Announcement

Collapse
No announcement yet.

AMD Zen 4 AVX-512 Performance Analysis On The Ryzen 9 7950X

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • MadCatX
    replied
    Originally posted by coder View Post
    That was true of their 14 nm products. However, Ice Lake (10 nm+) then brought AVX-512 to laptops and Tiger Lake (10 nm++) continued this trend.
    True, but AVX512 in very power-constrained laptop chips is probably worth less that an "AVX512-ready!" sticker you could slap on a laptop with such a chip.

    Originally posted by coder View Post
    Rocket Lake then brought it to the desktop, but that was still 14 nm, because Intel's 10 nm++ process still couldn't deliver high clock frequencies or the yield needed for desktop products.
    It's quite difficult to draw some exact conclusions but RKL's AVX512 seems worse than Zen4 at least in perf-per-watt. (https://www.phoronix.com/review/rocket-lake-avx512)

    Originally posted by coder View Post
    ‚Äč
    It's not until Alder Lake that Intel removed AVX-512 from consumer desktop & laptop platforms, and that was largely seen as a necessary compromise of their decision to go Big.Little. Some people debate whether it was truly necessary, but show me another Big.Little CPU with heterogeneous ISA.
    I'm aware of the technical intricacies here but nevertheless it's a step back. Unless Intel fixes this for Raptor Lake, they could be at risk at looking even worse in some benchmarks if more consumer-grade software adopts AVX512. Now that we have chips with "no compromises" AVX512 implementations, it would make more sense to finally do that.

    Leave a comment:


  • coder
    replied
    Originally posted by MorrisS. View Post
    So AVX5 yields a significant increase.
    Calling it that implies there was an AVX3 and AVX4. No, it's called AVX-512. It's inconsistent, but the reality of AVX-512 is that it's a different family of ISA extensions, just like AVX/AVX2 are a different family than the SSE/SSE2/SSE3/SSE4x extensions. Each family differs from its predecessor in more ways than mere register width.

    Leave a comment:


  • coder
    replied
    Originally posted by MadCatX View Post
    Let me see if I got this right. Intel introduces AVX512, limits it to super-duper expensive server-grade CPUs and makes it a tricky tradeoff between performance, power consumption and performance-per-watt.
    That was true of their 14 nm products. However, Ice Lake (10 nm+) then brought AVX-512 to laptops and Tiger Lake (10 nm++) continued this trend.

    Rocket Lake then brought it to the desktop, but that was still 14 nm, because Intel's 10 nm++ process still couldn't deliver high clock frequencies or the yield needed for desktop products.

    It's not until Alder Lake that Intel removed AVX-512 from consumer desktop & laptop platforms, and that was largely seen as a necessary compromise of their decision to go Big.Little. Some people debate whether it was truly necessary, but show me another Big.Little CPU with heterogeneous ISA.

    Leave a comment:


  • MadCatX
    replied
    Let me see if I got this right. Intel introduces AVX512, limits it to super-duper expensive server-grade CPUs and makes it a tricky tradeoff between performance, power consumption and performance-per-watt. AMD has AVX512 that pretty much just works. Intel needed a massive slap on the face after the 25th refresh of WeDontCareLake but this is getting ridiculous.

    Michael: Is there any chance you could do a quick round of Zen4/ADL AVX512 benchmarks and compare the relative benefits of AVX512 on both uarchs?

    Leave a comment:


  • MorrisS.
    replied
    Originally posted by coder View Post
    Hopefully, this spurs the adoption of glibc's hwcaps feature by more libraries & distros.
    So AVX5 yields a significant increase.

    Leave a comment:


  • aufkrawall
    replied
    The videos used with dav1d aren't 10 bit? Performance gains should be much bigger vs. 8 bit video.

    Leave a comment:


  • bridgman
    replied
    Originally posted by schmidtbag View Post
    Pretty good results for what is effectively a half-baked solution. I'm impressed the power draw didn't change much despite the performance increase.
    It's not half-baked, it's half-sized and perfectly baked

    In fairness we do also have multiple FP execution ports/pipes (6 vs 3 for Golden Cove) so there can still be a lot of work happening in parallel.
    Last edited by bridgman; 26 September 2022, 03:43 PM.

    Leave a comment:


  • schmidtbag
    replied
    Pretty good results for what is effectively a half-baked solution. I'm impressed the power draw didn't change much despite the performance increase.

    Leave a comment:


  • coder
    replied
    Originally posted by ms178 View Post
    Finally, AVX-512 on a mainstream CPU people actually want to buy.
    Hopefully, this spurs the adoption of glibc's hwcaps feature by more libraries & distros.

    Leave a comment:


  • coder
    replied
    Originally posted by MorrisS. View Post
    A test between AVX2 and AVX5 performance would be useful.
    I believe most of these tests are using AVX2, when AVX-512 is disabled.

    Leave a comment:

Working...
X