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 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.
reply