By: Travis (travis.downs.delete@this.gmail.com), December 27, 2017 3:18 pm
Room: Moderated Discussions
-.- (someone.delete@this.example.com) on December 27, 2017 2:53 am wrote:
> Travis (travis.downs.delete@this.gmail.com) on December 20, 2017 7:52 pm wrote:
> > Travis (travis.downs.delete@this.gmail.com) on December 20, 2017 1:44 pm wrote:
> >
> > Test code is available. Anyone with a recent x86 box, I'm interested in your results.
>
> Haven't done a lot of reading into this thread or the code, or how to exactly interpret the results, but if
> it's of any use, here's my results on Skylake-X (i7 7820X frequency locked at 3.6GHz, mesh at 2.4GHz).
>
> Results: https://pastebin.com/PBg3QQwF
> Image for asm: https://i.imgur.com/pblYgX8.png
Thank you! They are very useful indeed.
The overall pattern looks similar, except there is no evidence of "slow mode" - i.e., there are no outliers were the performance suddenly becomes twice as bad (this is what turned me on to this issue in the first place).
I wonder if it's something actually fixed in SKX, or if it is an effect of turning off the LSD or something (I could write a bigger loop that wouldn't fit in the LSD to test that theory).
The other three tests are all quite consistent with my SKL hardware.
> Travis (travis.downs.delete@this.gmail.com) on December 20, 2017 7:52 pm wrote:
> > Travis (travis.downs.delete@this.gmail.com) on December 20, 2017 1:44 pm wrote:
> >
> > Test code is available. Anyone with a recent x86 box, I'm interested in your results.
>
> Haven't done a lot of reading into this thread or the code, or how to exactly interpret the results, but if
> it's of any use, here's my results on Skylake-X (i7 7820X frequency locked at 3.6GHz, mesh at 2.4GHz).
>
> Results: https://pastebin.com/PBg3QQwF
> Image for asm: https://i.imgur.com/pblYgX8.png
Thank you! They are very useful indeed.
The overall pattern looks similar, except there is no evidence of "slow mode" - i.e., there are no outliers were the performance suddenly becomes twice as bad (this is what turned me on to this issue in the first place).
I wonder if it's something actually fixed in SKX, or if it is an effect of turning off the LSD or something (I could write a bigger loop that wouldn't fit in the LSD to test that theory).
The other three tests are all quite consistent with my SKL hardware.