You are not logged in.
After reading through a subreddit thread about removing Intel drivers to allow DRI3 support for a potential performance gain, I attempted this by removing the x86-video-intel package, and rebooting. After doing this, I checked all my Xorg logs (both /var/log/Xorg.* and .local/share/xorg/Xorg.* for messages containing DRI3 and glamor, of which there are none. In my Xorg.1.log, which is handled by GDM, I see the following output:
[ 17265.711] (II) Loading /usr/lib/xorg/modules/drivers/intel_drv.so
[ 17265.711] (II) Module intel: vendor="X.Org Foundation"
[ 17265.711] compiled for 1.18.1, module version = 2.99.917
[ 17265.711] Module class: X.Org Video Driver
[ 17265.711] ABI class: X.Org Video Driver, version 20.0
This indicates the Intel driver is still being loaded, which shouldn't occur since the package was removed. After much investigating to determine why this is installed, I came to an unusual conclusion: the driver doesn't exist. No file is actually present at /usr/lib/xorg/modules/drivers/intel_drv.so, which leaves me baffled.
Offline
You sure that's not an old log? I only ask because the timestamp [ 17265.711 ] indicates themessages were generated ~5 hours after boot.
But whether the Constitution really be one thing, or another, this much is certain - that it has either authorized such a government as we have had, or has been powerless to prevent it. In either case, it is unfit to exist.
-Lysander Spooner
Offline
Aha, it seems you are correct. The log I was looking at was old, and the newest one was specifically in .local/share/xorg/Xorg.1.log, and this one shows the intel driver not being loaded.
Now I'll have to figure out why it's not using DRI3 though. Thanks!
Offline
Modesetting should use DRI3 by default.
Run LIBGL_DEBUG=verbose glxinfo | grep libgl to verify if DRI3 is active.
Disliking systemd intensely, but not satisfied with alternatives so focusing on taming systemd.
(A works at time B) && (time C > time B ) ≠ (A works at time C)
Offline
Yes it is indeed active. I was mistaken in believing that it would show up in logs.
Offline
Did you notice a performance gain? I am using xf86-video-intel drivers myself and I am noticing a certain sluggishness when windows open and are moved. Curious to see if this fixes it.
Offline
I personally haven't noticed much difference. I feel like the window opening and workspace switching is a bit smoother, but it's difficult for me to objectively say that it's significantly better. It definitely isn't worse though. That being said, my laptop has hybrid nvidia graphics, so while I'm running a more demanding application, I don't suffer from any performance loss to the desktop environment.
Offline
I think you can use a tool like glmark2 (AUR) to compare the two drivers.l
EDIT: I did this on my machine (repeating now to get some replicates to verify) and running with xf86-video-intel is a little faster on my i7-3970k. I will update the post with the data once I get n=3 with and without the driver.
Last edited by graysky (2016-05-17 22:28:31)
CPU-optimized Linux-ck packages @ Repo-ck • AUR packages • Zsh and other configs
Offline
For my i7-4790k, I ran the benchmark 3 times per driver (either with xf86-video-intel or without it). In general, th driver provided a boost in performance as measured by the benchmark. YMMV.
With the driver: average glmark2 score=2,842.67 +/- 27.6
Without the driver: average glmark2 score=2,590.33 +/- 5.7
And the individual benchmarks:
Last edited by graysky (2016-05-18 14:52:48)
CPU-optimized Linux-ck packages @ Repo-ck • AUR packages • Zsh and other configs
Offline
What did you use to generate that graph? I'm planning on posting results once mesa 11.3 reaches stable release.
Offline
It's commercial software my university has called spotfire dxp.
CPU-optimized Linux-ck packages @ Repo-ck • AUR packages • Zsh and other configs
Offline
After doing some digging, I don't think the glmark2 benchmark accurately reflects the purpose of changing drivers. glmark2 only tests a set of OpenGL ES features, whereas changing the driver is meant to take advantage of DRI3 features, which (as far as I can tell) use more than what that benchmark covers. I'm not saying that the intel drivers aren't better than the modesetting ones, as you've demonstrated in certain applications it is indeed preferable. I think a more suitable benchmark is needed, though I'm have been unable to find one that covers this particular use-case.
Offline
So, does system without xf86-video-intel uses intel driver (i915) or what? I do not understand this, but I just removed xf86-video-intel and everything works...Only sddm loads a bit slower...
I do not speak English, but I understand...
Offline
pyntux, linux graphics stack uses multiple components.
Here's a simplified description :
1. kernel module
2 .drm driver
3. an OpenGL , OpenCL, OpenGL ES etc provider - ( mesa for open sourcedrivers)
4 graphic videodriver ( x11 , wayland , mir )
modesetting and xf86-video-intel both run at level 4 and share everything below that.
There are some big differences between modesetting and xf86-video-intel :
modesetting uses glamor for & dri 3 for HW acceleration
xf86-video-intel uses dri2 and SNA or UXA for HW acceleration.
Disliking systemd intensely, but not satisfied with alternatives so focusing on taming systemd.
(A works at time B) && (time C > time B ) ≠ (A works at time C)
Offline
pyntux, linux graphics stack uses multiple components.
Here's a simplified description :
1. kernel module
2 .drm driver
3. an OpenGL , OpenCL, OpenGL ES etc provider - ( mesa for open sourcedrivers)
4 graphic videodriver ( x11 , wayland , mir )modesetting and xf86-video-intel both run at level 4 and share everything below that.
There are some big differences between modesetting and xf86-video-intel :
modesetting uses glamor for & dri 3 for HW acceleration
xf86-video-intel uses dri2 and SNA or UXA for HW acceleration.
Nice explanation... Thank you very much...
So I am using modesetting now...
I do not speak English, but I understand...
Offline