Radeonsi is awesome, beats Catalyst!

[darkbasic]

I did some benchmarks of my AMD Radeon HD 7950 using kernel 3.15-rc4 + PTE patches (VRAM page table entry compression) + hyperz (R600_DEBUG=hyperz). I?m also using libdrm git, xf86-video-ati git, llvm 3.5 git, mesa git (OpenGL core profile version string: 3.3 (Core Profile) Mesa 10.3.0-devel (git-cf93f86)) and Keith Packard?s xorg-server glamor-server branch (1.16.0 RC 2). Catalyst version is 14.4 (kernel 3.14.3, xorg-server 1.15.1).

http://www.linuxsystems.it/2014/05/radeonsi-awesome-beats-catalyst/

[0xBADCODE]

For me RadeonSI was a bit worse experience on R9 270 GPU so far.
1) Xonotic is unable to use 1920x1080 resolution - display shows "no signal" if I try to use this resolution. On other hand 1920x1200 works fine (native display resolution is 2560x1440). But wait, HD5770 can set up 1920x1080 on the very same display. WTF? I guess RadeonSI does something wrong compared to R600 in terms of modesettings?
2) Xonotic seems to have performance issues on some maps. Performance just unstable. Most of time it fits 60FPS (vsync acrive) but suddenly can start lagging. Not sure why, but on some maps in some cases performance can drop down to ~40FPS temporarily (which is a shame for R9 270 @ 1920x1200 @ normal settings, isn't it?). I use recent MESA from oibaf PPA and 3.15rc4 kernel so my setup seems to be quite similar. Any ideas what can screw up performance of R9 270 THAT much?
3) One time GPU has encountered lock-up when using UVD and GPU recovery failed, "as usually". Dear AMD, your "recent" GPU recovery code stinks! I remember some years ago driver was able to recover from GPUs lock-ups. But last couple years or so GPU reset code seems to be insane as it brings GPUs to the very same (faulty) state and usually process getting stuck or loops in cycle after facing the very same GPU fault after "recovery". Needless to say, such "recovery" fails to "recover" anything in most GPU lockups cases I've seen.

Other than that, RadeonSI seems to be quite good thing. And R9 270 on its own just amazing piece of hardware, being both powerful GPU and at same time really cool in idle modes (about 30C or so). Really decent work on power management!

[darkbasic]

1) Xonotic is unable to use 1920x1080 resolution - display shows "no signal" if I try to use this resolution. On other hand 1920x1200 works fine (native display resolution is 2560x1440). But wait, HD5770 can set up 1920x1080 on the very same display. WTF? I guess RadeonSI does something wrong compared to R600 in terms of modesettings?

Really? I have the very same problem (only with Xonotic): 1920x1080 works fine with one of my monitors but not with the others.

2) Xonotic seems to have performance issues on some maps. Performance just unstable. Most of time it fits 60FPS (vsync acrive) but suddenly can start lagging. Not sure why, but on some maps in some cases performance can drop down to ~40FPS temporarily (which is a shame for R9 270 @ 1920x1200 @ normal settings, isn't it?). I use recent MESA from oibaf PPA and 3.15rc4 kernel so my setup seems to be quite similar. Any ideas what can screw up performance of R9 270 THAT much?

I don't play xonotic but I didn't see any slowdown while benching (I did in the past). On the contrary the are lots of slow downs with Catalyst!

3) One time GPU has encountered lock-up when using UVD and GPU recovery failed, "as usually". Dear AMD, your "recent" GPU recovery code stinks! I remember some years ago driver was able to recover from GPUs lock-ups. But last couple years or so GPU reset code seems to be insane as it brings GPUs to the very same (faulty) state and usually process getting stuck or loops in cycle after facing the very same GPU fault after "recovery". Needless to say, such "recovery" fails to "recover" anything in most GPU lockups cases I've seen.

I completely agree, this is what happens if I max out Unvanquished and raise the resolution to 2560x1600: https://www.youtube.com/watch?v=VzQr...ature=youtu.be

I suggest you to patch your kernel with PTE (it really helps) and most of all to use Keith Packard’s xorg-server glamor-server branch (otherwise 2D acceleration is a pain in the ass).

[darkbasic]

Steam benchmarks!

http://www.linuxsystems.it/2014/05/radeonsi-faster-catalyst-steam-games/

[jockinator]

I also have a radeon HD7950, and with catalyst driver, most steam games have stuttering/lag issues.

since kernel 3.13, I was able to play L4d2, Killing Floor, FORCED, and Mark of the Ninja, with radeon SI.
I saw a big regression with kernel 3.14, and this games were not playable anymore.

looking at your results, I will try with 3.15

[Dukenukemx]

Steam benchmarks!

http://www.linuxsystems.it/2014/05/r...t-steam-games/

Those benchmarks use PTE. Any idea if oibaf drivers come with this patch?

[dungeon]

Those benchmarks use PTE. Any idea if oibaf drivers come with this patch?

They are today at version 5 of the "Large PTE" patchset... so better wait or try that .

[dungeon]

Forgot to add links

http://lists.freedesktop.org/archives/dri-devel/2014-May/059273.html

http://lists.freedesktop.org/archives/dri-devel/2014-May/059274.html

[0xBADCODE]

Really? I have the very same problem (only with Xonotic): 1920x1080 works fine with one of my monitors but not with the others.

Interesting. Kernel 3.15-rc5 seems to fix this issue. Now Xonotic can set 1920x1080 as well without troubles. Ok, one bug down .

I don't play xonotic but I didn't see any slowdown while benching (I did in the past). On the contrary the are lots of slow downs with Catalyst!

Interestingly, it seems to perform better without vsync, usually doing over 100FPS most of time @ 1920x1080 and no huge lags.

I completely agree, this is what happens if I max out Unvanquished and raise the resolution to 2560x1600: https://www.youtube.com/watch?v=VzQr...ature=youtu.be

in this case GPU recovery likely was able to re-initialise GPU at least to degree it outputs some picture. Most cases I encountered were complete failures to recover, when screen left blank and log indicates GPU recovery totally failed and driver either just gave up on fatal error (like IB allocation issues) or it looping without any progress and restores GPU to very same faulty state, facing the very same lockup issue. Which is rather silly.

[gururise]

Latest Oibaf PPA with RadeonSI on Ubuntu 14.04 results in "LLVM ERROR: ran out of registers during register allocation" when trying to run Torchlight II on wine. It used to work about a month or two ago, so I'm not sure what changed.