4 ms·

"Not only Youtubers test on Windows, virtually the whole tech reviewer community tested on Windows exclusively. And many concluded that Threadripper problems we

by Jhsto 8y ago

"Not only Youtubers test on Windows, virtually the whole tech reviewer community tested on Windows exclusively. And many concluded that Threadripper problems were due to memory problems, and only few blamed Windows for it." https://www.phoronix.com/forums/forum/phoronix/latest-phoronix-articles/1040679-a-look-at-the-windows-10-vs-linux-performance-on-amd-threadripper-2990wx?p=1040719#post1040719 https://www.phoronix.com/forums/forum/phoronix/latest-phoron... Still, over 2x increase in performance on Linux over Windows seems insane.

ksec 8y ago
There are reasons why it is called the WinTel alliance. Fortunately Linux dominate the server market now, hopefully this will give AMD some breathing space.
- api 8y ago
  Never ascribe to conspiracy what can equally well be explained by stupidity... or in this case a broken scheduler.
  - cptskippy 8y ago
    Broken or optimized for a different set of processors? Don't forget that AMD can submit changes directly to the Linux Kernel but has to request or help Microsoft to make changes to Windows.
    - api 8y ago
      Oh I'm sure they're optimized for Intel, but that's easily explained by Intel's near-monopoly for the past 10 years. Now that AMD is gaining market share again we'll probably see this change. MS doesn't want to look bad in benchmarks.
      - cptskippy 8y ago
        We saw a similar situation when AMD released the Bulldozer line of CPUs and it took a minute for Microsoft to release a Hot Fix to address the issue. https://www.anandtech.com/show/5448/the-bulldozer-scheduling-patch-tested https://www.anandtech.com/show/5448/the-bulldozer-scheduling... The reality is that Windows doesn't recognize the new CPU and so it's not using a scheduler optimized for it. Where as AMD could have upstreamed any necessary changes for Linux to fully support it months ago.
    - mmrezaie 8y ago
      And also we should keep in mind that in Kernel level improvements, Linux by far is agiler than Windows. Windows process scheduler does not have fairness for ages now. When was CFS was added to Linux? a decade?
      - cptskippy 8y ago
        Linux isn't free of performance regressions with new architecture. There were patches last year to better support Epyc's NUMA. https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/?id=051f3ca02e46432c0965e8948f00c07d8a2f09c0 https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-n... It just seems that Linux avoided any regressions this year but Microsoft didn't.
  - ksec 8y ago
    I wouldn't call it a conspiracy. But why would M$ optimise for NUMA, or AMD when 90% of their use cases and customer have no use of it?
    - rbanffy 8y ago
      Because 10% of their used base is about the same size as a first-world country.
rbanffy 8y ago
Windows is the first thing I blame ;-) In this case, it looks like the thread scheduler is trying to do something clever that doesn't quite work well with this specific CPU. Maybe it's improperly localizing threads to cores or causing more cache evictions than it should. Without seeing performance counters it's hard to guess.
- gsnedders 8y ago
  My guess would simply be the scheduler doesn't have knowledge of the NUMA architecture of the processor, and cannot therefore efficiently schedule tasks to avoid large memory access costs.
  - mmrezaie 8y ago
    One can confirm this which sounds true by profiling the benchmark tools on windows and Linux specifically, remote page loads and such.
  - caf 8y ago
    The firmware has long made the NUMA topology available to the OS through ACPI tables.
    - rbanffy 8y ago
      Has a NUMA topology like this ever been used with Windows? They could be misinterpreting it. They have quite a reputation for partial buggy implementations of ACPI-related stuff that end up becoming de-facto standards.
      - snuxoll 8y ago
        I think it's less a matter of misinterpreting the topology, and more that the Windows scheduler isn't prepared to handle the somewhat exotic topology; whereas the Linux scheduler seems to be more intelligent about handling the memory-less NUMA nodes.
Nexxxeh 8y ago
There's some additional interesting stuff in the comments. edwaleni points out that the Phoronix Test Suite uses a very outdated 7-Zip (16.02) and the current Windows client (18.05) includes a lot of changes since then. https://www.phoronix.com/forums/forum/hardware/processors-memory/1040679-a-look-at-the-windows-10-vs-linux-performance-on-amd-threadripper-2990wx?p=1040736#post1040736 https://www.phoronix.com/forums/forum/hardware/processors-me...

by Jhsto 8y ago

ksec 8y ago

There are reasons why it is called the WinTel alliance. Fortunately Linux dominate the server market now, hopefully this will give AMD some breathing space.

api 8y ago
Never ascribe to conspiracy what can equally well be explained by stupidity... or in this case a broken scheduler.
- cptskippy 8y ago
  Broken or optimized for a different set of processors? Don't forget that AMD can submit changes directly to the Linux Kernel but has to request or help Microsoft to make changes to Windows.
  - api 8y ago
    Oh I'm sure they're optimized for Intel, but that's easily explained by Intel's near-monopoly for the past 10 years. Now that AMD is gaining market share again we'll probably see this change. MS doesn't want to look bad in benchmarks.
    - cptskippy 8y ago
      We saw a similar situation when AMD released the Bulldozer line of CPUs and it took a minute for Microsoft to release a Hot Fix to address the issue. https://www.anandtech.com/show/5448/the-bulldozer-scheduling-patch-tested https://www.anandtech.com/show/5448/the-bulldozer-scheduling... The reality is that Windows doesn't recognize the new CPU and so it's not using a scheduler optimized for it. Where as AMD could have upstreamed any necessary changes for Linux to fully support it months ago.
  - mmrezaie 8y ago
    And also we should keep in mind that in Kernel level improvements, Linux by far is agiler than Windows. Windows process scheduler does not have fairness for ages now. When was CFS was added to Linux? a decade?
    - cptskippy 8y ago
      Linux isn't free of performance regressions with new architecture. There were patches last year to better support Epyc's NUMA. https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/?id=051f3ca02e46432c0965e8948f00c07d8a2f09c0 https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-n... It just seems that Linux avoided any regressions this year but Microsoft didn't.
- ksec 8y ago
  I wouldn't call it a conspiracy. But why would M$ optimise for NUMA, or AMD when 90% of their use cases and customer have no use of it?
  - rbanffy 8y ago
    Because 10% of their used base is about the same size as a first-world country.

rbanffy 8y ago

Windows is the first thing I blame ;-) In this case, it looks like the thread scheduler is trying to do something clever that doesn't quite work well with this specific CPU. Maybe it's improperly localizing threads to cores or causing more cache evictions than it should. Without seeing performance counters it's hard to guess.

gsnedders 8y ago
My guess would simply be the scheduler doesn't have knowledge of the NUMA architecture of the processor, and cannot therefore efficiently schedule tasks to avoid large memory access costs.
- mmrezaie 8y ago
  One can confirm this which sounds true by profiling the benchmark tools on windows and Linux specifically, remote page loads and such.
- caf 8y ago
  The firmware has long made the NUMA topology available to the OS through ACPI tables.
  - rbanffy 8y ago
    Has a NUMA topology like this ever been used with Windows? They could be misinterpreting it. They have quite a reputation for partial buggy implementations of ACPI-related stuff that end up becoming de-facto standards.
    - snuxoll 8y ago
      I think it's less a matter of misinterpreting the topology, and more that the Windows scheduler isn't prepared to handle the somewhat exotic topology; whereas the Linux scheduler seems to be more intelligent about handling the memory-less NUMA nodes.

Nexxxeh 8y ago

There's some additional interesting stuff in the comments. edwaleni points out that the Phoronix Test Suite uses a very outdated 7-Zip (16.02) and the current Windows client (18.05) includes a lot of changes since then. https://www.phoronix.com/forums/forum/hardware/processors-memory/1040679-a-look-at-the-windows-10-vs-linux-performance-on-amd-threadripper-2990wx?p=1040736#post1040736 https://www.phoronix.com/forums/forum/hardware/processors-me...