The Apple A15 SoC Performance Review: Faster & More Efficient
by Andrei Frumusanu on October 4, 2021 9:30 AM EST- Posted in
- Mobile
- Apple
- Smartphones
- Apple A15
Conclusion & End Remarks
Today’s investigation into the new A15 is just scratching the tip of the iceberg of what Apple has to offer in the new generation iPhone 13 series devices. As we’re still working on the full device review, we got a good glimpse of what the new silicon is able to achieve, and what to expect from the new devices in terms of performance.
On the CPU side of things, Apple’s initial vague presentation of the new A15 improvements could either have resulted in disappointment, or simply a more hidden shift towards power efficiency rather than pure performance. In our extensive testing, we’re elated to see that it was actually mostly an efficiency focus this year, with the new performance cores showcasing adequate performance improvements, while at the same time reducing power consumption, as well as significantly improving energy efficiency.
The efficiency cores of the A15 have also seen massive gains, this time around with Apple mostly investing them back into performance, with the new cores showcasing +23-28% absolute performance improvements, something that isn’t easily identified by popular benchmarking. This large performance increase further helps the SoC improve energy efficiency, and our initial battery life figures of the new 13 series showcase that the chip has a very large part into the vastly longer longevity of the new devices.
In the GPU side, Apple’s peak performance improvements are off the charts, with a combination of a new larger GPU, new architecture, and the larger system cache that helps both performance as well as efficiency.
Apple’s iPhone component design seems to be limiting the SoC from achieving even better results, especially the newer Pro models, however even with that being said and done, Apple remains far above the competition in terms of performance and efficiency.
Overall, while the A15 isn’t the brute force iteration we’ve become used to from Apple in recent years, it very much comes with substantial generational gains that allow it to be a notably better SoC than the A14. In the end, it seems like Apple’s SoC team has executed well after all.
Facebook
Twitter
RSS
204 Comments
View All Comments
unclevagz - Monday, October 4, 2021 - link
Thanks, since Anandtech does have data on Spec 2017 subtests with various x86 processors it may also be helpful to show these results for selected x86 CPUs in the displayed graphs for ease of comparisions.Andrei Frumusanu - Monday, October 4, 2021 - link
I thought about it but didn't want to complicate it too much given the power disparity.Andrei Frumusanu - Monday, October 4, 2021 - link
I added in performance marks for the x86 folks. Obviously no power data.Kangal - Tuesday, October 5, 2021 - link
Hey Andrei,The graphs for Spec-2017 Efficiency looks quiet off. It's showing the Cortex-A55 cores consuming considerable more energy than Apple's E-cores, and sometimes even more than the Cortex-A78 cores too. Whilst performance seems as expected.
The worst offender seems to be the 544.nab_r, with the a discrepancy of 0.60 perf / 682 J = ~0.001 p/J compared to the 2.70 perf / 280 J = ~0.01 p/J. So that's an efficiency difference of ~x10 which is massive. And the best case for the A55 seems to be in the 541.leela_r test. Here we have 1.00 perf / 295 J = ~0.003 p/J compared to the 2.49 perf / 264 J = ~0.009 p/J. So in this best-case scenario the efficiency difference is ~x3 which is still huge.
I mean, I remember when Apple's E-cores were running slightly slower than the Cortex-A73 whilst using slightly more power than the Cortex-A53. But what we have here is just ridiculous. We have even less power draw than the Cortex-A55 or even the Cortex-A53, but performance is somewhere above the legendary Cortex-A76.
I can't wrap my head around it. It feels like an impossibility. Is my maths checking out? Or does there seem to be an issue someplace in the data?
Andrei Frumusanu - Tuesday, October 5, 2021 - link
Perf per joule is a bit of a weird metric that is superfluous, you want either perf/W or simply just Joules consumed for energy efficiency, so either 0.60 / 0.24W = 2.5ppW & 2.7 / 0.45W = 6ppW. You can argue about power curves and ISO-perf or ISO-power.In any case, the other thing to consider is that we're not just measuring the core, we're measuring the efficiency of the whole SoC, power delivery, DRAM as well. Some vendors aren't running things as efficiently as they should be, that's how you end up with those Exynos A55 results, contrasted for example to the MediaTek A55 results.
Kangal - Wednesday, October 6, 2021 - link
I didn't know that, I thought we had the software just churn out how much power the module was using on its own. With that said, I don't think it would be a factor. Apple doesn't have anything special in the makeup of their silicon to make it more efficient than competitors. And even if they did have a notable advantage in the make-up of their silicon, this would be against something like a RockChip SoC, and not against a flagship Qualcomm SoC. The more feasible explanation would be that the QSD chip might be activating other co-processors like it's NPU, and the task isn't actually being hardware-accelerated by it, but "software-encoded" by its targeted CPU (eg A55). Thus its still running slow, but now its wasting power by having other co-processors become active and not actually compute anything......Would something like this be a cause for concern, for future testing?
Secondly, I used the Joules as that's what the graph was visually showing. I basically used it to find the best-case and worst-case scenario. I didn't really think hard about it. Since you've graphed it, and since you've recorded it, I figured you knew something that I didn't and prioritised Joules over Watts.
Converting them to Watts, we instead get:
(nab_r) 2.70/0.45 = 6.00 vs 2.50 = 0.6/0.24 ---> a difference of x2.4
(leela_r) 2.49/0.40 = 6.23 vs 5.56 = 1.00/0.18 ----> a difference of x1.1
But now, the graphs themselves need to be switched. For instance, the New Worst-case scenario is now: 520.omnetpp_r (~x3.4) from what I can see. Maybe I'll go through these benchmark figures properly on a weekend or something, unless you guys plan on doing something of the sort.
So yes, these ranges do seem more reasonable. For starters, here we see the "IceStorm v2" cores are actually using about double the power of the "Cortex-A55" on half of the tests. This shatters my previous impression, that Apple's small cores were faster-than Cortex-A73 and used less-power than Cortex-A53. And that fits much neater into our general understanding about them, comparing small in-order cores, versus medium out-of-order cores.
Can we change how the graphs are displayed from now on? Plot the Watts on the Right/Second x-axis instead of Joules. Or better yet, let's just strip out Joules entirely. I mean the third graph, the Energy-Axis should probably be deleted, and just keep the Power-Axis there instead? No?
Ppietra - Wednesday, October 6, 2021 - link
Kangal,Joules will always be the most correct parameter to assess efficiency, since it is the actual energy expended to do all the work.
Power, on the other hand, can fluctuate through time while doing the work, so the Power value can be very deceiving, firstly because it might not be the actual average power usage, secondly because you need to do another calculation to actually measure efficiency.
Kangal - Wednesday, October 6, 2021 - link
Do you know how they calculated the Watts? And how they calculated the Joules?To me, Watts makes much more sense in this context/comparison. Joules is more "universal" measurement, and it might be useful in a niche, but I feel like it could me mis-used/abused easily when put out of context.
How do we explain the HUGE discrepancy in the measurements between Watts and Joules? There is something else here I am not understanding.
Ppietra - Wednesday, October 6, 2021 - link
For that you need to understand what is Power and what is Energy.If there is one parameter that can be misused to assess efficiency while doing a task it’s Power not Energy. What you don’t seem to account for it’s the Time variable that affects how you can interpret Power.
michael2k - Wednesday, October 6, 2021 - link
I wanted to specifically bring something up:Apple doesn't have anything special in the makeup of their silicon to make it more efficient than competitors.
A14: TSMC 5nm (N5)
A15: TSMC 5nm (N5P)
D1200: TSMC 6nm (N6)
SD888: TSMC 5nm (N5)
Technically Apple is one year ahead of Qualcomm and two or so ahead of MediaTek in terms of process.
Looking at the SPECin2017 Power Axis graph we see on average that the A15 IceStorm v2 consumes 0.44W/2349J to achieve a 2.42 score, which puts them on par with the D1200 A78 with it's 2.57 score, but at far higher power cost of 1.13W/6048J
In other words the A78 and A15 have very similar performance, which makes sense since there are many similarities in terms of number of execution units, width, etc. If you look at the older style charts you can see that the efficiency cores were far closer in performance to the A76 'performance' cores on the Kirin 990:
https://www.anandtech.com/show/14892/the-apple-iph...
https://www.anandtech.com/show/14892/the-apple-iph...
Long story short, there doesn't seem to be any surprises. Apple has a process advantage, uses cores similar to ARM's performance cores for efficiency purposes, and does so by clocking them at 3/4 the speed to dramatically reduce the power draw. The A15e only hits 2.016GHz and the A14e maxed at 1.823GHz, and the A13e at 1.728GHz