I love the consistent CPU performance gains that Apple generates, always between between 15% to 30%.
I know this is the standard since the Apple got on the ARM train, but for over a decade from 2008 to 2020 in the Intel world, there was very few significant gains like this. The best you got was like 10% and it wasn't even a yearly release cadance.
And 20% YOY compounds quickly… If the speed doubles in 4 years and the RAM situation gets better, we could be looking at some very capable on-device models in 2030.
Very quick, but I wonder how much of it comes from the clock speed bump (4.93 GHz from 4.26 GHz, +15.7%) instead of overall platform IPC improvements.
If it were largely clock speed, I worry we'll find that thermal throttling makes these numbers only generally available in the lab or very short bursts before the device gets too warm.
That said, very excited for what that means for the Neo laptops with more thermal headroom once they get a refresh, and if it happened to be this chip. Though I suppose that is unlikely, as they seem to be adopting a strategy of using past-gen chips for Neo.
> "I wonder how much of it comes from the clock speed bump (4.93 GHz from 4.26 GHz, +15.7%)"
About 15.7%.
> "If it were largely clock speed, I worry we'll find that thermal throttling makes these numbers only generally available in the lab"
Going from TSMC's 3nm N3P to 2nm N2 process reduces overall power consumption and thermal output by more than 15%. The reason they're able to boost the clock is because they have new headroom. It also increases the thermal density, which means the heat that's generated needs to be moved more quickly, which is probably why they also have new thermal management.
Absolutely massive improvement in CPU performance in a single year. Looks like TSMC N2 is going to be a very good node.
+22% faster single thread
+34% faster multi thread
+40% GPU
This is suppose to be the generation that Apple wants to fast forward because they want to put more resources into the M7 generation for local AI. There are no M6 Pro/Max/Ultra planned.
I think the key AI improvement is that these chips double the memory bandwidth and this has been the main limit for AI performance recently, became LLMs do not fit into cache.
I know they said the A20 was desktop class, but beating the M5 is wild to me. I can’t wrap my head around the fastest chip (at least for single-core performance) is in a phone instead of a desktop.
I thought maybe they clock the M5 down for the iPad, but when I go over to the Mac benchmarks, I see the same thing.
I think it depends on the MacBook Neo refresh frequency.
When the MacBook Neo was released this year, it used the latest A-series chip (although non Pro version.). So when it is refreshed I suspect it will again use the latest A-series chip, but the non Pro-version.
Is it going to be refreshed yearly or bi-yearly? We will find out I guess around March 2027.
Because it ruffles x86 feathers, despite geekbench corresponding very closely to SPEC as well and there being no indication that the results aren’t comparable.
What people willfully conflate this with often is that you shouldn’t compare FLOP numbers between GPU arches and instead rely on benchmarks.
Geekbench when run on Asahi Linux is practically the same as when run on macOS, just like Linux and Windows in the same x86 chip don’t give significant differences.
I love the consistent CPU performance gains that Apple generates, always between between 15% to 30%.
I know this is the standard since the Apple got on the ARM train, but for over a decade from 2008 to 2020 in the Intel world, there was very few significant gains like this. The best you got was like 10% and it wasn't even a yearly release cadance.
And 20% YOY compounds quickly… If the speed doubles in 4 years and the RAM situation gets better, we could be looking at some very capable on-device models in 2030.
Not to mention GPU performance, which has also steadily improved.
The A20 Pro has about the same GPU performance as an M1 Pro, according to the Metal GPU benchmark. Very impressive.
Sometimes Intel chips would regress generation on generation. Broadwell in 2015 was sometimes 10% slower than the generation it replaced.
Very quick, but I wonder how much of it comes from the clock speed bump (4.93 GHz from 4.26 GHz, +15.7%) instead of overall platform IPC improvements.
If it were largely clock speed, I worry we'll find that thermal throttling makes these numbers only generally available in the lab or very short bursts before the device gets too warm.
That said, very excited for what that means for the Neo laptops with more thermal headroom once they get a refresh, and if it happened to be this chip. Though I suppose that is unlikely, as they seem to be adopting a strategy of using past-gen chips for Neo.
> "I wonder how much of it comes from the clock speed bump (4.93 GHz from 4.26 GHz, +15.7%)"
About 15.7%.
> "If it were largely clock speed, I worry we'll find that thermal throttling makes these numbers only generally available in the lab"
Going from TSMC's 3nm N3P to 2nm N2 process reduces overall power consumption and thermal output by more than 15%. The reason they're able to boost the clock is because they have new headroom. It also increases the thermal density, which means the heat that's generated needs to be moved more quickly, which is probably why they also have new thermal management.
I can imagine that the new vapour chamber they’re introducing helps or is necessary to run this chip at high speeds.
Absolutely massive improvement in CPU performance in a single year. Looks like TSMC N2 is going to be a very good node.
+22% faster single thread
+34% faster multi thread
+40% GPU
This is suppose to be the generation that Apple wants to fast forward because they want to put more resources into the M7 generation for local AI. There are no M6 Pro/Max/Ultra planned.
I think the key AI improvement is that these chips double the memory bandwidth and this has been the main limit for AI performance recently, became LLMs do not fit into cache.
They should use LPDDR6 which is a sizable improvement but the other thing they can do is increase prompt processing even further.
A blazing fast chip, even by Apple standards. Exceeds M5 iPad Pro in single core for a new record.
It even surpasses the iPad Pro M3 on multi-core, so very capable indeed.
All iOS devices on Geekbench: https://browser.geekbench.com/ios-benchmarks
Also, a max frequency of 4.93Ghz according to Geekbench!
I know they said the A20 was desktop class, but beating the M5 is wild to me. I can’t wrap my head around the fastest chip (at least for single-core performance) is in a phone instead of a desktop.
I thought maybe they clock the M5 down for the iPad, but when I go over to the Mac benchmarks, I see the same thing.
https://browser.geekbench.com/mac-benchmarks
Apparently the A20 Pro Performance Core is a new design that is different to the M6 Super Core.
Also GPU test: https://browser.geekbench.com/v6/compute/compare/6874665?bas...
This is interesting because the single Core is even fasting that the M5 suggest the two 2nm chip might not share the same Core.
This is like the good ol' days of the x86 wars. Mo' cache, mo' GHz.
I find it impressive that these pocket computers are running at almost 5 GHz now.
looking forward to have this in a macbook neo
also, i don't know much memory that chip has, but i guess the A20 Pro could also be viable for low-end macbooks air.
edit: nevermind, just 12gb. still a good candidate for a macbook neo.
I wonder if we could get MacBook Neos with this chip instead of the A18.
I think it depends on the MacBook Neo refresh frequency.
When the MacBook Neo was released this year, it used the latest A-series chip (although non Pro version.). So when it is refreshed I suspect it will again use the latest A-series chip, but the non Pro-version.
Is it going to be refreshed yearly or bi-yearly? We will find out I guess around March 2027.
You have that exactly backwards. The MacBook Neo released with the previous years A series Pro chip.
It’s currently got an A18 Pro.
2027 April will be MacBook Neo with A19 Pro and 12GB Memory.
2028 April we should be getting A20 Pro MacBook Neo. But I wouldn't be surprise if they bump the price by $100 again.
Give it a few years. I’m sure it will come eventually.
Imagine what a bunch of the could do in some kind of blade servers, very low power usage beasts for just hosting things or low to medium computing.
Just dreaming.
Reminder that despite their claims, geekbench results cannot be compared across platforms.
They don't only cover CPU and GPU but some OS components that can look very different on different platforms.
Because it ruffles x86 feathers, despite geekbench corresponding very closely to SPEC as well and there being no indication that the results aren’t comparable.
What people willfully conflate this with often is that you shouldn’t compare FLOP numbers between GPU arches and instead rely on benchmarks.
No, because the software layers between GeekBench and the hardware are not identical on all targets.
Geekbench when run on Asahi Linux is practically the same as when run on macOS, just like Linux and Windows in the same x86 chip don’t give significant differences.
Is it fast enough for Apple to disable transitions in the UI?
just kidding, I know Apple knows better - UIs are meant to have lag :)