't make the cut for first-gen Monaka
That was regarding the HPC Fugaku , this is next generation
But Arm is owned by SoftBank, a Japanese conglomerate.
(I think Japanese sovereignty is the main point of the article.)
Interesting to note that Fujitsu did this when the SPARC architecture was popular: they enhanced/developed a CPU for the same ABI.
Seems they have decided they don’t want to create their own architecture, but take the popular architecture and build on it.
Thanks kinda how Japan works. Inventing something new is hard, but taking something that's a cool idea and making it actually usable is well within their wheelhouse.
See: - Blue LEDs - Quartz Watches - Lithium-ion Batteries - Bidets
I'm sure there's plenty of other stuff invented in Japan, but that's really beside the point.
The industry is now mature enough that nobody really wants the headache of a new architecture at this point in time, Risc-V got a pass for being a grassroots movement growing out off FPGA's since the IP situation was more or less clear or even predatory with all existing architectures, Power, Sparc,SuperH,68k,etc are buried for good reasons.
POWER is still holding on just barely but I do suspect that IBM are looking to sunset it in the next few years.
Given that IBM just announced a dual-ISA mainframe with support for both z/arch and ARM, they might indeed be looking to sunset POWER.
Powers not dead yet! IBM is still clinging to it and I can't blame them, it's a pretty good ISA
Yep, I have said it for a good 20 years now.
POWER/PowerPC is what happens when you take a brilliant design to execute it poorly. x86 is when you take a poor design and execute is brilliantly.
PPC could have any number of merits, the 2 nail in the coffin imho is that the community/implementations seems to have held on to a big-endian tradition whilst the mass of users/developers firmly shifted to little-endian when Apple went to x86 and that IBM didn't push it properly open in the 2005-2008 timeframe when PS3/Xbox360 were still relevant (OpenPower seems to have been created in 2013, after the creation of Risc-V and PS4/XboxOne going to x86).
UC Berkeley developed RISC-V, so no doubt their influence within the computer science academia sphere had something to do with it.
Dunno, I think Risc-V's simplicity gave it the initial popularity rush rather than being from a certain institution (the institution perhaps gave it visibility, but it does have own merits and no idea if other archs from the time had those?), don't remembere exactly but iirc it was possible to implement with a ridiculously small amount of FPGA gates. Conversly heard some rumblings that it had parts t hat were unsuitable for scaling up but it seems to have gained enough momentum for people to engineer past stuff like that.
The 1980s called, it wants its stereotypes back.
I had this idea that Fujitsu was one of those Japanese conglomerates that had operations in all kinds of markets, from electronics to heavy earth moving equipment to air conditioners to locomotives. But that doesn't appear to be the case so I'm wondering which company I'm confusing them with.
Edit I think I was thinking of Mitsubishi.
Fujitsu was a telco equipment spinoff from Fuji Electric, so Fujitsu do/did make a lot of things that would be in a phone exchange building over its history, from air conditioners to terminals to actual mainframes. Just not as everything as Mitsubishi, which was a do-all procurement company for the new government when the Tokugawa stepped down.
About 25 years back here in Australia they were advertising Fujitsu air-conditioners, they had the promise of some like "They will get the air down to 17c in all conditions."
About a month later there was a summer heat wave and I went to a friends house who had just bought one. It was wild walking in from a 45c day into, as advertised, a 17c house. Had to give it to them, they lived up to the claim.
That they come from the telecom/server space, it makes sense they had a good eye for aircon engineering.
Probably Panasonic. They're like the Apple of the B2B world. Except they also make everything
Fujitsu is basically the Japanese IBM.
Sun invented SPARC, Fujitsu just happened to be the last producer since they built mainframes on it. Reading it seems like they plan to stop producing Sparc processors and machines in a near timeframe so that makes this announcement less surprising.
For many years the Fujitsu SPARC implementations were much faster than the Sun SPARC implementations.
During the decade 1995-2005, until the 64-bit AMD Opteron servers with Linux offered an alternative that crushed any kind of SPARC CPU with Solaris by a much higher performance combined with a much lower price, the SPARC Sun or Fujitsu servers dominated the market for servers used to host the CAD/EDA design tools used in electronics engineering, for the design of integrated circuits or electronic equipment. This means that all the EDA software from vendors like Mentor, Cadence, Synopsis etc. was available only for Solaris. That kind of software could not be run on Windows, due to the 32-bit memory limit. During that time, whoever got Fujitsu servers instead of the sluggish Sun servers, was very lucky.
It was an issue of fundamentally differing directions.
In the mid-2000's, Sun decided to take SPARC towards designs with many small SMT cores. In the era of single-core processors, the UltraSPARC T1 had 8 cores x 4 threads per core. This was at the same time Intel released the Pentium 4 with hyper-threading, so it was an industry trend.
This of course works great for very specific applications, particularly considering efficiency, but is awful for others. Scientific computation was especially bad because the T1 had only one FPU for 8 cores.
Fujitsu's SPARC64 didn't go in this direction, and stayed with a conventional design (2 way SMT at most). Sun realized this and started to also sell the Fujitsu SPARC64 for customers who couldn't use the thread level parallelism, an arrangement that lasted until the end.
The idea of lots of slow cores is still a thing today: Intel's Sierra Forest Xeon is 144 E-cores.
1 FPU for 8 cores is wild, but then again they were selling webservers.
Still, iirc FPU's were silicon heavy back in those days and it'd be interesting to know how far ahead the foundries Sun and Fujitsu were, maybe it was simply a factor of being too far behind in the foundry race that left Sun with few options.
Intel's E-Cores still are functionally complete for most parts though (excl Avx512?)? Todays limits seems to be memory bandwidth and power and I guess many of the 144 core customers are in it for virtualization and servers?
Thanks. Looked for that in the article and couldn't find it.
What are the GPU capabilities?
Likely none? I believe they just don't give graphics out to non-x86 servers. If you mean GPU as in SIMD processor, I believe this is supposed to have it in forms of massive SIMD capabilities on CPU itself.
These things feel like, if they have any video output, it would be a single VGA port for if you REALLY need to get a signal from it.
x86 servers are like that. PC/AT memory map have the range for VRAM, but RS232C is optional, so you can't just have the RS232C and not VGA on an x86 as its debug/recovery interface.
Platform designers can make up their own memory maps, so non-x86 ones can just have UART I/O buffer and hardware interface wired up in whatever address you want and forgo VGA.
I have a 32bit SPARC machine on a shelf I got from a junk hardware shop, it has a DB25 for console. It's ok.
Thanks I was, of course, looking and very curious what their tool chain story will be.
Not mentioning the architecture is rather strange. Being Fujitsu I'd expect SPARC, but apparently not. Maybe we're reaching a point where some circles either don't care or just defaults to ARM?
Obviously they can just ignore the license in the future and continue development out of a local branch, but it's also a bit disingenuous to speak of "sovereign infrastructure" and then use licensed processor design.
It think we are now heading towards the final long term ISA setup. x86 and ARM at the top, ARM in the middle and RISC-V for smaller embedded machines that don't need the ARM toolchain.
Fujitsu dropped SPARC for HPC ~seven years ago with their A64fx. That's where ARM SVE comes from in the first place.
I feel old now.
Yes but it's their own design, not licensing Arm cores.
Named after monaka, a traditional Japanese confection of sweet red bean paste sandwiched between two wafers.
https://en.wikipedia.org/wiki/Monaka
It's not terribly popular even in Japan, since it's dry, bland, fragile and has to be assembled on the spot or it goes soggy.
Good metaphor, then; do what we must, not what others want from us.
"sandwiched between two wafers"
Pun intended?
Almost certainly.
Where are the CPUs fabbed? They make a big deal about "sovereign" but is the CPU actually made in Japan, or do they rely on TSMC?
Also odd how big an emphasis they put on AI inference when they don't build the GPU?
According to a Feb press release, they're fabbed by TSMC:
> For FUJITSU-MONAKA, the 2nm semiconductors will tape out next year. Due to this, Rapidus would be unable to meet the deadline, so TSMC is handling the manufacturing.
Source: https://global.fujitsu/en-global/pr/news/2026/02/12-01?utm_s...
According to Rapidus webpage, they are explicitly starting mass production of their 2nm in 2027.
Source: https://www.rapidus.inc/en/iim/?utm_source=chatgpt.com
Their CPUs are faster at inference than many GPUs. These are server CPUs with 144 cores and wide memory interface.
Fujitsu has added some ISA extensions for AI inference to the standard Arm ISA, e.g. instructions for inference with FP8.
Even their previous CPU generation was for some time the champion in energy efficiency, with better performance per watt than the NVIDIA GPUs, until a newer generation of NVIDIA GPUs has leapfrogged them.
There are good chances that this new Fujitsu CPU might be again for some time the CPU with the best energy efficiency, but it remains to be seen how it compares with the recent GPUs.
> put on AI inference when they don't build the GPU?
GPU in AI world is essentially a set of specific matrix calculations that this CPU supports on hardware-level. Thought memory speed seems low.
The article says that it’s manufactured in Japan.
where exactly does it say that? I don't consider "designed and developed in Japan" to be the same as manufactured there
At the very top of the article: "Achieving world-class AI inference performance through Japan-developed 2nm 3D-stacked CPU and server integrated, developed, and manufactured in Japan"
Pretty sure it's worded in a way that it applies only to the server, so it should be read as two distinct things:
- Japan-developed 2nm 3D-stacked CPU
- server integrated, developed, and manufactured in Japan
"Japan-developed CPU..." plus the server itself (parts, power supplies, case, etc.) integrated, developed and manufactured in Japan. It sounds like they're weasel-wording it a bit, but I don't think the CPU itself is manufactured in Japan, only "developed."
Yep. I'd say the CPU is manufactured outside, and the rest of the server in Japan.
Combining info from https://news.ycombinator.com/item?id=49740907 and https://news.ycombinator.com/item?id=49740901, it seems that they're relying on TSMC for the first batch and are planning to move production to JASM once that's ready?
Ever seen the phrase "diffused in Taiwan, assembled in Malaysia"?
Yeah but TSMC have fab on japan soil
so its as sovereign as you can get
Not for 2nm
I think the repeat of Fujitsu, Fujitsu- MONAKA is truly annoying. Is there just this press release, or other better sources?
Wow a corporate TLD in a wild, never seen this outside of the erstwhile domains.google
Funny they have money to buy TLD or even develop such a fancy CPU, but don't have money to compensate people affected by Post Office scandal.
there's quite a few
like:
- home.kpmg - global.honda
and probably more I forgot about
but as you can see, they're pretty terrible for replacing .com domains
Could someone knowledgeable please explain why an AI-focused CPU is superior to a GPU-based solution with an ARM CPU driving the high-level operations, please, especially when the RAM isn't on-chip?
I know that for GPU's the model weights have to be transferred over the bus initially, but that only has to occur once for inference use cases, so is the Fujitsu system more about training scenarios? Or is the focus more about efficiency, as these are ARM-based cores with AI additions?
Honestly I think this is an HPC CPU that has been AI-washed. Then you might ask why not use GPUs for HPC and I think the simple answer is that Fujitsu just doesn't want to take on the effort of building a GPU.
I fail to see how they will take a significant market share or even break even on this venture.
It's an overcrowded market. Far better would be to focus on semiconductor supply chain, which Japan already supplies some elements, to sell to fabs.
Because the bottleneck is the fabs. If a new 2 nm fab came online today, it would immediately sell all its capacity to 2030 no problem, without Fujitsu trying this gambit.
It's not 'sovereign', the architecture is not designed in Japan, the silicon is not fabbed in Japan. This whole thing is sideways.
>the architecture is not designed in Japan
The text says "next-generation CPU, FUJITSU-MONAKA, designed and developed in Japan".
It's ARM.
> I fail to see how they will take a significant market share or even break even on this venture.
Doesn't matter. You have to start somewhere and this is a good start.
It's a custom Fujitsu micro-architecture.
It's ARM with extensions.
Previous generation of this was a monstrosity based on a custom SPARC with ARM front end grafted on top, supporting the version of SPARC extensions they created backported to ARM as a new extension.
So there's a good chance that it's still an optimized half-SPARC inside.
Japan historically behaves as if they have nukes and need a totally sovereign HPC capability to support their stockpile.
Every developed country is going to push for their own homemade chips. Japan used to make Sparc CPU's back in the day. Probably many others I'm not aware of.
Even x86! Bandai WonderSwan ran on an NEC-made 80186 compatible CPU.
They made 6502 and Z80 clones as well, nintendo used them iirc
Every developed country except Australia, perhaps. This government is atrocious. The record for highest number of bureaucrats per capita in the world is owned by Australia. All they know how to do is tax people. All the wealthy people have left, or are leaving.
> The record for highest number of bureaucrats per capita in the world is owned by Australia
You have been convinced by fake news, please recheck this claim, its false.
> wealthy people
*rich people
> All they know how to do is tax people
Oh god you're one of those. People who complain about taxes have historically been on the wrong side of arguments.
> Oh god you're one of those. People who complain about taxes have historically been on the wrong side of arguments.
Oh you're one of those. The arrogance is astounding. You never been a country which overly taxes it's citizens and squanders the gains on bureaucracy and monetary black holes? Grow up, your ideological blind spots are making you look like a fool.
Here's the previous generation, two years ago: https://news.ycombinator.com/item?id=42415754
And here's more technical information about this generation: https://news.ycombinator.com/item?id=49443040
Monaka? Hope it's as sweet and satisfying as the dessert. Always good to see more domestic CPU efforts.
Yes, AI is that important and every big actor should develop their own, avoiding any strongarming to slowdown and comply to "save the world" by certain bully we all know
Kudos to Japan
One problem we’re going to have with AI hardware is coming up with a standard set of specifications that are comparable. I don’t really care about the CPU GHz and the memory bandwidth, at least not directly. What I really want to know is how many tokens per second this will deliver, but that also depends on the model. We need a standard metric for that. Perhaps we agree on a specific open weight model (e.g. GLM 5.3 Flash or Qwen vWhatever) and then measure TPS on the hardware of interest.
I feel like even TPS is becoming less of a good metric as we're seeing certain models handle similar problems while burning far fewer tokens.
> I don’t really care about the CPU GHz and the memory bandwidth ... What I really want to know is how many tokens per second this will deliver
There is a fairly direct link between the two numbers. You can predict the latter from former reasonably well
I always heard of GB/s as most important number ... AI told me their 8800 MT/s on 8 Byte, 12 DDR5-Channels means 845 GB/s. A Nvidia RTX 4090 has 1008 GB/s. Nvidia B200 has 8000 GB/s. Is this the right way of looking at it?
You have to load the model weights into VRAM over PCI-E (from RAM). So the (PCI-E) bandwidth strongly affects time to first token.
You have to run inference on the GPU by reading and writing to VRAM. So TFLOPS of the compute matters, and bandwidth to the VRAM (Always integrated with the GPU, rarely a bottleneck), and this strongly affects tokens/s
If you're doing training workloads or offloading to system RAM, it gets more complicated. (And mostly bound up trying to feed compute on time)
(Edits for clarity.)
> So the (PCI-E) bandwidth strongly affects time to first token
On dedicated inference hardware I'd expect model weights to never leave the RAM, and you'd probably load them on startup before even starting to serve requests
Would be cool to get ahold of one of these in the US.
I do find it odd that the 2U model has less storage capacity than the 1U on their chart. That doesn't make much sense to me.
The two made-in-Japan, premium grade, Fujitsu laptops I had to work with were by far the worse pieces of electronic I have ever used. Both exhibited identical defects: fans running continuously at maximum speed, and the batteries would completely die within hours of the devices being powered off. Windows or Linux. And don't talk about that tiny unresponsive trackpad to me. Never again.
The Fujitsu Lifebooks were legendary for their robustness back in the day, I'm genuinely sad you had that experience.
The only fujitsu laptops I used was back in 2011- it was a budget version for on-call and it was fine..
Sounds like the fans running 100% probably contributed significantly to the issue here. If I had to guess it was likely that the C-States were disabled in BIOS somehow so the CPU clock was running at full tilt the whole time, maybe combined with a bad thermal paste job.
I'm not here to defend Fujitsu but I've had really bad experiences with basically every major laptop brand (Dell: majorly bad coil whine and especially faulty soldered RAM, HP: keys vanishing from the keyboard and very weak hinges that break all the time, Apple: The GPU unsoldering itself and the butterfly keyboard shenanigans).
I don't really know any brand with a flawless track record sadly.
Before anyone brings up Thinkpads, they're trading on a reputation that hasn't been true for over a decade. If you have fond memories of a thinkpad it's most likely that you had it from before 2016, or it's "fine", but certainly not great.
I do recall that Fujitsu had a very... unique approach to exposing the Setup option in their firmware. It was just a UEFI boot entry. If that entry was removed, you couldn't get back into the firmware setup; you had to find the name of the setup EFI file and then run that from the EFI console.
Their consumer electronics arm is completely unrelated and also majority owned by Lenovo
I was about to say if they hadn't said it was a Fujitsu laptop I would have guessed it was a ThinkPad. Describes my experience with a P1 exactly.
Someone high up at Fujitsu is obsessed with making excessively miniaturized laptops. I once had an online fight on whether it's safe to run laptops closed, and realized that only Fujitsu manuals explicitly states that the machine uses keyboard top surfaces as a heatsink and running closed for extended periods is neither safe nor covered by warranty. IIRC Panasonic ones did say similar, but mostly towards noting that it ain't a server. Theirs were also one of the few that didn't seem to have been tested for Jack Bauer subset of MIL-STD-810 criteria, like thrown into a black SUV and flown somewhere in Middle East on a C-5 then C-130 then UH-60 and all the way back to the local soccer field.
Reaching further back, the Sylistic line was mostly quite good/durable, and I still use my Stylstic ST-4110 when I need to use an old scanner or control a CNC machine on my back deck when cutting tropical hardwoods.
Really miss the transflective display and wish that there were newer devices with such technology
I just have Fujitsu Lifebooks at Home since 2007. I think they are good, came with a bloat free Windows. I bought even one for my Mom over 10 years ago. I just changed once the HDD with a SSD. Now all my Lifebook run with Linux just flawless.
What year? The ones I had were the best. Very expensive, but service at the office & at home, very sturdy, very long battery life. That was somewhere early 2000s.
2012 and 2017 Celsius H. Very expensive and both maxed up but those fans always running even with limited usage made them the worse to use.
If they can’t build real enterprise sales and software support outside Japan, it’s just another cool Arm chip no one can actually buy
I'm kind of a Fujitsu fanboy. Since i came across the fact that a pretty modern xeon ecc ram nvme Workstation boards can be optimized to only draw ~10 Watt idle is just impressive, that is less than a gaming router. Even older ones like D3417-B12 were crazy effizient.
I also liked their Primergy Servers, but never got one for a reasonable price to test it out.
Another thing was the Futro series (e.g. S930) that could be used as opnsense firewall or low budget proxmox host.
Unfortunately it was always hard to obtain the high quality stuff as a consumer and years ago they sold their mainboard section to kontron. So no new Fujitsu Mainboards for now... What a pity.
I've been calling for high-multicore CPUs (at least 100 cores) with local memories for a quarter century now. No winners so far.
A Pentium 4 hit 3.8 GHz longer than that ago in 2004, so that's not special. And the estimated price will be $7,000-10,000, which isn't special either since that's about 10x more than it should be.
Hot take: GPUs disrupted the CPU industry to such a degree that CPUs never recovered, and like the k-shaped economy, the current status quo only serves a small fraction of customers. We can and should do better, but sadly we won't. Still, it's good that Fujitsu did this, for the competition if nothing else.
Unless they are used for multi-processing in classical UNIX fashion, or proper microkernels, most applications will hardly take advantage of them.
Managed languages runtimes are probably the ones that would be better equipped to take advantage of them, for distributed JIT, GC and asynchronous code.
The Connection Machine style with StarLisp.
Very few devs can write optimal multi-threaded code that explores the single digit count of cores on their laptops or phones already.
IMO we are still far from discovering a truly usable concurrency paradigm (if one even exists).
Maybe if designing software like digital circuits, which are more parallel in execution.
However one could argue that FPGAs are already that, and still we come back to a skill issue.
I am also not an expert, cannot make heads or tails about SIMD algorithms, and in what GPUs goes, only traditional shading languages.
Let alone algorithms that on top of that, should take optimal advantage of all available CPUs.
Ya functional programming is key, as it's equivalent to a spreadsheet or digital circuit. In other words, it removes the element of time and often dynamic state, which imperative programming generally fails to do.
IMHO most concurrency problems don't actually need async behavior, they need higher-order methods and models which behave deterministically by encapsulating async behavior internally.
Another maturing solution is to use SAT solvers to prove that all exceptional behavior and failure modes are handled. I'm hopeful that AI will help with that and allow us to exercise programs fully, rather than rely on unit tests and fuzzing.
Personally I think that the UNIX model of orchestrating small async programs that do one thing well is the only proven mainstream solution. Erlang and Go come really close, but unfortunately we need a hybrid of the two, which doesn't currently exist. The pattern for that is functional core, imperative shell. Which mimics the real world where business logic can be formerly proven correct or constrained by types and categories, then we wire up programs cookie cutter style. That avoids the use of monads (promises/futures in imperative languages), which are the main footguns. The closest language that does that is ClojureScript, whose runtime is analogous to suspending and resuming a coroutine or green thread that makes Lisp calls (although that's a poor fit and I'm sure I'm wrong about it).
But you're right that we don't currently have a language that can recruit multicore CPUs. I used MATLAB/GNU Octave in the past, but think that Julia probably has a brighter future since it can already run on GPU mostly unmodified. I would not try to do it with a mainstream language like C# or Python, or even PHP for that matter. Although some of PHP's multiprocessing metaphors are pretty solid, since they intentionally use processes instead of threads like in Ruby. There the problem is latency introduced by poor process models used by Microsoft and Apple, not something fundamental with spawning processes. Real-time Linux attempts to unify the kernel under one process model to provide deterministic timing, which is hard enough for regular Linux and probably out of reach of the big OS companies, because money can't buy everything.
Chapel tries to do a bit that language, however it looks they only care about HPC workloads, that are not everyday programming anyway.
Very nice overview, also agree with the points that you presented.
I think what's most notable about this for me is the reduced environmental impact. The specifications list liquid-cooling as optional and it highlights the CPU as the main processor instead of a power hungry GPU. My takeaway is that the next generation of AI HPC's will focus more heavily on sustainable operations.
Iam very curious as to how well it'll perform outside of Asia.
Performance would probably be the same regardless of where you run the CPU.
And if they'll sell it, seeing as they market it as 'sovereign AI'.
Anyone have idea the likely cost range?
> The FUJITSU-MONAKA CPU integrates Fujitsu's extensive expertise in processor ... development
Does Fujitsu have extensive expertise in processor development? I don't remember any notable Fujitsu processors.
Although, as irusensei notes, this seems to be an ARMv9 processor.
Japan technology is back ????
Never buy products of this dodgy company.
https://en.wikipedia.org/wiki/British_Post_Office_scandal
New CPU:
:D
For AI infrastructure:
:(
https://en.wikipedia.org/wiki/British_Post_Office_scandal
Their software is so poor that it caused probably the biggest miscarriage of justice in history. Well that and Fujitsu staff colluding with post office executives to put innocent people in prison rather than admit there was a problem
The 'Fujitsu' that made the shitty software was a British consulting company that was acquired by Fujitsu [1] to farm public contracts. This is made by the Japanese parent company.
[1] https://en.wikipedia.org/wiki/International_Computers_Limite...
It's seriously underselling ICL to call it a consulting company, though by the time of the Fujitsu acquisition that might be accurate.
Not trying to be obnoxious, but it's still Fujitsu.
In a very specific legal sense, it isn't, we are talking about two entirely separate legal entities (and ultimately groups of people) which have a relation of ownership, not directorship.
It is ultimately Fujitsu responsibility.
You're right, but the scandal is so upsetting that I personally cannot ever look at Fujitsu the same way even. Also, I think you're neglecting that a large part of why it became so bad was that Fujitsu claimed there was no issue with their software. Even if it were built by a company they acquired, the response to the developing situation was something that the parent company should have stepped in to oversee.
I'm surprised at the downvotes for this, does anyone think I'm misrepresenting the facts? I genuinely don't think I am.
I wouldn't trust that this CPU can do maths properly.
Will it lock up thousands of innocent self-employed post office contractors as part of its instruction set?
https://en.wikipedia.org/wiki/British_Post_Office_scandal
Conglomerates in general, and especially Japanese ones, are heterogeneous enough that those are basically two different companies. The British "Fujitsu" was an acquisition that I'm certain they now deeply, deeply regret.
(Which in no way excuses their utter evil.)
Probably not, Japanese firms excel at hardware and generally build extremely poor software. This a hardware product, so consequently it is probably fairly good.
What caused this?
Japanese software don't get much recognition. Nintendo Switch(1 and 2) isn't running Linux, and it supports NVIDIA graphics stack. That shouldn't be possible unless Nintendo is actually capable of building a graphical operating system, which it is. Hardly anyone realizes this. There are also tons of (trashy)software embedded in Japanese hardware and none of those somehow mentally register.
Japan is also extremely weary of getting in the way of US. Software is one of few remaining consumer high tech domains that US leads. Challenging that without a good reason isn't a great idea.
I think Ruby would be a counterpoint - although it's the only one I know of.
I worked for a Japanese company although in a local office with local management. Through all those layers of protection I got the impression that they suffer from what all corporates suffer from - no software engineers that have made it to the top. Software companies run by people who have no clue.
In the US, for example, I think some people who have a clue or a partial clue have made it up the ladder in some places and therefore although everyone there has their horror stories of idiotic behavior it is probably somewhat better.
msgpack, mold, and lld are a few other examples of exemplary software.
This is just my hypothesis, but I believe the language barrier is an issue. English is effectively a necessity for programming because real programming languages with ecoystems, error diagnostics, documentation, and useful articles/blogs/books are written in English. Japanese software developers know enough to get by, but they're still by-and-large insulated from the Anglosphere learning culture, where ideas and best practices rapidly propagate in the open. I'm not involved in hardware, but I'm guessing this is less of a barrier for hardware because hardware/documentation tends to be proprietary rather than open anyways, so whereas the Anglosphere gains a cumulative learning advantage from an open source + technical blogging culture in their fluent language, this does not manifest so much in hardware.
Honestly, no idea. In my career I have worked with a lot of Japanese firms and have always found their engineers to be excellent. I used to think it was a lack of consideration for UX, but that obviously can't be true when you see excellent examples of UX in hardware (e.g. Switch controls), even though its exceedingly rare on the software side.
If anyone on HN knows (there are several other commenters more deeply steeped in Japanese engineering culture), I would love to know why as well.
Japanese culture and work culture is very hierarchical/top down. They excel at waterfall but it's basically impossible to do agile in a Japanese office environment, and waterfall is suitable for hardware but terrible for building software.
＞Japanese culture and work culture is very hierarchical/top down.
This, and the 'rise or climb to the top' is more about seniority and social standing than anything else, especially at large Japanese conglomerates. The result us that there are almost none of the software competencies in senior management, and even if there were those would be software competencies from the 1970' and 1980's who are still leading while being 80 years old. There is also a 'filter' put in place at most organizations which are sufficiently large or diverse, such that middle-management is moved across multiple divisions every two or three years during their career. This is presented as a cross-training/cross-competency, but has an effect of preventing those with the most direct contact to the production staff from building enough localized social capital and systemic capability which would enable changes to the workflow not directly from the top. The middle stays in the middle, and the top gets to keep their seat well past any reasonable capability to adapt to the changing world of business.
This leads to the situation, where if leadership were changing with incoming capacity and points of view, there would be a motive and opportunity to admit to company errors; such as acquiring an international subsidiary which produced a faulty product that directly harmed people, because the newer leadership could say: "Sorry, that was a series of bad decisions, and as new leadership my role is to correct the course of business operations."
However, when the leadership is still the same clique that made the errors in either acquisition, oversight, response to the negative outcomes, then that "Sorry" statement would be a career ending move and result in massive social penalties beyond simply a forced retirement which would have lasting socio-economic impacts on the entire clique's family in every facet of business and later prospects. Combined with the strict social responsibility of a large enough scale of harm could lead to actual, not-figurative, suicide.
So, no, there's not going to be any 'mea culpa' for the direct actions of a foreign subsidiary, because those negative outcomes of those poor decisions are not in the direct social awareness in Japan of those surrounding the entrenched leadership and thus do not impact the daily social standing of that clique to the extent that a formal 'taking responsibility' would.
Instead, there will be very public announcements of improved capability which has the indirect and implied meaning of: "Engineering a way to make a better product will keep that from happening again" without ever directly addressing the existence of the faults. Social and economic uplift to counter any negative pressure in the local social awareness equals a success, not a failure.
In order for Japan to actually become successful at software in the near and medium time-frames, there will need to be a large shift in the work-life balance and internal social structure of new startups. This is beginning to happen in small scale, and to the extent that those newer 'brands' can avoid becoming subsidiaries of the larger established conglomerations will determine the slope to that success.
The groups which can empower and embrace a 'star' network topology inside and among organizations instead of 'waterfall' hierarchy will both; struggle to reach the social equity available through corporate-government-financial integration that is strongly entrenched in the established systems, and will outperform those entrenched systems in terms of real productivity through reliable delivery of improved capacity.
The socio-economic inertia of the giant keiretsu-gaisha are immense, but the geriatric cliff is already here and the nation and it's business cultures are at an 'adapt or cease operations' threshold. It's questionable whether any of the globally well know names will remain operational in their prior fields of production over the next 50 to 100 years without a significant shift in developing a way to build direct connections to capacity and applied skills at every tier of each company.
The 25 to 30 year old employees group in every main workforce is barely a fraction of what it used to be, and handling that small group as if it needs a 'filter' to keep the authority concentrated at the top is a significant risk for these organizations.
Those workers at the start of their career who are looking hard at the work-life balance aspect of employment have an inertia which is at odds with that of the organizational behemoths. Many of them are in a position to be drawn out of the crowded city, embrace a rural lifestyle and remote work without the centralized office and it's toxic overwork for underproduction culture. The rural municipalities are developing stronger incentives to attract this mobility, including relocation assistance and direct support with childcare and family building resources. Some prefectures and town offer up to several years of basic food and housing subsidies, low to zero cost fibre internet; next-generation FLET'S Hikari Cross or NTT's eqivalent 10Gbps in a HOUSE instead of a tiny apartment. The major central large conglomerates are going to find that the local in-office work pool will stagnate as the most capable and best informed no longer flow in directly from university and fill every available desk from a waiting list.
The "bigs" have insulated the ledger books with massive foreign investments, so they have the capability to stay 'in-the-game' for quite some time, so long as those investments continue to be productive, but the hollowing out of the main offices in Tokyo is pretty much guaranteed in the next two generations as rural Japan's liveability and economics for raising a family continue to grow, because the 80+ hour workweek labor staff in the city has no incentive to pair up and have kids, and the kids who are graduating in this and the next generation's rural school graduates are less and less likely to stay in the city after university if they want a family.
It's pretty much Japan's only way out of the population crisis that is still worsening. So, either they will adjust and become a productive, blended high tech semi-rural powerhouse without the high stress office culture that developed in postwar recovery, or they risk becoming a history lesson about over-concentration of inflexible management culture and loss of productivity.
The third option: succeed AND keep hyper-concentrated power is dependent on the types of moves that Fujitsu et al. are making today, find a way to automate more and more capacity via Machine Learning and robotics in the hopes of not actually needing an influx of skilled workforce for production.
If this 'hail Mary' play actually were to succeed it only cements a productive income for the shrinking population without resolving any of the causes of that shrinkage, i.e. the population still works 80+ hours a week in a socially disconnected state that in antithetic for raising healthy youthful progeny.
Taken to an 'in extremis' view: Within another 60 to 80 years, this might result in a well funded country of the elderly being waited on by robot butlers with until there is no one left to be waited on or fed the produce grown in automated fields.
I think that he most likely path will be a mixture of all three, in varying measure and with an array of outcomes. We live in interesting times indeed.