18-Way GPU Linux Benchmarks, Including The Radeon RX 460 & RX 470 On Open-Source

[Michael]

In addition, would it be possible to split the graphs, e.g., have the 460/950/960 and similar level cards have their own graph (at 1080p Medium) and have the 470/480/970/980/1060 have their own 1080p/1440p Ultra graph? I understand it's extra work, but it would mean that results were segregated better by product class.

If someone wants to submit/sponsor patches to PTS for adding that support together.... All the graphs otherwise are auto-generated, so would need to patch the grapher and stuff to do that.

[Michael]

Maybe I have that backwards then... I thought Heaven was newer. Hold on, will check...

Nope, looks like you are right and Heaven did come out first. I guess I just thought Heaven was newer because it seems to use more advanced graphics tech, but maybe it's just tesselation in Heaven vs. more detailed textures in Valley.

So yeah, I guess running both would be best and we'll try to figure out why Heaven runs faster than Valley. Thanks !

Valley is newer than Heaven, but heaven makes heavier use of tessellation.

[cj.wijtmans]

4K results are only interesting for the GTX 1080 and maybe 1070, it makes no sense to test everything in 4K. At most 2% Steam users have 4K monitors according to the Steam hardware survey.

and my CPU is never ever a bottleneck anyway. unless i am playing a game with lots of AI.

[starshipeleven]

4K results are only interesting for the GTX 1080 and maybe 1070, it makes no sense to test everything in 4K. At most 2% Steam users have 4K monitors according to the Steam hardware survey.

hmmm... 2% users with 4k monitors.... 0.9% linux users..... that's double the linux users right there.

Clearly all linux gamers have a 4k monitor, and also dualboot into windows, there is no other rational explanation.

[bridgman]

Clearly all linux gamers have a 4k monitor, and also dualboot into windows, there is no other rational explanation.

You missed the part about how we only run Windows for the Steam survey

[smitty3268]

Hey Michael, a couple suggestions. (not criticism)

1. I realize people will probably ask you to compare every card you have, but i actually think this would have been cleaner to show fewer. You could limit it to the other cards in it's class, for example, along with a few others like the 1060 to compare to nvidia's current architecture and maybe some higher-end last gen AMD gpus. But including outliers like the 1080 and 980ti isn't useful.

2. Reduce the tests to run at 1080p for this hardware. It's slow enough that it makes sense to do so, or at least 1440p. 4K just isn't going to run well on this hardware, and it doesn't make sense to do the comparisons there. Taking out the 1080 results per suggestion 1 might have made this more obvious to you, since it's the high-end nvidia cards that are capable of running 4k. Sub-30 fps tends to indicate the resolution is too high.

[M@GOid]

I opened a bug report for the regression performance of the R9 290 in kernel 4.7:

https://bugs.freedesktop.org/show_bug.cgi?id=97260

If you have been hit by a performance regression in kernel 4.7, please make a comment on the link above and if you weren't (IF you are running kernel 4.7), comment too, so we can understand if it is a kernel ou packaging problem.

[M@GOid]

Hey Michael, a couple suggestions. (not criticism)

1. I realize people will probably ask you to compare every card you have, but i actually think this would have been cleaner to show fewer. You could limit it to the other cards in it's class, for example, along with a few others like the 1060 to compare to nvidia's current architecture and maybe some higher-end last gen AMD gpus. But including outliers like the 1080 and 980ti isn't useful.

2. Reduce the tests to run at 1080p for this hardware. It's slow enough that it makes sense to do so, or at least 1440p. 4K just isn't going to run well on this hardware, and it doesn't make sense to do the comparisons there. Taking out the 1080 results per suggestion 1 might have made this more obvious to you, since it's the high-end nvidia cards that are capable of running 4k. Sub-30 fps tends to indicate the resolution is too high.

I support this. Limiting for price class will make Michel work less and open a opportunity to use older cards of the same class, like a HD6870 or a GTX750Ti, so people still running them can make a decision to upgrade or not.

[marek]

Valley is newer than Heaven, but heaven makes heavier use of tessellation.

No just "heavier". Valley doesn't use tessellation at all.

[dungeon]

xrandr -s 1920x1080

Exactly that should fix Xonotic too... if Michael don't want to change his 4K monitor, he can do this manually before running any FullHD benchmarks.