I'm puzzled by Espressif's naming here. We had the ESP32-S3, so "S31" sounds like "S3, variant 1," but this part doesn't really look like a simple S3 variant. And then there's an ESP32-E22, but no E21 or even a plain E2 anywhere.
It was because IA-64 was a completely different unrelated architecture that until AMD succeeded with K8 was "the plan" for both 64bit intel roadmap and the roadmap to kill off compatible vendors (AMD, VIA)
They claim that the chip has an "MMU". But unfortunately this doesn't seem to be a true RISC-V MMU (according to the Sv32 specification) integrated into the CPU core itself, but just a peripheral designed for memory mapped SPI flash and PSRAM. So as far as I understand there is no true process isolation with page faults and dynamic paging.
Maybe Espressif will notice that there are no RV32 chips with MMU so far (at least to my knowledge); we only have 32 bit MCUs or then only 64 bits for the CPUs. Something like Cortex-A7 is missing.
Interesting that they made a new chip with BLE+BR/EDR again. all the chips after the original ESP32 were BLE only.
Hope this chip has good low power options so we can use it in Bluetooth audio workloads.
Nah, ESP32's have had ethernet capability for a while and ESP-IDF supports it well. I've been using one I built for 5+ years now. Unfortunately RMII (ethernet phy) interface takes up a lot of the GPIO pins. This part looks like it'll remedy that issue.
There's two ESP32 boards that have been around for a while with PoE:
> I'm more hopeful for single-pair ethernet to gain momentum though!
I keep looking for a reasonably priced 10baseT to 10Base-T1L bridge... everything commercial seems too expensive (for me) and the two hobby designs [1] [2] I've seen are not orderable :(
But I'm seeing more commercial options lately, so that's hopeful.
Multidrop SPE isn't going to outperform newer CAN versions though. Somewhere in the sub-100Mb/s (e.g. 10-20Mb/s range) is the practical maximum speed of a multidrop bus at useful lengths, and that essentially applies equally to CAN or SPE. The only way to really get faster in a "multidrop-like" sense is with logically loop-like systems like ethercat and Fibre Channel where each network segment is point-to-point and the nodes are responsible for the routing.
On that note, why does the PoE capability often add such a big proportion of the price of various items? Is the technology really costly for some reason, or is it just more there's fairly low demand and people are still willing to pay?
PoE is not obvious to implement (take it from someone who has done it with a fair share of mistakes), uses more expensive components that normal ethernet, takes up more space on the board, makes passing emissions certification more complex, and is more prone to mistakes that ruin boards in the future, causing support/warranty issues. In other words, a bag of worms: not impossible to handle, but something you would rather avoid if possible.
I wouldn't call it "better", but the least-effort path among hobbyists and low end gear is often 12v or 24v sent over a pair with Gnd and a forgiving voltage regulator on the other end.
A full-module add-on in this power class is about $7 at 1,000 unit scale [0]. It would be around $3 with your own custom PCB design in terms of BoM addon at scale. That’s power only. Add another dollar or two for 10/100 PHY.
The trick is as others have said in what adding it to your design does in terms of complicating compliance design.
PoE power supplies need to be isolated (except in rare exceptions) and handle much higher voltages than common USB-C or wall wart power supplies.
They have to use a transformer and a more complex control strategy, not a simple buck regulator with an inductor. PoE inputs need to tolerate voltages several times higher than the highest USB-C voltages, so more expensive parts are used everywhere.
Any Ethernet (well, any RJ45 you expect in a home/office) has to have at least 1500V isolation from the RJ45 wire to anything metal that can be touched or is a connector on the device.
A PoE-only device with no electrical connectors besides the RJ45 can just use a very cheap RJ45 port with integrated magnetics and PoE allowance (tiny bit bigger wires and a center pin exposed, less than 50ct more than the cheapest RJ45 with integrated magnetics) and a cheap buck from 40~80V to e.g. 5V.
Oh, and a cheap bridge rectifier and some signaling resistors to take care of input polarity and signal to the source that we in fact want the approximately 50V that could hurt a device not made for it.
It sounds like the PoE spec was designed before the arrival of “IoT” type things like the esp32, raspberry pi’s, etc.
How much of the complexity is a “fundamental electrical engineering problem” and how much of it is just a spec written to solve a different set of problems?
Whenever you combine two things into one, the complexity and cost go up considerably. A regular coffee machine is pretty cheap. Add high pressure so it can make espresso and it gets considerably more expensive. Add milk so it can make cappuccino, again more complex and expensive. The same holds for electronics. Isolating power when it's alone is fairly straightforward. It gets considerably more tricky and hence more expensive the moment you want to place any kind of a meaningful data signal in its vicinity.
> You don't need long cables, just a local power source
Which means batteries that have to be replaced and maintained or cables... So ethernet with PoE or even better SPE (single pair Ethernet) with PoDL (power over data lines which is PoE for SPE) is the best from my point of view
I mean, if I just look at my house. There is just one ethernet outlet, but many power sockets. If I want to connect devices all over my house, the best way is to use wifi and usb power adapters. Not ethernet.
Both solutions require 1 cable per device, but the first solution would require only short and thin cables, and the second solution would require very long cables which I don't know even how to do properly without milling my walls.
Yep. Mains electricity is ubiquitous, highly interoperable, very reliable, very high power available per drop, can be outdoor capable, common standards, understandable by users, requiring no active components, with many on-call experts available who can come to fix problems or extend/alter connectivity. Mains power wall plates with inbuilt USB power outlets are even available at quite small cost if the look of the bigger plug and wiring is not appealing.
PoE is much fewer of those things. Difficult to recommend it these days with wifi being fast and reliable and so widely used. Certainly not for average residential user.
That's half the equation. The other half is the reliability and security of wifi, which is less than that of ethernet for people without physical access to my wall innards
On the other hand, _all_ the WiFi devices that I had at some point fell off the network, at least once. Including doorbells and cameras. While PoE devices just work.
Another point is that mains power in my area can go down periodically. My PoE switch is powered by a Li-Ion UPS and can provide power for about a day.
Can't you run a 5V supply from where your router is all the way to every god damn device in your house, and then pretend the wifi is also going through it? If you just want it to be inconvenient, there's no reason to let a lack of PoE stop you!
I don't understand what possesses these folks to continue making 2.4ghz devices. I understand there are use cases for low bandwidth, high range. But surely we've passed the point where that is more desirable to most than lower latency and high throughput, right?
Is what you described a truth for all IoT devices? If I have los of my AP, why do I need 2.4Ghz? Even so, what SNR do you truly need for this low bandwidth application? Where is the engineering here?
I have a unique position of having a data set over 8000 APs with 40k unique devices. If you design properly, there is no need for 2.4 ever. 2.4Ghz congestion (with nearly no actual 802.11 traffic) is very high. To the point where the IoT folks are struggling.
My 2.4ghz is basically all IOT these days. Things that matter are on 5 or 6 ghz. Busy moving the entire thing to be entirely firewalled off given how clean the separation is
Yes. And 2.4 lives and dies by that sword. What downsides might there be in areas where dozens of APs hear each other and 100s of clients hear each other?
It's an IoT device, not a laptop. It does not really need 5ghz to fulfill its purpose as an embedded CPU, and adding 5ghz likely would require making some room for it by removing other functionality.
Yes and in some uses cases it works against you. 2.4 is incredibly crowded without adding 802.11 to the mix. My IoT admins would have less complaints if they could take advantage of my small cell 5Ghz spectrum. This isn't 2005 with widely deployed asymmetrical wireless networks.
Can't you just underpower the antenna on a 2.4 radio if you need networks that don't bleed into each other as badly? Unless it's an issue because of the tiny antennas that usually come on microcontrollers.
I assume their chips don't really exist until they're actually supported by ESP-IDF. The ESP32-C5 was announced in June 2022, received initial support in -IDF in August 2025, and more complete support in December. It seems to have only recently started getting third party dev boards.
Realistically 2.4Ghz is far from "greatest backward compatibility" since there is a real benefit of running 5Ghz and 6Ghz only networks.
2.4Ghz makes sense because this tiny device does not need high speeds Wi-Fi connection, and deployment scenarios benefit from 2.4 GHz penetration more.
It would be good if this chip had good idle current comparable to other MCUs. I have used the ESP32S3 and it's idle current with the radio enabled, but not transmitting, is quite terrible.
My application needed both can bus and Bluetooth (though no wifi) so the S3 was one of the only options available. I'm sure the high current draw is because the wifi and ble share the same radio?
Without being hands on, it's difficult to make a direct comparison. There's 2 processors according to CNX [0], and the HP core's instruction set might roughly be comparable to M55.
Don’t know the specifics of the Espressif RISC-V cores, but in general they can’t really compete on those aspects with ARM.
ARM is a much more mature platform, and the licensing scheme helps somewhat to keep really good physical implementations of the cores, since some advances get “distributed” through ARM itself.
Compute capabilities and power efficiency are very tied to physical implementations, which for the best part is happening behind closed doors.
you can run linux on riscv without an MMU. There is mainline support for Kendryte K210 chip, so it should be possible port to this chip provided you have enough PSRAM.
Although, I'd like to seem some non-paid blogger head-to-head reviews benchmarking instruction cycle efficiency per power of comparable Arm vs. ESP32 Xtensa LX6* and RISC-V parts.
* Metric crap tons of WROOM parts are still available and ancient ESP8266 probably too.
That native sdk and the vscode plugin are very professional. There is a bit of a learning curve to get into it, but once you do, it's very functional and the developers are super supportive. They have fixed bugs for me in days.
Still requires using rust compiled against their llm fork. 'espup' makes it easy if you're okay with using it.
Other than that it works pretty well. This is if you run ESP-IDF, with bare-metal rust it's either best thing ever or meh. Rust community seems to use stm32 and picos more.
It’s not like creating a chip gives you unfettered access to it. You _can_ add 0-day flaws and backdoors, but these can be discovered, leaked, etc. Has there been any case of such a backdoor built in consumer chips like theses? I’m not talking about CIA ops like snowden described, that’s supply chain interception. I mean, has anybody ever found such a backdoor?
Well, that depends on what you count as a backdoor, but Espressif has had some questionable flaws:
- Early (ESP8622) MCUs had weak security, implementation flaws, and a host of issues that meant an attacker could hijack and maintain control of devices via OTA updates.
- Their chosen way to implement these systems makes them more vulnerable. They explicitly reduce hardware footprint by moving functionality from hardware to software.
- More recently there was some controversy about hidden commands in the BT chain, which were claimed to be debug functionality. Even if you take them at their word, that speaks volumes about their practices and procedures.
That’s the main problem with these kinds of backdoors, you can never really prove they exist because there’s reasonable alternative explanations since bugs do happen.
What I can tell you is that every single company I’ve worked which took security seriously (medical implants, critical safety industry) not only banned their use on our designs, they banned the presence of ESP32 based devices on our networks.
Except if you penetrate the market with modules that cost 5% of similar US made solutions, you start to win mindshare. At least some of those hobbyists start making a product, and sometimes the determination of whether a product is "safety critical" isn't agreed upon until after it's failed catastrophically.
I'm puzzled by Espressif's naming here. We had the ESP32-S3, so "S31" sounds like "S3, variant 1," but this part doesn't really look like a simple S3 variant. And then there's an ESP32-E22, but no E21 or even a plain E2 anywhere.
Edit: found an article explaining some of their naming logic, and said that the SoC naming will get its follow-up article, but sadly it never happened. https://developer.espressif.com/blog/2025/03/espressif-part-...
It reminds me a bit of the new STM32s (STM32MP2) which are actually 64 bit, but they kept the name STM32 because everyone knows it
Didn’t Intel also try to brand the 64bit x86 extensions as ia-32e initially? Seemed like wasting an opportunity to me.
(Disclaimer: I work at Intel but this was way before my tenure.)
It was because IA-64 was a completely different unrelated architecture that until AMD succeeded with K8 was "the plan" for both 64bit intel roadmap and the roadmap to kill off compatible vendors (AMD, VIA)
I stopped following the producer logic when Intel went from Pentium 4 to Pentium D
They claim that the chip has an "MMU". But unfortunately this doesn't seem to be a true RISC-V MMU (according to the Sv32 specification) integrated into the CPU core itself, but just a peripheral designed for memory mapped SPI flash and PSRAM. So as far as I understand there is no true process isolation with page faults and dynamic paging.
That’s a shame, it’d be a cool and, afaik, unique feature for this niche.
Maybe Espressif will notice that there are no RV32 chips with MMU so far (at least to my knowledge); we only have 32 bit MCUs or then only 64 bits for the CPUs. Something like Cortex-A7 is missing.
The upcoming Baochip is an RV32 chip with an MMU, I believe.
https://www.bunniestudios.com/blog/2026/baochip-1x-a-mostly-...
Edit - Oops GeorgeHahn beat me to it
There’s one incoming, https://baochip.github.io/baochip-1x/ It would be great to see more.
Does anyone know what CPU this uses? Is it their own first party design?
Interesting that they made a new chip with BLE+BR/EDR again. all the chips after the original ESP32 were BLE only. Hope this chip has good low power options so we can use it in Bluetooth audio workloads.
I believe this is the first ESP to gain Ethernet capability?
I totally wish that a board would come with PoE…
Because as it is right now, powering a fleet of those with USB power supplies is annoying as fsck…
Nah, ESP32's have had ethernet capability for a while and ESP-IDF supports it well. I've been using one I built for 5+ years now. Unfortunately RMII (ethernet phy) interface takes up a lot of the GPIO pins. This part looks like it'll remedy that issue.
There's two ESP32 boards that have been around for a while with PoE:
- https://www.tme.com/us/en-us/details/esp32-poe/development-k... - https://wesp32.com/
I'm more hopeful for single-pair ethernet to gain momentum though! Deterministic, faster than CANBUS, single pair, with power delivery:
https://www.hackster.io/rahulkhanna/sustainable-real-time-la...
> I'm more hopeful for single-pair ethernet to gain momentum though!
I keep looking for a reasonably priced 10baseT to 10Base-T1L bridge... everything commercial seems too expensive (for me) and the two hobby designs [1] [2] I've seen are not orderable :(
But I'm seeing more commercial options lately, so that's hopeful.
[1] http://robruark.com/projects/10BASE-T1L/10BASE-T1L.html
[2] https://matthewtran.dev/2024/08/10base-t1l-converter/
Waiting for my ManT1S:
https://www.crowdsupply.com/silicognition/mant1s
I really wish there was a camera option. You’d have made wired doorbell cameras possible without a retrofit.
I’d buy in a heartbeat
SPE with multidrop and PoDL would be awesome ! They are working on that and it will be everywhere.
Multidrop SPE isn't going to outperform newer CAN versions though. Somewhere in the sub-100Mb/s (e.g. 10-20Mb/s range) is the practical maximum speed of a multidrop bus at useful lengths, and that essentially applies equally to CAN or SPE. The only way to really get faster in a "multidrop-like" sense is with logically loop-like systems like ethercat and Fibre Channel where each network segment is point-to-point and the nodes are responsible for the routing.
M5stack has PoE offerings
The original ESP32 has Ethernet as well, I believe in the form of RMII. Then it has been removed from the chip, never specified the reason.
This would be great indeed.
On that note, why does the PoE capability often add such a big proportion of the price of various items? Is the technology really costly for some reason, or is it just more there's fairly low demand and people are still willing to pay?
PoE is not obvious to implement (take it from someone who has done it with a fair share of mistakes), uses more expensive components that normal ethernet, takes up more space on the board, makes passing emissions certification more complex, and is more prone to mistakes that ruin boards in the future, causing support/warranty issues. In other words, a bag of worms: not impossible to handle, but something you would rather avoid if possible.
And what would a better alternative look like ?
I wouldn't call it "better", but the least-effort path among hobbyists and low end gear is often 12v or 24v sent over a pair with Gnd and a forgiving voltage regulator on the other end.
There is none, I never said PoE is "bad": it's a very good solution, it's just difficult to implement.
A full-module add-on in this power class is about $7 at 1,000 unit scale [0]. It would be around $3 with your own custom PCB design in terms of BoM addon at scale. That’s power only. Add another dollar or two for 10/100 PHY.
The trick is as others have said in what adding it to your design does in terms of complicating compliance design.
[0] https://www.digikey.com/en/products/detail/silvertel/AG9705-...
PoE power supplies need to be isolated (except in rare exceptions) and handle much higher voltages than common USB-C or wall wart power supplies.
They have to use a transformer and a more complex control strategy, not a simple buck regulator with an inductor. PoE inputs need to tolerate voltages several times higher than the highest USB-C voltages, so more expensive parts are used everywhere.
Any Ethernet (well, any RJ45 you expect in a home/office) has to have at least 1500V isolation from the RJ45 wire to anything metal that can be touched or is a connector on the device. A PoE-only device with no electrical connectors besides the RJ45 can just use a very cheap RJ45 port with integrated magnetics and PoE allowance (tiny bit bigger wires and a center pin exposed, less than 50ct more than the cheapest RJ45 with integrated magnetics) and a cheap buck from 40~80V to e.g. 5V.
Oh, and a cheap bridge rectifier and some signaling resistors to take care of input polarity and signal to the source that we in fact want the approximately 50V that could hurt a device not made for it.
It sounds like the PoE spec was designed before the arrival of “IoT” type things like the esp32, raspberry pi’s, etc.
How much of the complexity is a “fundamental electrical engineering problem” and how much of it is just a spec written to solve a different set of problems?
Ethernet is already one of the most expensive standards because you need magnetics for isolation. Adding power on top of that is genuinely expensive.
Whenever you combine two things into one, the complexity and cost go up considerably. A regular coffee machine is pretty cheap. Add high pressure so it can make espresso and it gets considerably more expensive. Add milk so it can make cappuccino, again more complex and expensive. The same holds for electronics. Isolating power when it's alone is fairly straightforward. It gets considerably more tricky and hence more expensive the moment you want to place any kind of a meaningful data signal in its vicinity.
I’m sure the other commenters are right, but I’m guessing market segmentation may play a role here too.
> Because as it is right now, powering a fleet of those with USB power supplies is annoying as fsck…
Therefore, wifi is more convenient than ethernet.
You don't need long cables, just a local power source.
> You don't need long cables, just a local power source
Which means batteries that have to be replaced and maintained or cables... So ethernet with PoE or even better SPE (single pair Ethernet) with PoDL (power over data lines which is PoE for SPE) is the best from my point of view
Well, yes, but then you need to be "in range" of PoE switch and drag the ethernet cable from it vs the nearby socket. Still, nice to have options
I mean, if I just look at my house. There is just one ethernet outlet, but many power sockets. If I want to connect devices all over my house, the best way is to use wifi and usb power adapters. Not ethernet.
Both solutions require 1 cable per device, but the first solution would require only short and thin cables, and the second solution would require very long cables which I don't know even how to do properly without milling my walls.
Yep. Mains electricity is ubiquitous, highly interoperable, very reliable, very high power available per drop, can be outdoor capable, common standards, understandable by users, requiring no active components, with many on-call experts available who can come to fix problems or extend/alter connectivity. Mains power wall plates with inbuilt USB power outlets are even available at quite small cost if the look of the bigger plug and wiring is not appealing.
PoE is much fewer of those things. Difficult to recommend it these days with wifi being fast and reliable and so widely used. Certainly not for average residential user.
That's half the equation. The other half is the reliability and security of wifi, which is less than that of ethernet for people without physical access to my wall innards
On the other hand, _all_ the WiFi devices that I had at some point fell off the network, at least once. Including doorbells and cameras. While PoE devices just work.
Another point is that mains power in my area can go down periodically. My PoE switch is powered by a Li-Ion UPS and can provide power for about a day.
Esp32 's wifi is only 2.4ghz though.
Can't you run a 5V supply from where your router is all the way to every god damn device in your house, and then pretend the wifi is also going through it? If you just want it to be inconvenient, there's no reason to let a lack of PoE stop you!
Are there any US vendors with wifi/BLE-integrated MCUs -- a single package that does it all?
I don't understand what possesses these folks to continue making 2.4ghz devices. I understand there are use cases for low bandwidth, high range. But surely we've passed the point where that is more desirable to most than lower latency and high throughput, right?
> I understand there are use cases for low bandwidth, high range.
Use cases like IoT? The very thing this is for?
Is what you described a truth for all IoT devices? If I have los of my AP, why do I need 2.4Ghz? Even so, what SNR do you truly need for this low bandwidth application? Where is the engineering here?
I have a unique position of having a data set over 8000 APs with 40k unique devices. If you design properly, there is no need for 2.4 ever. 2.4Ghz congestion (with nearly no actual 802.11 traffic) is very high. To the point where the IoT folks are struggling.
My 2.4ghz is basically all IOT these days. Things that matter are on 5 or 6 ghz. Busy moving the entire thing to be entirely firewalled off given how clean the separation is
>My 2.4ghz is basically all IOT these days.
Yup. And it's exactly why some of my IoT admins are struggling. There is only so much spectrum to go around.
2.4 GHz has the advantage of it passing through obstacles easier. The higher the frequency the more easily it gets blocked.
Yes. And 2.4 lives and dies by that sword. What downsides might there be in areas where dozens of APs hear each other and 100s of clients hear each other?
It's an IoT device, not a laptop. It does not really need 5ghz to fulfill its purpose as an embedded CPU, and adding 5ghz likely would require making some room for it by removing other functionality.
Yes and in some uses cases it works against you. 2.4 is incredibly crowded without adding 802.11 to the mix. My IoT admins would have less complaints if they could take advantage of my small cell 5Ghz spectrum. This isn't 2005 with widely deployed asymmetrical wireless networks.
Can't you just underpower the antenna on a 2.4 radio if you need networks that don't bleed into each other as badly? Unless it's an issue because of the tiny antennas that usually come on microcontrollers.
I don’t trust Espressif’s releases, I am still waiting for ESP32-P4 to hit distributors. It is now more than 2 years and 3rd chip revision
Can also be ordered on JLCPCB in a custom PCB: https://www.lcsc.com/product-detail/C22387510.html?s_z=n_ESP...
I assume their chips don't really exist until they're actually supported by ESP-IDF. The ESP32-C5 was announced in June 2022, received initial support in -IDF in August 2025, and more complete support in December. It seems to have only recently started getting third party dev boards.
HN title entropy record?
[delayed]
Pardon for off topic. The designer in me hates that they used AI for every single asset in this release post, looks so amateurish.
Argh…Wifi 6 , but 2.4ghz.
Adding 2 frequency ranges is more expensive. If you're going to choose just one, you should choose the one with the greatest backward compatibility.
Realistically 2.4Ghz is far from "greatest backward compatibility" since there is a real benefit of running 5Ghz and 6Ghz only networks.
2.4Ghz makes sense because this tiny device does not need high speeds Wi-Fi connection, and deployment scenarios benefit from 2.4 GHz penetration more.
It would be good if this chip had good idle current comparable to other MCUs. I have used the ESP32S3 and it's idle current with the radio enabled, but not transmitting, is quite terrible.
My application needed both can bus and Bluetooth (though no wifi) so the S3 was one of the only options available. I'm sure the high current draw is because the wifi and ble share the same radio?
Love ESP boards, and with Raspberry pi pricing though the roof, I’m hoping more will discover the love of getting the job done on a 10mm2 package.
I suspect a lot of the things people are using RPi for are better served by things like this (and virtualisation for the heavier end)
Perfect timing. I just started planning to build a DIY smartwatch and was looking into the S3. Having native zigbee support could be nice.
I hope this one has multiple radios so you can actually use BT/Wifi/Thread simultaneously.
> high-speed 250 MHz 8-bit DDR PSRAM with concurrent flash and PSRAM access
This is perhaps lost in the noise but IMO a large deal. PSRAM starting to get serious bandwidth.
For reference, the 4-bit PSRAM interface on the ESP32-S3 normally runs at 80 MHz (maximum 120 MHz) and shares bandwidth with the external flash.
I wonder if it will be possible to (ab)use the faster PSRAM interface on the ESP32-S31 as a general purpose 8-bit parallel interface, eg. for ADCs...
Oh neat. Zigbee support.
I wonder if I at some point can create low power devices with EspHome for home assistant. I assume this should use less power than connecting to wifi?
The C6 and the H2 already support ZigBee. Their SDK has a thin layer on top of zboss.
You already can with nRF52 boards. Presumably they'll add ESP32 support soon too.
https://esphome.io/components/zigbee/
How do Espressif’s RISC-V cores compare to existing ARM or RISC-V options in terms of power efficiency (computational power / electrical power)?
Without being hands on, it's difficult to make a direct comparison. There's 2 processors according to CNX [0], and the HP core's instruction set might roughly be comparable to M55.
[0] https://www.cnx-software.com/2026/03/24/esp32-s31-dual-core-...
Don’t know the specifics of the Espressif RISC-V cores, but in general they can’t really compete on those aspects with ARM.
ARM is a much more mature platform, and the licensing scheme helps somewhat to keep really good physical implementations of the cores, since some advances get “distributed” through ARM itself.
Compute capabilities and power efficiency are very tied to physical implementations, which for the best part is happening behind closed doors.
Soon espressif will add TPU to their chips.
Hopefully it will be just added to https://github.com/espressif/esp-nn on software side.
Is there something that match those elsewhere ?
Roughly how much do you think this costs?
Given their history, I would guess <$6 a piece for a dev board, <$2 for the chip at scale.
Does it run Linux?
No, missing a MMU.
There is μClinux [1] although it is not clear to me how much alive is the project
I wish I could run DiscoBSD/RetroBSD [2] on an ESP32, I like the idea of running on a MCU something that was originally meant for a PDP/11 (2.11 BSD)
[1] https://en.wikipedia.org/wiki/%CE%9CClinux
[2] https://github.com/chettrick/discobsd
you can run linux on riscv without an MMU. There is mainline support for Kendryte K210 chip, so it should be possible port to this chip provided you have enough PSRAM.
why would you do that? (unless for the fun of it)
Just to get an idea of its capabilities.
Imagine installing Claude Code in it with full root access. Just by asking a few things, you can have a semi-broken ESP32 doing stuff.
Interesting.
Although, I'd like to seem some non-paid blogger head-to-head reviews benchmarking instruction cycle efficiency per power of comparable Arm vs. ESP32 Xtensa LX6* and RISC-V parts.
* Metric crap tons of WROOM parts are still available and ancient ESP8266 probably too.
The ESP32 boards I own have bad support and are a bit of a hit and miss. (arduino nano esp32) Did this get better? Or is the support still messy?
That native sdk and the vscode plugin are very professional. There is a bit of a learning curve to get into it, but once you do, it's very functional and the developers are super supportive. They have fixed bugs for me in days.
What do you mean by that? bad support for what?
Arduino nano are made by arduino using Espressif chips, and Arduino IDE support is indeed hit and miss.
ESP-IDF, the official C SDK, is a bit more work, and there is drama around platform-io, but it’s significantly more stable.
> there is drama around platform-io
What do you mean ?
Don't use the Arduino framework, use ESP-IDF or Rust.
For those using PlatformIO, the folks at pioarduino[0] are doing a great job keeping up with Arduino Core 3.x support.
[0]: https://github.com/pioarduino/platform-espressif32how's Rust on the xtensa cores ?
Still requires using rust compiled against their llm fork. 'espup' makes it easy if you're okay with using it.
Other than that it works pretty well. This is if you run ESP-IDF, with bare-metal rust it's either best thing ever or meh. Rust community seems to use stm32 and picos more.
Since the Snowden leaks in 2013, it just doesn't make sense that *any* foreign customers would put US technology inside their firewall. But they do.
It shocks me even more that any Western customer would do the same with network-connected Chinese chips. But we do.
The Espressif chips are truly incredible value, but what are we doing here?
Is there any doubt that these don't represent a major attack surface if a conflict were to heat up?
If you had network-connected chips of your own design inside every household of your adversary, what could you do with that?
It’s not like creating a chip gives you unfettered access to it. You _can_ add 0-day flaws and backdoors, but these can be discovered, leaked, etc. Has there been any case of such a backdoor built in consumer chips like theses? I’m not talking about CIA ops like snowden described, that’s supply chain interception. I mean, has anybody ever found such a backdoor?
Well, that depends on what you count as a backdoor, but Espressif has had some questionable flaws:
- Early (ESP8622) MCUs had weak security, implementation flaws, and a host of issues that meant an attacker could hijack and maintain control of devices via OTA updates.
- Their chosen way to implement these systems makes them more vulnerable. They explicitly reduce hardware footprint by moving functionality from hardware to software.
- More recently there was some controversy about hidden commands in the BT chain, which were claimed to be debug functionality. Even if you take them at their word, that speaks volumes about their practices and procedures.
That’s the main problem with these kinds of backdoors, you can never really prove they exist because there’s reasonable alternative explanations since bugs do happen.
What I can tell you is that every single company I’ve worked which took security seriously (medical implants, critical safety industry) not only banned their use on our designs, they banned the presence of ESP32 based devices on our networks.
You can hide malicious intent, so the repeated negligence patterns you’re pointing out make a better signal. Smart. Thx for the perspective
Obviously... They are not made for safety critical systems. It's for hobbyists.
Except if you penetrate the market with modules that cost 5% of similar US made solutions, you start to win mindshare. At least some of those hobbyists start making a product, and sometimes the determination of whether a product is "safety critical" isn't agreed upon until after it's failed catastrophically.