| Bug ID | 109171 |
|---|---|
| Summary | AMD's GPU is Slower than Intel's |
| Product | Mesa |
| Version | 18.2 |
| Hardware | x86-64 (AMD64) |
| OS | Linux (All) |
| Status | NEW |
| Severity | critical |
| Priority | medium |
| Component | Drivers/Gallium/r600 |
| Assignee | dri-devel@lists.freedesktop.org |
| Reporter | ali.sherif10@yahoo.com |
| QA Contact | dri-devel@lists.freedesktop.org |
glxgears and CS: Go are much slower with AMD's GPU. Performanc= e's much higher with Intel's. Outputs: $ glxgears 306 frames in 5.0 seconds =3D 61.029 FPS 299 frames in 5.0 seconds =3D 59.761 FPS 299 frames in 5.0 seconds =3D 59.762 FPS 299 frames in 5.0 seconds =3D 59.762 FPS 299 frames in 5.0 seconds =3D 59.762 FPS $ DRI_PRIME=3D1 glxgears 304 frames in 5.0 seconds =3D 60.666 FPS 299 frames in 5.0 seconds =3D 59.762 FPS 299 frames in 5.0 seconds =3D 59.762 FPS 299 frames in 5.0 seconds =3D 59.763 FPS 299 frames in 5.0 seconds =3D 59.760 FPS $ vblank_mode=3D0 glxgears ATTENTION: default value of option vblank_mode overridden by environment. 32606 frames in 5.0 seconds =3D 6521.047 FPS 31516 frames in 5.0 seconds =3D 6302.010 FPS 32145 frames in 5.0 seconds =3D 6428.891 FPS 31443 frames in 5.0 seconds =3D 6288.595 FPS 31029 frames in 5.0 seconds =3D 6205.721 FPS $ DRI_PRIME=3D1 vblank_mode=3D0 glxgears ATTENTION: default value of option vblank_mode overridden by environment. 12031 frames in 5.0 seconds =3D 2406.195 FPS 10110 frames in 5.0 seconds =3D 2021.870 FPS 10122 frames in 5.0 seconds =3D 2022.474 FPS 9782 frames in 5.0 seconds =3D 1956.206 FPS 10175 frames in 5.0 seconds =3D 2034.901 FPS Laptop: HP Pavilion g6. AMD's GPU: Radeon HD 7670M. OS: 64-bit Linux (Fedora 29 KDE Spin).
CS: Go is faster with Intel's GPU even without vblank_mode=3D0= .
Please attach your xorg log and dmesg output. Note that using= the dGPU involves extra copies to copy the rendered frame to the integrated GPU for display so there is additional overhead. You need a relatively advanced 3D game to see an advantage on the dGPU as you have to overcome the overhead of the additional copies.
Created attachment 142914 [details]
Xorg log files.
I attached all files.
| What | Removed | Added |
|---|---|---|
| URL | https://bugzilla.redhat.com/show_bug.cgi?id=3D1662535 |
Created attac=
hment 142951 [details]
I used something like dmesg | vim.
The dmesg is only a fraction. Please provide the full dmesg di= rectly after boot. That glxgears is getting slower numbers on the secondary GPU is perfectly normal. It is a test how fast we can page flip and not how fast the GPU is.=
| What | Removed | Added |
|---|---|---|
| Attachment #142951 is obsolete | 1 |
Created attachment 142953 <=
a href=3D"attachment.cgi?id=3D142953&action=3Dedit" title=3D"Output of =
dmesg.">[details]
Output of dmesg.
Created attachment=
142954 [details]
Output of DRI_PRIME=3D1 dmesg
I restarted, logged into KDE then $ dmesg >> dmesg. Afte= r a minute, $ DRI_PRIME=3D1 dmesg >> dmesg2. Did I do it correctly? (In reply to Alex Deucher from comm= ent #2) > You need a relatively advanced 3D game to see an= advantage on the dGPU as you > have to overcome the overhead of the additional copies. Counter-Strike: Global Offensive is much slower, when I run Steam with DRI_PRIME=3D1. Additional Info: I installed Steam from RPMFusion.
Created attachment=
142955 [details]
Steam output wit Intel's GPU.
Created attachment 1=
42956 [details]
Steam output wit AMD's GPU.
I found that only CS: GO's main GUI is slow with AMD's GPU, bu= t it's much faster during matches. On Intel's GPU, the main GUI's smooth, but there's s= ome lag during matches. I tried the tutorial only. On both cases, fullscreen was enabled.
Created attachm=
ent 142957 [details]
glxinfo with(out) DRI_PRIME=3D1.
This can help.
| What | Removed | Added |
|---|---|---|
| Status | NEW | RESOLVED |
| Resolution | --- | MOVED |
-- GitLab Migration Automatic Message -- This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity. You can subscribe and participate further through the new bug through this = link to our GitLab instance: https://gitlab.freedesktop.org/mesa/mesa/issues/649.