4 ms·

I think people generally expected larger improvements between iterations. Intel and AMD continue to deliver sizeable performance and efficiency gains every 1-2

by Sakos 2y ago

I think people generally expected larger improvements between iterations. Intel and AMD continue to deliver sizeable performance and efficiency gains every 1-2 years while it feels like the Apple M-series isn't getting comparable gains. It definitely seems like Apple has suffered significant enough brain drain in recent years that they're finding it difficult to iterate on the M1.

TylerE 2y ago
Hasn't each gen been like 20-30% performance? Isn't that better than what AMD and Intel have managed?
- ben_w 2y ago
  20-30% per year may (I'm not sure) be better than what AMD and Intel have managed since the M1 came out, but it's definitely not as good as the 40-60% per year improvements that I remember when I was a kid. I'm anchored on those historical performance gains, so I'm naturally (and, I recognise, unreasonably) disappointed by "only" 20-30% improvements from Apple.
- Sakos 2y ago
  Not really? https://arstechnica.com/gadgets/2023/11/testing-apples-m3-pro-more-efficient-but-performance-is-a-step-sideways/ https://arstechnica.com/gadgets/2023/11/testing-apples-m3-pr... I've been waiting for an upgrade to my M1 and I still haven't seen one worth spending that much money on. I'd rather just sink that into upgrading my Windows tower.
  - bhouston 2y ago
    https://docs.google.com/spreadsheets/d/1i5dBe_rsiNaQATH-D-Pj3u1h7bSu4M8GMR9vC-mdgpo/edit?gid=0#gid=0 https://docs.google.com/spreadsheets/d/1i5dBe_rsiNaQATH-D-Pj... Single core performance went from 2300 with the M1 to 3800 with the M4. That is a huge improvement for my workflow (large TypeScript mono repositories) which is dependent a lot of single core performance (even with parallel builds, because one rarely re-builds everything, rather hot reloads.)
    - zamadatix 2y ago
      These statements are orthogonal though i.e. 2300->3800 is still less than 20% per generation (17% per if you use the exact single core numbers for the M1 vs M4 iPad). That might be meaningful for your workload but it also means 20-30 percent per generation is quite a bit off.
      - ksec 2y ago
        I am sorry I have to say this out loud. But seriously? We expect 20% per generation? When the word generation used to mean 2 years in CPU terms because that was what Moore's law or Intel's tick tock dictate. Even ignoring the word generation, For the past 20 years how many times did we see a successful uArch bringing 20% IPC improvement over the current leading IPC? Where did that 20-30% per gen coming from. The worst part about Apple, its media is that it seems we have a whole generation of people who were never interested in CPU performance suddenly coming up with these expectation and figures.
        zamadatix 2y ago
        On the contrary, _not_ expecting at least 20% per year (more for generations farther apart) is actually only a recent take for CPUs [1][2] which formed in what I like to call "the great stagnation" of the mid 2010s where AMD bombed, ARM was still a low performance mobile play, and Intel didn't have any stiff competition. After that things have started to pick up a little again now that there is innovation on said fronts [3] I remember there was even a period Moore's law was commonly conflated with "a doubling in performance every 2 years" (implies people being used to an increase of at least 41% per year) instead of a doubling in transistor count. Because Intel's tick-tock model started ~5 years prior to the great stagnation people commonly claimed it's the end of Moore's law and that's why we have tick-tock's now. Of course these days it's common knowledge that Moore's law (the actual transistor count version) has been holding steady all these years still and that the performance lull wasn't related to hitting and innate technical scaling barriers. Marketing departments are definitely going to be marketing departments but they aren't the origin of the idea CPUs can have more than minor increments in performance each year. [1] https://preshing.com/images/integer-perf.png https://preshing.com/images/integer-perf.png [2] https://preshing.com/images/float-point-perf.png https://preshing.com/images/float-point-perf.png [3] https://i.imgur.com/HlGbJE9.png https://i.imgur.com/HlGbJE9.png (note this chart is linear not log)
  - bee_rider 2y ago
    CPUs got good enough for most applications a decade ago, it is hard to talk about upgrades being “worth spending money on” without specific workload info.
- bhouston 2y ago
  I've been tracking the performance increase via GeekBench for M1, M2, M3 and now M4 and they are good incremental and consistent improvements: https://docs.google.com/spreadsheets/d/1i5dBe_rsiNaQATH-D-Pj3u1h7bSu4M8GMR9vC-mdgpo/edit?gid=0#gid=0 https://docs.google.com/spreadsheets/d/1i5dBe_rsiNaQATH-D-Pj...
  - ksec 2y ago
    Thank You for the Google Spreadsheets. I did my comparison on my Apple Numbers, dont know why it never occurred to me I could do it on Google Spreadsheets and shared it out on the web. For M1 - M3 the IPC improvement were minimal. Absolute performance came from Node and Clock Speed improvement. And for M4, most of the single core improvement actually came from floating point improvements. If we look into each test results and in terms of Integer I remember last time I checked it was less than 5%. In total, M1 to M4 has less than 10% of IPC improvement in terms of integer performance. This is not to say I am disappointed or surprised like all others. A lot of the CPU improvement went from A12 to M1 / A14 which was barely showcased by Apple developer toolkit during ARM transition. The ultra wide design which Apple were perfecting for many years before it reached M1, has reached a plateau. I am now wondering if there could be some other Integer IPC improvement, but judging from M4 we will likely have to wait until 2026 before we see anything new in terms of CPU uArch design.
- zamadatix 2y ago
  "Up to 20-30% faster" or "20-30% faster than <couple generation old version you might upgrade from". Comparing the "plain" M1 model vs M4 model (not pro/max as those aren't out yet, though follow nearly identical trends so far) there has been a total ~56% single core and ~59% multi-core performance uplift, or an average of ~16% per M* generation. M1 released in November 2020. In almost the exact time frame AMD went from Zen3 -> Zen 5 (so one less generation) with a total gain of ~34% multicore and ~59% multicore gain. Intel went from ~10900k to ~14900k for a total of 76% single core and 128% multicore gain (i.e. more than double). Two disclaimers before the conclusion: This is just from one lazy benchmark (Geekbench 6) and of the high performing model. That doesn't necessarily tell the whole story on both accounts, but it at least gives some context around general trends. E.g. passmark is going to say the differences on Intel are a little smaller, comparisons between Max and non-Max generations confuse multi core growth, and changes in low-mid market chips may follow a different slope than the high end products. Also there are other considerations like "and what about integrated graphics performance, which eats up a huge amount of die space and power budget?" Anyways, my conclusion is that Apple is doing reasonably well with generational improvements vs the competition. Maybe not the absolute best, but not the worst. Being on top in the single core realm with a mobile passively cooled device makes equivalent gains all the more difficult but they're still doing it. Apple may be a victim of its own success in that the M1 from 2020 is still such a performance powerhouse that a 50% gain 4 generations later just isn't that interesting whereas with Intel a catchup of ~doubling multicore performance in the same time seems more impressive especially when M*'s story isn't as tantalizing on that half of things.
  - ggus 2y ago
    > ~59% multi-core performance uplift, or an average of ~16% per M* generation. +59% in 4 years is equivalent to 12% every year. percentages compound exponentially.
    - phonon 2y ago
      What are you correcting? 59% from M1->M4 = 1.59^⅓ = 16.7% increase per generation No one said "years" either...
mrbungie 2y ago
I'm still waiting for chips to compete with the perf/watt ratios and general efficiency of the M-series chips. Snapdragon X Elite + Windows ARM is getting there, but their GPU perf leaves a lot to be desired. Is Apple a victim of their own success? Idk, but diminishing returns are a thing you know.

by Sakos 2y ago

TylerE 2y ago

Hasn't each gen been like 20-30% performance? Isn't that better than what AMD and Intel have managed?

ben_w 2y ago
20-30% per year may (I'm not sure) be better than what AMD and Intel have managed since the M1 came out, but it's definitely not as good as the 40-60% per year improvements that I remember when I was a kid. I'm anchored on those historical performance gains, so I'm naturally (and, I recognise, unreasonably) disappointed by "only" 20-30% improvements from Apple.
Sakos 2y ago
Not really? https://arstechnica.com/gadgets/2023/11/testing-apples-m3-pro-more-efficient-but-performance-is-a-step-sideways/ https://arstechnica.com/gadgets/2023/11/testing-apples-m3-pr... I've been waiting for an upgrade to my M1 and I still haven't seen one worth spending that much money on. I'd rather just sink that into upgrading my Windows tower.
- bhouston 2y ago
  https://docs.google.com/spreadsheets/d/1i5dBe_rsiNaQATH-D-Pj3u1h7bSu4M8GMR9vC-mdgpo/edit?gid=0#gid=0 https://docs.google.com/spreadsheets/d/1i5dBe_rsiNaQATH-D-Pj... Single core performance went from 2300 with the M1 to 3800 with the M4. That is a huge improvement for my workflow (large TypeScript mono repositories) which is dependent a lot of single core performance (even with parallel builds, because one rarely re-builds everything, rather hot reloads.)
  - zamadatix 2y ago
    These statements are orthogonal though i.e. 2300->3800 is still less than 20% per generation (17% per if you use the exact single core numbers for the M1 vs M4 iPad). That might be meaningful for your workload but it also means 20-30 percent per generation is quite a bit off.
    - ksec 2y ago
      I am sorry I have to say this out loud. But seriously? We expect 20% per generation? When the word generation used to mean 2 years in CPU terms because that was what Moore's law or Intel's tick tock dictate. Even ignoring the word generation, For the past 20 years how many times did we see a successful uArch bringing 20% IPC improvement over the current leading IPC? Where did that 20-30% per gen coming from. The worst part about Apple, its media is that it seems we have a whole generation of people who were never interested in CPU performance suddenly coming up with these expectation and figures.
      - zamadatix 2y ago
        On the contrary, _not_ expecting at least 20% per year (more for generations farther apart) is actually only a recent take for CPUs [1][2] which formed in what I like to call "the great stagnation" of the mid 2010s where AMD bombed, ARM was still a low performance mobile play, and Intel didn't have any stiff competition. After that things have started to pick up a little again now that there is innovation on said fronts [3] I remember there was even a period Moore's law was commonly conflated with "a doubling in performance every 2 years" (implies people being used to an increase of at least 41% per year) instead of a doubling in transistor count. Because Intel's tick-tock model started ~5 years prior to the great stagnation people commonly claimed it's the end of Moore's law and that's why we have tick-tock's now. Of course these days it's common knowledge that Moore's law (the actual transistor count version) has been holding steady all these years still and that the performance lull wasn't related to hitting and innate technical scaling barriers. Marketing departments are definitely going to be marketing departments but they aren't the origin of the idea CPUs can have more than minor increments in performance each year. [1] https://preshing.com/images/integer-perf.png https://preshing.com/images/integer-perf.png [2] https://preshing.com/images/float-point-perf.png https://preshing.com/images/float-point-perf.png [3] https://i.imgur.com/HlGbJE9.png https://i.imgur.com/HlGbJE9.png (note this chart is linear not log)
- bee_rider 2y ago
  CPUs got good enough for most applications a decade ago, it is hard to talk about upgrades being “worth spending money on” without specific workload info.
bhouston 2y ago
I've been tracking the performance increase via GeekBench for M1, M2, M3 and now M4 and they are good incremental and consistent improvements: https://docs.google.com/spreadsheets/d/1i5dBe_rsiNaQATH-D-Pj3u1h7bSu4M8GMR9vC-mdgpo/edit?gid=0#gid=0 https://docs.google.com/spreadsheets/d/1i5dBe_rsiNaQATH-D-Pj...
- ksec 2y ago
  Thank You for the Google Spreadsheets. I did my comparison on my Apple Numbers, dont know why it never occurred to me I could do it on Google Spreadsheets and shared it out on the web. For M1 - M3 the IPC improvement were minimal. Absolute performance came from Node and Clock Speed improvement. And for M4, most of the single core improvement actually came from floating point improvements. If we look into each test results and in terms of Integer I remember last time I checked it was less than 5%. In total, M1 to M4 has less than 10% of IPC improvement in terms of integer performance. This is not to say I am disappointed or surprised like all others. A lot of the CPU improvement went from A12 to M1 / A14 which was barely showcased by Apple developer toolkit during ARM transition. The ultra wide design which Apple were perfecting for many years before it reached M1, has reached a plateau. I am now wondering if there could be some other Integer IPC improvement, but judging from M4 we will likely have to wait until 2026 before we see anything new in terms of CPU uArch design.
zamadatix 2y ago
"Up to 20-30% faster" or "20-30% faster than <couple generation old version you might upgrade from". Comparing the "plain" M1 model vs M4 model (not pro/max as those aren't out yet, though follow nearly identical trends so far) there has been a total ~56% single core and ~59% multi-core performance uplift, or an average of ~16% per M* generation. M1 released in November 2020. In almost the exact time frame AMD went from Zen3 -> Zen 5 (so one less generation) with a total gain of ~34% multicore and ~59% multicore gain. Intel went from ~10900k to ~14900k for a total of 76% single core and 128% multicore gain (i.e. more than double). Two disclaimers before the conclusion: This is just from one lazy benchmark (Geekbench 6) and of the high performing model. That doesn't necessarily tell the whole story on both accounts, but it at least gives some context around general trends. E.g. passmark is going to say the differences on Intel are a little smaller, comparisons between Max and non-Max generations confuse multi core growth, and changes in low-mid market chips may follow a different slope than the high end products. Also there are other considerations like "and what about integrated graphics performance, which eats up a huge amount of die space and power budget?" Anyways, my conclusion is that Apple is doing reasonably well with generational improvements vs the competition. Maybe not the absolute best, but not the worst. Being on top in the single core realm with a mobile passively cooled device makes equivalent gains all the more difficult but they're still doing it. Apple may be a victim of its own success in that the M1 from 2020 is still such a performance powerhouse that a 50% gain 4 generations later just isn't that interesting whereas with Intel a catchup of ~doubling multicore performance in the same time seems more impressive especially when M*'s story isn't as tantalizing on that half of things.
- ggus 2y ago
  > ~59% multi-core performance uplift, or an average of ~16% per M* generation. +59% in 4 years is equivalent to 12% every year. percentages compound exponentially.
  - phonon 2y ago
    What are you correcting? 59% from M1->M4 = 1.59^⅓ = 16.7% increase per generation No one said "years" either...

mrbungie 2y ago

I'm still waiting for chips to compete with the perf/watt ratios and general efficiency of the M-series chips. Snapdragon X Elite + Windows ARM is getting there, but their GPU perf leaves a lot to be desired. Is Apple a victim of their own success? Idk, but diminishing returns are a thing you know.