I pulled agd5f's drm a few days ago (9/28). I can't remember if that is the latest update or not (drm-2.3.0-2634-gb323796). The last commit:
Last edited by DanL; 10-01-2009 at 12:00 PM.
after i rebuilt drm and mesa again, it finaly works fine again, openarena has no glitches, indirect rendering only for quake3 still, but it has no glitches too, i have drawprim disabled with patch posted here earlier.
Is anyone playing things using the free drivers on r6xx+ ?
What i'm observing on my HD4550 is frame rates between 40 and 80 in OpenArena, with very little relation to detail/complexity level and resolution.
I've tried many different combinations, and basically I get similar frame rates using vertex lighting and minimum detail at 800x600 and full-blown everything on at 1920x1080. There is some difference, but I can't hit 125fps, whatever I do.
What does this mean? It's hard to imagine that the card is not powerful enough (at low res), the engine is 10 years old. I guess that the fill rate is not the issue, as the resolution doesn't matter much. Where is the bottleneck? Is there hope for improvement?
Another thing I've noticed is that the frame rate is very unstable and fluctuates wildly, even when you're staring at a wall. On top of this, there are sticky points every 2 seconds or so (screen freezes for a split second), which gets worse the longer X has been running. Upgrading to the latest drm-next kernel has improved this significantly, but it's still there.
This is not moaning, just a discussion, and I do appreciate the efforts that have gone into the drivers. We've had a few trolls here recently, so I wanted to make that clear.
If the frame rate doesn't change much with resolution that means you're either CPU bound or limited by the driver's ability to overlap GPU and CPU processing.
Are you running with KMS ? If so the new memory manager code is probably introducing some frame rate variations, but that will improve over time.
I wonder about:I get this with glxgears. But maybe 60fps is for all OpenGL apps?Code:*** NOTE: Don't use glxgears as a benchmark. OpenGL implementations are not optimized for frame rates >> 60fps, thus these numbers are meaningless when compared between vendors.
Yeah, the drivers are written for real world apps that do a lot of drawing in each frame, ie the drawing code gets more attention than the "once per frame" code.
The glxgears program draws so little each frame that you end up measuring the performance of the "once per frame" code rather than the drawing code -- so the correlation between glxgears performance and real world application performance is weak at best.
That's why the devs make rude comments about glxgears being used as a benchmark for anything other than buffer-flipping. Any frame rate faster than your display refresh rate is largely wasted anyways.
Last edited by bridgman; 11-20-2009 at 11:28 AM.