LLVMpipe & OpenSWR OpenGL Riding Off Threadripper

[LinuxID10T]

They need to improve multicore scaling first, *massively*. llvmpipe doesn't even scale properly to 4 cores.

Yep, my stock i3-6100 is faster than my Ryzen 7 1700 overclocked to 3.8 GHz.

[LinuxID10T]

Interesting. I couldn't find any common tests between the systems (OpenArena had different resolution, Xonotic had different resolution and quality) but what struck me was the big jump in frame rates for Xeon between Mesa 17 and 17.3 with no corresponding jump for Threadripper. Looks like some llvmpipe changes might have gone in between the versions which were more Intel-specific than intended.

I think it has to do with AVX support.

[linuxgeex]

interesting. I couldn't find any common tests between the systems (openarena had different resolution, xonotic had different resolution and quality) but what struck me was the big jump in frame rates for xeon between mesa 17 and 17.3 with no corresponding jump for threadripper. Looks like some llvmpipe changes might have gone in between the versions which were more intel-specific than intended.

AVX-512.

Michael I've edited that to AVX-512 (caps) 6 times and it keeps getting written as lowercase?!

And now it finally lets me save it uppercase. bizarre.

And I just noticed that it lcase'd the first word of bridgman's quote as well.

[linuxgeex]

avx-512.

Michael i've edited that to avx-512 (caps) 6 times and it keeps getting written as lowercase?!

And now it finally lets me save it uppercase. Bizarre.

And i just noticed that it lcase'd the first word of bridgman's quote as well.

avx-512

[linuxgeex]

Avx-512

[linuxgeex]

Avx-512

post 973170 I cannot edit it to caps, and note that the quote converted the content to lowercase!
post 973172 I cannot edit it to caps but it is forcing it to title-case?!

AFAIK this is new... time to roll back the code.

[bridgman]

AVX-512.

Geez, I hope not... that would drop out most of Intel's CPUs as well. Will go back and look at it again, thanks...

I thought the new code was using lower AVX levels where Ryzen and Intel CPUs have comparable support (AFAIK Ryzen is faster in common AVX operations because it has 4 128-bit FPUs vs 2 256-bit FPUs but Intel is faster for 256-bit FMAC).

[Kemosabe]

They need to improve multicore scaling first, *massively*. llvmpipe doesn't even scale properly to 4 cores.

This is concerning ... :-(

[coder]

Is nobody else bothered by the similarity of the software results? Both this & the 40-core test, in fact. It seems unlikely that they're so closely optimized to the limits of the hardware. More likely that all tests are using the same software backend.

[coder]

Very interesting. I'm sure that with a lot of optimization work a purely software renderer could become very competitive against low-end graphic cards.

No, but I wonder if the gap would be reduced by using games with more sophisticated graphics. I'm not very familiar with the titles tested, but suspect they're mostly fill-rate limited.