Announcement

Collapse
No announcement yet.

The Surprises When Testing A Radeon RX 470 With AMDGPU's DRM-Next Linux 4.11

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • L_A_G
    replied
    Originally posted by SomeoneElse View Post

    The Test 'The Surprises When Testing A Radeon RX 470 With AMDGPU's DRM-Next Linux 4.11' used a 4.11 Kernel with DRM-Next. The older test 'DeusExMD-January-2017-RS' used a 4.9 Kernel ... so there is a good possibility that not only the regression got fixed but also more improvements will be introduced. When comparing with the performance of DX:MD on windows, there was/is room for more fps under linux.
    The second paragraph of the post-update benchmark article explicitly states the tests are with kernel version 4.10 and Mesa 13.1-dev as of January 6th. I'm not doubting that they may have introduced some decent improvements since than and in the commits to kernel version 4.11, but reaching almost double the performance figures of an RX 480 in a month and a half looks a bit suspicious. Either it's a bug or then they've done some massive improvements they for some reason haven't told anyone about.

    Leave a comment:


  • SomeoneElse
    replied
    Originally posted by L_A_G View Post

    There may have been regressions, but if you look at the RX 480 benches after the initial performance patches to Deus Ex: Mankind Divided the RX 470 does way better. On low settings at 1920x1080 the 480 did 36.30 FPS when the 470 now does 58.10 and that's pretty much the same performance figures the R9 Fury did. These new tests used "high" rather than "ultra" settings, but even with high settings the 470 performs better than the 480 did on low settings (45.50 vs 36.30).

    Here's a link to Michael's tests just after the performance patch:

    https://www.phoronix.com/scan.php?pa...nuary-2017-RSI
    The Test 'The Surprises When Testing A Radeon RX 470 With AMDGPU's DRM-Next Linux 4.11' used a 4.11 Kernel with DRM-Next. The older test 'DeusExMD-January-2017-RS' used a 4.9 Kernel ... so there is a good possibility that not only the regression got fixed but also more improvements will be introduced. When comparing with the performance of DX:MD on windows, there was/is room for more fps under linux.

    Leave a comment:


  • Mystro256
    replied
    bridgman I had the exact same issue when I was looking at it last night. Seems fine this morning.

    Leave a comment:


  • master_andreas
    replied
    do those benefits translate to all GCN GPU or polaris only or 470 only?

    Leave a comment:


  • marek
    replied
    Memory management improvements might be responsible for this, but I'm not completely sure.

    Leave a comment:


  • L_A_G
    replied
    Originally posted by theriddick View Post
    Like many have said, the rx470 is a special case and had regressions previously. Other AMD cards will likely not see a huge improvement for Deus Ex.
    There may have been regressions, but if you look at the RX 480 benches after the initial performance patches to Deus Ex: Mankind Divided the RX 470 does way better. On low settings at 1920x1080 the 480 did 36.30 FPS when the 470 now does 58.10 and that's pretty much the same performance figures the R9 Fury did. These new tests used "high" rather than "ultra" settings, but even with high settings the 470 performs better than the 480 did on low settings (45.50 vs 36.30).

    Here's a link to Michael's tests just after the performance patch:

    https://www.phoronix.com/scan.php?pa...nuary-2017-RSI

    Leave a comment:


  • theriddick
    replied
    Like many have said, the rx470 is a special case and had regressions previously. Other AMD cards will likely not see a huge improvement for Deus Ex.

    Leave a comment:


  • L_A_G
    replied
    Whenever you see an improvement like that when there haven't been any major changes you should probably actually look at the images being produced. I get the feeling that there may be a bug somewhere and some part of the render pipeline is being skipped. Maybe anti-aliasing isn't working?

    I remember that a few years ago Nvidia got caught cheating in game benchmarks by having the drivers override the anti-aliasing settings to something less taxing and blamed a bug when people found out that if you changed the name of the .exe file the game would actually use the correct anti-aliasing settings and as a result run slower than if you didn't change the name of the .exe file.

    Leave a comment:


  • agaman
    replied
    Originally posted by rabcor View Post
    Impressive, all those improvements without any signs of regression, how rare.
    Compared to the performance of the RX480, seems to me that there was a regression for RX470 in linux 4.10 (and maybe in older kernels too). Now, with drm-next 4.11, it has the expected performance.

    Leave a comment:


  • ramrod
    replied
    The game works fine for me with the 4.8 kernel on my rx470, It must have regressed with 4.9 or 4.10.

    Leave a comment:

Working...
X