You are not logged in.

#1 2009-10-15 11:28:31

fijam
Member
Registered: 2009-02-03
Posts: 244
Website

[solved] Serious performance regression with xorg stack on intel 915GM

I jumped on to testing today (xorg-server 1.7.0.901, mesa 7.6 and xf86-video-intel 2.9) and sadly, with i915GM, I am experiencing a serious performance regression in comparison to stable (xorg-server 1.6.3, mesa 7.5 and xf86-video-intel 2.8.1). Previously smooth World of Goo in fullscreen resolution is now choppy to the point of no longer being playable. Sadly, it does not have a fps counter.

Glxgears dropped by 25% (just for reference).

I'll install openarena or something similar later and give some real fps counts.

in dmesg:

i915 0000:00:02.0: setting latency timer to 64
[drm] DAC-6: set mode 640x480 0
render error detected, EIR: 0x00000010
page table error
  PGTBL_ER: 0x00000100
[drm:i915_handle_error] *ERROR* EIR stuck: 0x00000010, masking
render error detected, EIR: 0x00000010
page table error
  PGTBL_ER: 0x00000100

Xorg.0.log clean.

edit: AIGLX is enabled, direct rendering is enabled.

edit 2: solved

Last edited by fijam (2009-10-24 06:57:52)

Offline

#2 2009-10-15 13:11:54

skualito
Member
Registered: 2008-11-19
Posts: 203

Re: [solved] Serious performance regression with xorg stack on intel 915GM

I also upgraded today Xorg to testing, and can confirm the serious performance regression. I also have an 915GM. But the error mentionned in dmesg was there already before the upgrade. It started with update from kernel26-2.6.30 to kernel26-2.6.31 series.

render error detected, EIR: 0x00000010
page table error
  PGTBL_ER: 0x00000100
[drm:i915_handle_error] *ERROR* EIR stuck: 0x00000010, masking
render error detected, EIR: 0x00000010
page table erro

Offline

#3 2009-10-15 13:54:09

fijam
Member
Registered: 2009-02-03
Posts: 244
Website

Re: [solved] Serious performance regression with xorg stack on intel 915GM

skualito wrote:

But the error mentionned in dmesg was there already before the upgrade. It started with update from kernel26-2.6.30 to kernel26-2.6.31 series.

You are right, thanks for pointing that out. That narrows the regression down to xorg stuff. With my totally lousy internet connection and more important things to do at the moment I'll probably download openarena from Daenyth's arch-games repo overnight and do any serious testing tomorrow/Saturday.

Last edited by fijam (2009-10-15 14:02:31)

Offline

#4 2009-10-15 14:58:24

goran'agar
Member
From: Nothern Italy
Registered: 2009-05-19
Posts: 171

Re: [solved] Serious performance regression with xorg stack on intel 915GM

I can confirm it on an eeePC 900 (also i915). I use stellarium as a benchmark and the performance dropped from 18fps to 3 fps (!!), also worsening the video quality.

I use kernel 2.6.32-rc4.

It was a PITA rolling back to the previous versions but after doing that all is fine again.


Sony Vaio VPCM13M1E  - Arch Linux - LXDE

Offline

#5 2009-10-15 15:25:02

TigTex
Member
From: Portugal
Registered: 2008-06-19
Posts: 301

Re: [solved] Serious performance regression with xorg stack on intel 915GM

I can also confirm this but on GMA950 acer aspire one. Glxgears dropped to near 200fps and games are unplayable (from 50fps to 2fps).


.::. TigTex @ Portugal .::.

Offline

#6 2009-10-15 15:45:01

GogglesGuy
Member
From: Rocket City
Registered: 2005-03-29
Posts: 610
Website

Re: [solved] Serious performance regression with xorg stack on intel 915GM

965GM here which seems to be working fine. GLXGears around 400 FPS, which is a lot lower than I used to get (around a 1000), but those numbers were from a year ago or so, so I'm not sure if you can compare those.

Last edited by GogglesGuy (2009-10-15 15:50:12)

Offline

#7 2009-10-15 15:51:03

fijam
Member
Registered: 2009-02-03
Posts: 244
Website

Re: [solved] Serious performance regression with xorg stack on intel 915GM

GogglesGuy wrote:

965GM here which seems to be working fine. GLXGears around 400 FPS.

glxgears has only some relevance in comparison to previous results. My glxgears also oscillates about 400fps now yet it does not really correlate with real performance. -> http://qa-rockstar.livejournal.com/7869.html

edit: You edited your message while I was writing mine smile Yeah, basically it would be best to compare testing to non-testing in some kind of standardized real-workload way.

Last edited by fijam (2009-10-15 15:53:49)

Offline

#8 2009-10-23 16:59:08

fijam
Member
Registered: 2009-02-03
Posts: 244
Website

Re: [solved] Serious performance regression with xorg stack on intel 915GM

This is solved for me by the recent mesa and intel-dri update build without gallium. Performance is back to normal or even a bit better.

Offline

Board footer

Powered by FluxBB