24-Way AMD Radeon vs. NVIDIA GeForce Linux Graphics Card Comparison

[Michael]

Someone needs to write the PTS code, hint hint nudge nudge: http://phoronix.com/forums/showthrea...in-all-GL-apps

Otherwise the only frame times are from engines that report it, ie OA/quake derivatives.

Should be easy for anyone to do considering PTS already supports generating line graphs out of frame times (for engines that expose it), etc.

In terms of box plots, PTS also supports generating them. IIRC, the reason I don't use them by default in the UI is I couldn't make them look really nice / graphically pleasing,

[curaga]

Someone needs to write the PTS code, hint hint nudge nudge: http://phoronix.com/forums/showthrea...in-all-GL-apps

Otherwise the only frame times are from engines that report it, ie OA/quake derivatives.

[FourDMusic]

i feel like things like _min frame rate_ ie graph frame rate analysis, is also important. Avg fps doesn't show things like microstutter or other problems encountered with gpus.

For instance on the pc, i get solid 60fps with rage but the game stutters like a bitch because it caches from the hard drive. Even the 64bit version of the game (i have 8gb of memory) yet people with ssd don't have this problem. I remember fallout games having a similar issue with anti-aliasing enabled as well. I think therefore it's important to include the frame rate graph showing min frame rate, max and medium.

box plots ftw!!!

[b15hop]

I feel like things like _Min frame rate_ ie graph frame rate analysis, is also important. Avg fps doesn't show things like microstutter or other problems encountered with GPUs.

For instance on the PC, I get solid 60fps with RAGE but the game stutters like a bitch because it caches from the hard drive. Even the 64bit version of the game (I have 8GB of memory) yet people with SSD don't have this problem. I remember fallout games having a similar issue with anti-aliasing enabled as well. I think therefore it's important to include the frame rate graph showing min frame rate, max and medium.

[brosis]

Michael, excuse me please if it was already answered, but why did you use tests with different settings?

What I mean is - virtually none of the results intersect with earlier 25-way opensource comparison, but both are done on near identical hardware.
The closest test you done was Xonotic, but opensource got tested on "high" topmost, where closed-source starts from "ultra".

The only thing I noticed is that opensource driver has much more concentrated fps distribution, where catalyst jumps everywhere between 40 and 500 fps, coming with less than middle (for example 140 fps) average. This is observed by comparing Xonotic high vs ultra sections of the articles.

Thanks!

[Herem]

Read the article why it wasn't tested...

That'll teach me to read the article rather than scanning the results.

[Michael]

@Michael don't you have the R290 anymore? It would be interesting to see if the performance is getting any better yet.

Read the article why it wasn't tested...

[Herem]

@Michael don't you have the R290 anymore? It would be interesting to see if the performance is getting any better yet.

[mmstick]

The closed-source drivers can handle Unigine Valley/Heaven and other more demanding tests where as the open-source ones can't (at least not without overriding environment variables and other non-default tweaks and likely rendering issues).

That doesn't mean you can't run some of the same benchmarks for both. Besides, Valley runs fine without rendering issues with just adding two export variables before launching it. Nonetheless, that wasn't the point of my comment.

[smitty3268]

The closed-source drivers can handle Unigine Valley/Heaven and other more demanding tests where as the open-source ones can't (at least not without overriding environment variables and other non-default tweaks and likely rendering issues).

There are no rendering issues. You just have to override the GLSL version, or wait for the GL3.3 support to be committed to master, which looks like it should be coming very soon now.