When TPU jumpted the reviews, I was surprised that on the apps side only really WinRAR performed much better. I do suspect that the build-in 7zip benchmark (easy to use) does not see any benefit because of the settings: default is a 32MB dictionary whereas for solid archives larger ones tend to get better results. Default 7z Ultra setting is to use a 64MB dictionary but the memory usage estimate is far above that:
I suspect to really take advantage of the huge cache will require developer work.
From the Phoronix reviews of Milan-X:
https://www.phoronix.com/scan.php?page=article&item=amd-epyc-7773x-linux&num=1
and the kernel 5.18 follow-up:
https://www.phoronix.com/scan.php?page=article&item=amd-milanx-linux518&num=1
The kernel made big difference. And while their reviews stick to their standard test suite, being able to re-compile (which I assume they do) makes a big difference.
Obviously, HPC workloads tend to be used to having to re-compile. Although even then I would think more can be squeezed out of certain workloads rather than just telling the compiler that the target has lots of cache.
Bethesda's Creation Engine will an interesting to see results, we just need a volunteer:
Will interested to see your results. I would also appreciate posting in
@CAT-THE-FIFTH's FO4 benchmark thread:
https://forums.overclockers.co.uk/t...hmark-thread-need-some-zen3-results.18946938/
Yes, the power figures in the CB review are pretty telling. We all know the 12900KS is the Special-power-hug-Editon, but since 5800X3D uses less power than the 5800X due to the lower clocks, the 12900KS uses closer to
three times the power:
https://www.computerbase.de/2022-04...2/#abschnitt_die_leistungsaufnahme_in_spielen
58W vs 151W is a very poor showing for Intel's KS model. And the extra 46W compared the normal 'K' gains them at most 3% at 720P (and even a regression in RT performance for whatever reason). Guess power usage is no longer important.
