The Very Latest RadeonSI Git Code Provides Minor Benefits For Total War: WARHAMMER

[phoronix]

Phoronix: The Very Latest RadeonSI Git Code Provides Minor Benefits For Total War: WARHAMMER

Feral Interactive released Total War: WARHAMMER for Linux this week. On launch-day we provided NVIDIA Linux benchmarks as well as RadeonSI GPU benchmarks for this game over many different GPUs. Also landing on launch-day in Mesa Git were support for compiling optimized shader variants asynchronously in RadeonSI. So here are some benchmarks with the very newest Git to show the performance difference, which some have claimed is up to 25% faster...

http://www.phoronix.com/scan.php?page=news_item&px=TW-WARHAMMER-Fresh-Git

[FireBurn]

Zzzzz

[dungeon]

Zzzzz

Related/unreleated... Wake Up, FireBurn

[Pontostroy]

I have not seen the numbers, but the difference between the first and second benchmark launch can be 10 fps.

[cj.wijtmans]

I have not seen the numbers, but the difference between the first and second benchmark launch can be 10 fps.

shader cache?

[Pontostroy]

shader cache?

in memory shader cache

[droste]

I have not seen the numbers, but the difference between the first and second benchmark launch can be 10 fps.

AFAIK pts runs the test multiple times to make sure to compensate such things. IIRC it even makes sure that the standard deviation isn't too large and does additional runs if that's the case.

[tomtomme]

AFAIK pts runs the test multiple times to make sure to compensate such things. IIRC it even makes sure that the standard deviation isn't too large and does additional runs if that's the case.

yeah right.
but I think this time Michael did not run multiple times.
Usually there is a "SE +-Number" below the name of the card. SE would be standard error afaik. This is missing. So maybe Michael did disable the multiple runs this time?
Michael : could you confirm this and rerun if so?

[Michael]

yeah right.
but I think this time Michael did not run multiple times.
Usually there is a "SE +-Number" below the name of the card. SE would be standard error afaik. This is missing. So maybe Michael did disable the multiple runs this time?
Michael : could you confirm this and rerun if so?

As explained in the original article, this game doesn't have ability for automated benchmarking from the cli or scripted, so no PTS support, just hooked into it for graphing.

[mannerov]

Was the test with llvm git ? Some people report lower performance, probably safer to use last llvm stable.