Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
New cores (A15)
- Thread starter cmaier
- Start date
- Sort by reaction score
Apple put out a release saying the new A15 was 40% faster than the A12, which is what they said last year for the A14.
The above is all we’ve really got to go on.
Anandtech will be the best barometer as they are by far the best of the "processor tech" sites out there.
Yeah, a little hard to know what that ends up meaning, since we don’t know how they’re clocking things, etc. I would seriously doubt that A15 has 0% improvement over A14 single core CPU performance.
Apple now deals with multiple SoC configuration not only with iPhone and iPad but also Mac as well. I don't think they will maintain the yearly cadence on new CPU cores every year. Looking like the performance improvement will come from the improved process rather than the architecture this time around.
From the Anandtech article, mentions two possible theories.
From the Anandtech article, mentions two possible theories.
I doubt that. Any new core design is easily leveraged into each product, due to the SoC methodology. And it’s unlikely that the design team has been sitting around doing nothing other than a process shrink for the last year.
Well, if that is the case, I think that is the worse sign since it would indicate they just hit the wall in terms of architecture improvement.
My sense was they chose power over performance. All the performance you used to have, a bit more GPU and neural, brighter displays all around and still 2 extra hours of runtime.
If nobody is using all that CPU yet, makes sense to take a breath and improve efficiency.
edit: yeah, basically what @Kpjoslee found:
Yes, I bet the cores are underclocked to save battery, well I hope the cores are underclocked
We'll see the benchmarks soon.
A ten percent bump would place it near the M1 in single core performance, ideally with the same power consumption as the A14.
They likely chose lower power per core over single core performance. Thus allowing them to (a) improve battery life on iDevices, or (b) stuff a lot more cores in Mac systems without needing exotic cooling.
Yeah we’ll just have to wait for them to be in the wild. As others mentioned, maybe more improvements to power efficiency rather than performance this time around … Another amusing possibility is that Apple’s marketing team got sloppy and just copied stuff from last year and forgot to change the numbers.
I dunno, we’ll find out soon enough
So, back to Apple Kremlinology, how does this announcement, or more specifically lack of an announcement, compare with the information that you previously had from the sources that "you may or may not" have in regards to the next Mac SoC? I believe you mentioned an approximate 14% gain from the previous generation, all things being equal, but if that is the case, Apple had very little to say about what's new in the A15.
Apple must have used that increased transistor budget for something, but for whatever reason, they haven't yet highlighted those enhancements.
As the Anantech article points out:
Also, they had this comment, which may or may not explain some of the changes, or more specifically, lack of changes:
I know you've been skeptical that the engineers that left for Nuvia would have a significant impact on Apple's SoC plans going forward, but the timing is notable, according to the Anandtech article.
Obviously, it's too soon to understand exactly what is going on, but @cmaier has better insights into how this will possibly play out, considering your background. I admit that I'm confused about Apple not emphasizing what's new in regards to the A15 and the reasons behind that. Is it possible that the A15, and future chips derived from that design, are in fact a relatively minor updates to current designs?
I always thought the 8+2/10+2 rumors were iffy. As you say @cmaier, it doesn't make sense without a fundamentally different perf/W relationship between the "P" and "E" cores, and Apple has a history of designing new cores by making incremental improvements to the last generation.
On perf/W - according to powermetrics, the M1 4-core E cluster uses about 2.5W at max freq / 100% load and the 4-core P cluster about 25W. 1/10th the power is really important. Even though the E cluster offers ~1/3 the performance of the P cluster, that translates to ~3.3x perf/W. As for area, the E cluster needs ~1/4 the die area of the P cluster, or roughly ~1.3x perf/mm^2.
Giving up two E cores to get about half a P core just doesn't seem like a good tradeoff. Why starve low-priority background tasks of the best resource to run them on? And when you want to do a lot of computation with many parallel threads, E cores are arguably better.
You need a balance, because Amdahl's Law is real, but 8+2 doesn't seem like a great balance.
On perf/W - according to powermetrics, the M1 4-core E cluster uses about 2.5W at max freq / 100% load and the 4-core P cluster about 25W. 1/10th the power is really important. Even though the E cluster offers ~1/3 the performance of the P cluster, that translates to ~3.3x perf/W. As for area, the E cluster needs ~1/4 the die area of the P cluster, or roughly ~1.3x perf/mm^2.
Giving up two E cores to get about half a P core just doesn't seem like a good tradeoff. Why starve low-priority background tasks of the best resource to run them on? And when you want to do a lot of computation with many parallel threads, E cores are arguably better.
You need a balance, because Amdahl's Law is real, but 8+2 doesn't seem like a great balance.
Just for comparison, Apple M1 "only" has 16B transistors, and now the A15 has 15B.
Not sure if these numbers came to pass, but from EE Times Asia in 2019:
"The N5P [...] could squeeze anther [sic] 7% in speed or 15% in power from N5 using the same design rules. The gains come in part from enhancements to a fully strained high-mobility channel."
Makes it sound like that 6% was essentially what they'd get for free.
Last edited:
It's not only possible, it's likely. But not for the reasons you're speculating about.
Apple has a lot of revenue riding on shipping a new iPhone family every fall in high volume. This means Apple's silicon engineering group pays a lot of attention to the risk of schedule slips. One of the best ways to de-risk new chip designs where the spec is mostly "same as last year's chip, but better" is to not keep re-inventing the wheel. Anywhere you can build on last year's design without sweeping changes, you do that.
There's no reason why this would be any different after Gerard Williams' departure.
I believe the lack of CPU speed improvements can be explained by A15 focusing on battery life instead of absolute performance. They already outperform anyone else, it makes sense to slightly lower the clocks/use less aggressive power curve.
As to Jade Die, I still expect it to be based on a different infrastructure variant. A15 is a low-power product. Jade Die will be a scalable performance product.
As to Jade Die, I still expect it to be based on a different infrastructure variant. A15 is a low-power product. Jade Die will be a scalable performance product.
Yep. Or 14 percent less power. Or some combo that adds up to 15. That’s what I hear. So is that right? We dunno.