Apple's A20 Pro shatters Geekbench 7 single-core record — 2nm chip beats desktop Intel Core i9 and AMD Ryzen 9 by up to 32%

News
By
Published

Eats Intel's entry Panther Lake CPUs for breakfast too.

Apple A20 Pro
(Image credit: Apple)

Architectural enhancements and significantly higher clock speeds enable Apple's A20 Pro application processor (AP), used in the company's latest iPhones, to deliver not only a substantial generation-to-generation performance boost but also to outperform leading desktop CPUs from AMD and Intel by up to a whopping 32% in the single-thread Geekbench 7 benchmark, setting the record for the highest single-thread performance. While high-end PC CPUs still have more oomph for multi-threaded workloads, the tiny A20 Pro is still faster than mainstream laptop CPUs even when many threads are involved.

Fastest smartphone SoC

Latest Videos FromTom's Hardware
Anton Shilov
Anton Shilov
Contributing Writer

Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.

27 Comments Comment from the forums
  • Faiakes
    I suppose single thread scores are good for efficiency?

    But multi thread cannot be compared to desktop class.
    Reply
  • LordVile
    Faiakes said:
    I suppose single thread scores are good for efficiency?

    But multi thread cannot be compared to desktop class.
    It’s a phone though. Anything intensive is going to be lightly threaded and short. It’s intriguing for what the M6 will be like though
    Reply
  • circadia
    Admin said:
    Apple's A20 Pro shatters Geekbench 7 single-core record
    if only there are better multi-platform benchmarks that properly test everything on iPhones, Androids, PCs, Macs whatever... instead of.... bloody Geekbench.
    but I guess we have to take what we have.
    Reply
  • bit_user
    The article said:
    When compared to Intel's Panther Lake, the A20 Pro is 48.7% faster in single-thread performance than the flagship Core Ultra X9 388H, yet the 16-core Panther Lake processor is 63% faster in multi-thread workloads.
    So, that tells me you looked at a Panther Lake with a single-core score of about 2694. I'll come back to this, shortly.

    Before that, I just wanted to remind folks that Geekbench MT is not a pure MT-scaling metric. Even if it were, comparing a phone to a laptop CPU is putting you in a vastly different power & thermal envelope, so of course Panther Lake should do much better on MT. If you looked at a pure MT scaling metric, like Cinebench or Blender, the difference should be even greater.

    Now, back to the single-threaded. We really need to look at the sub-scores, to have any chance of understanding what's really going on with this chip. That's because some of the individual benchmarks can use features like Apple's AMX (which is similar but distinct from Intel's, which Panther Lake lacks). So, the reason I wanted to pin down what Panther Lake we should compare it to is so I could find an entry in Geekbench's database for doing that detailed comparison. I'm actually going with a higher one, which is 2737. I don't know of a way to sort search results in the GB database, so that's just the highest on the first page of hits I got. I wanted to be sure I was looking at an entry at least as good as what the article's author cited. So, here's the detailed comparison:
    https://browser.geekbench.com/v7/cpu/compare/263870?baseline=316653
    One thing to note about that results page is that the Difference column doesn't mean what I'd have assumed. It's not 100 * (B / A - 1) or even B - A. Rather, it's simply A / B. In other words, the Panther Lake is equal to the A20 when the "Difference" column is 100%! The greatest discrepancy will be where the "Difference" is lowest. This insanity makes me want to slap someone at Geekbench.

    To get a handle on this, I've put the values into a spreadsheet and computed a "Speedup" score that's 100 * (B / A - 1). Then, I sorted by this quantity to easily find the highest and lowest, as well as any other patterns in the data.
    BenchmarkX9 388HA20 ProSpeedupClang2379466396.01%Video Player3026541678.98%HTML5 Browser2749449363.44%PDF Viewer2933457055.81%Structure from Motion3075464651.09%Navigation3126466849.33%Photo Library2581383948.74%Game Physics2709400447.80%Video Encoder2868415945.01%Ray Tracer2536359241.64%Asset Compression2689375639.68%Text Processing2878400339.09%HDR2538346736.60%File Compression2531334832.28%Audio Encoder2480295419.11%Photo Editor2835331016.75%
    So, we see that Clang (i.e. C/C++ compilation) is fastest. This is very impressive, since compilation isn't really something you can vectorize much. It might benefit from huge caches and low memory latency. But, a nearly 2x speedup from that stuff is still pretty astonishing! I think it's worth noting that "Text Processing" and "File Compression" didn't favor the A20 Pro nearly as much.

    I could speculate about the rest, but there's not an obvious pattern that jumps out at me, like a handful of outliers primarily responsible for the large difference in the averages. Indeed, the median speedup is about 46%, which aligns pretty well with how the means compare.

    I guess I'll be trying to find more details about the A20 Pro to see if I can get any more insight into these results.
    Reply
  • bit_user
    circadia said:
    if only there are better multi-platform benchmarks that properly test everything on iPhones, Androids, PCs, Macs whatever...
    Anandtech used SPEC CPU for this. It's based on real world programs, unlike Geekbench. Still, SPEC CPU is primarily oriented towards server and workstation apps, with some nod towards a desktop-relevant workloads.

    You can find single-threaded SPEC CPU results (what they call "rate=1") for a lot of different CPUs, here:
    https://benchview.hjc.im
    Note the benchmark selection box, at the very top of the page. Stick to the results that stipulate rate-1, for single-threaded.
    Reply
  • gg83
    bit_user said:
    So, that tells me you looked at a Panther Lake with a single-core score of about 2694. I'll come back to this, shortly.

    Before that, I just wanted to remind folks that Geekbench MT is not a pure MT-scaling metric. Even if it were, comparing a phone to a laptop CPU is putting you in a vastly different power & thermal envelope, so of course Panther Lake should do much better on MT. If you looked at a pure MT scaling metric, like Cinebench or Blender, the difference should be even greater.

    Now, back to the single-threaded. We really need to look at the sub-scores, to have any chance of understanding what's really going on with this chip. That's because some of the individual benchmarks can use features like Apple's AMX (which is similar but distinct from Intel's, which Panther Lake lacks). So, the reason I wanted to pin down what Panther Lake we should compare it to is so I could find an entry in Geekbench's database for doing that detailed comparison. I'm actually going with a higher one, which is 2737. I don't know of a way to sort search results in the GB database, so that's just the highest on the first page of hits I got. I wanted to be sure I was looking at an entry at least as good as what the article's author cited. So, here's the detailed comparison:
    https://browser.geekbench.com/v7/cpu/compare/263870?baseline=316653
    One thing to note about that results page is that the Difference column doesn't mean what I'd have assumed. It's not 100 * (B / A - 1) or even B - A. Rather, it's simply A / B. In other words, the Panther Lake is equal to the A20 when the "Difference" column is 100%! The greatest discrepancy will be where the "Difference" is lowest. This insanity makes me want to slap someone at Geekbench.

    To get a handle on this, I've put the values into a spreadsheet and computed a "Speedup" score that's 100 * (B / A - 1). Then, I sorted by this quantity to easily find the highest and lowest, as well as any other patterns in the data.
    BenchmarkX9 388HA20 ProSpeedupClang2379466396.01%Video Player3026541678.98%HTML5 Browser2749449363.44%PDF Viewer2933457055.81%Structure from Motion3075464651.09%Navigation3126466849.33%Photo Library2581383948.74%Game Physics2709400447.80%Video Encoder2868415945.01%Ray Tracer2536359241.64%Asset Compression2689375639.68%Text Processing2878400339.09%HDR2538346736.60%File Compression2531334832.28%Audio Encoder2480295419.11%Photo Editor2835331016.75%

    So, we see that Clang (i.e. C/C++ compilation) is fastest. This is very impressive, since compilation isn't really something you can vectorize much. It might benefit from huge caches and low memory latency. But, a nearly 2x speedup from that stuff is still pretty astonishing! I think it's worth noting that "Text Processing" and "File Compression" didn't favor the A20 Pro nearly as much.

    I could speculate about the rest, but there's not an obvious pattern that jumps out at me, like a handful of outliers primarily responsible for the large difference in the averages. Indeed, the median speedup is about 46%, which aligns pretty well with how the means compare.

    I guess I'll be trying to find more details about the A20 Pro to see if I can get any more insight into these results.
    Almost 5ghz on a mobile chip is crazy right? Could it ever sustain the 4.9ghz in real world? How hot it must get! It does go into bigger profile devices, maybe they are using a Frore cooler or something?
    Reply
  • usertests
    gg83 said:
    Almost 5ghz on a mobile chip is crazy right? Could it ever sustain the 4.9ghz in real world? How hot it must get! It does go into bigger profile devices, maybe they are using a Frore cooler or something?
    It's crazy but it was expected. A19 Pro hit 4.26 GHz. 4.9 GHz is +15%. So it's line with the most optimistic gains you could make going from TSMC N3P to N2(P?). TSMC's N2 nodes have switched to GAAFETs from FinFETs.

    It's also known that the next Snapdragon flagship will hit 5 GHz:

    Qualcomm’s next Snapdragon flagship chip has 5 GHz Oryon CPU
    The upcoming Snapdragon chip will feature a new Oryon CPU that “is the first mobile CPU to reach 5 GHz.” For reference, the 8 Elite Gen 5 reached 4.74 GHz and the 8 Elite before it was at 4.47 GHz.
    Reply
  • bit_user
    gg83 said:
    Almost 5ghz on a mobile chip is crazy right? Could it ever sustain the 4.9ghz in real world? How hot it must get! It does go into bigger profile devices, maybe they are using a Frore cooler or something?
    It would be interesting to see how much power it uses for a single thread. You might be right that it drops down to something like 4.5 GHz, after a short time, but that's still more than fast enough to maintain a good performance lead.

    In a laptop form factor, it shouldn't need to throttle any more than comparable laptop chips of different makes. So, if we're looking at 1T performance as a predictor of what the M6 would do, that makes these results very relevant.
    Reply
  • ezst036
    Even with all of this performance, Apple still is capped at a certain market share because they are too bossy of a company, placing too many limits on user freedom within their software.

    IOW, performance isn't enough.
    Reply
  • bit_user
    ezst036 said:
    Even with all of this performance, Apple still is capped at a certain market share because they are too bossy of a company, placing too many limits on user freedom within their software.
    First, this is a phone SoC. iPhone has a pretty enviable marketshare, which I say as someone who's never owned nor wanted one.

    As for Macs, I tried to find a graph of marketshare. I think the ones people used to cite were from Statista, which are now behind a paywall, but they are indeed going up and to the right. I think the typical consumer doesn't care what OS their machine runs, beyond just wanting something they're familiar with. People like us are the exception, not the norm.

    As someone who has never owned a Mac, nor has any intention of getting one, the reason I find these announcements exciting is because they reveal what is possible. I also want to see someone keep the pressure on the rest of the industry not to get lazy.
    Reply