Catalyst 9.2 Released, Still Fails To Deliver XvBA

[oyvind]

Funny drivers There is so much bugs and everybody waits and they only fix the minor things. There is only one i can say:

Choose nvidia, have no problems I did so, now it's you turn ...

One cannot always choose that easily.. I have nvidia on some machines, and I'm very pleased with that. The ATI-card is in my work laptop, and it's a Lenovo that I really like, and I chose it because it's generally nice Also, Lenovo's nvidia-offerings were few and far apart at the time (couple of years ago, I guess).

However, life is not care-free with nVidia binary driver. Look at nvnews-forum, and you'll see quite a lot of people having problems with nvidia as well. I have another laptop with a nVidia GeForce Go 6600 in it, and only just now, with the latest 180.29-driver, and after quite a few years(!), are things working without any significant problem (except for VCs getting garbled beyond recognition when X initializes , and xrandr shows no available resolutions except flat panel native). Luck, if you can call it that, most definitely plays a role with nvidia-hw+Linux as well.

[oyvind]

I'm seeing *drastic* improvements with video playback using XV. I don't know if it applies across the board to all cards, but for me, it's dramatically better than 9.1. No more distortions and Xine doesn't flake out anymore.

Wow. Color me almost satisifed.

Cool 4 u

My main problem with XVideo is the general drop in performance in the later Catalyst-releases, and especially when using Compiz. I don't expect it to be perfect in Compiz, because I realize the current limitations in DRI framework and X, but at least AMD should be able to do as well as the open source alternatives. Maybe my GPU is getting too old for AMD to bother prioritizing it.

[bommelid]

I installed the leaked Catalyst 9.2 beta on 15th of january. This was before the official catalyst 9.1 was released. The version number of the files inside are fglrx_*_8.580_*.
The files inside the new catalyst 9.2 final are version 8.582.
Seams to be not so much progress. Don't know, what version number 9.1 was but I wonder what they do in over a month.

regards

[sundown]

Catalyst 9.2 Released, Still Fails To Deliver XvBA

Catalyst is generally a fail

[bridgman]

My main problem with XVideo is the general drop in performance in the later Catalyst-releases, and especially when using Compiz. I don't expect it to be perfect in Compiz, because I realize the current limitations in DRI framework and X, but at least AMD should be able to do as well as the open source alternatives. Maybe my GPU is getting too old for AMD to bother prioritizing it.

Oyvind, how would you summarize the performance difference :

- cpu util is higher so maxed out there ?

- seems like GPU is maxed out; gets slower when the playback window is bigger ?

- seems like something else is slow; CPU not maxed and speed doesn't change when playback window is bigger/smaller ?

My guess is that the problem is more related to your GPU being relatively small than to its age. We are reaching the point where it's getting difficult to make a single driver which takes full advantage of new hardware and doesn't overload older hardware (without making it essentially two different drivers duct-taped together).

AFAIK your X1400 is what we call an RV515 with 4 pixel shaders. The new fglrx video stack probably does more complex video processing than the code in the open source driver (to get better video quality) and the result is that the GPU is maxing out.

I'm not sure if there are any options in aticonfig or cccle for disabling video quality options but it would be worth checking. It would also be useful for someone with a high-end 5xx (same age but more shader power) to see how the recent drivers perform on their GPUs. I suspect that their video performance would be fine -- in which case the simpler shaders on the open source driver might be a better fit for your GPU.

[Qaridarium]

overall it works for me...

infobash -v3
Host/Kernel/OS "ass" running Linux 2.6.28-6-generic i686 [ sidux 2008-04 Πόντος - kde-lite - (200812222321) ]
CPU Info (1) AMD Engineering Sample 1024 KB cache flags( sse3 nx lm svm ) clocked at [ 2600.000 MHz ]
(2) AMD Engineering Sample 1024 KB cache flags( sse3 nx lm svm ) clocked at [ 2600.000 MHz ]
(3) Dual-Core AMD Opteron 8218 1024 KB cache flags( sse3 nx lm svm ) clocked at [ 1000.000 MHz ]
(4) Dual-Core AMD Opteron 8218 1024 KB cache flags( sse3 nx lm svm ) clocked at [ 1000.000 MHz ]
Videocard ATI RV710 [Radeon HD 4350] X.Org 1.4.2 [ 1920x1200@60.0hz ]
Network cards 2x Broadcom NetXtreme BCM5754 Gigabit Ethernet PCI Express
Processes 149 | Uptime 2:54 | Memory 394.2/3276.9MB | HDD ATA SAMSUNG SP2014N,ATA SAMSUNG HD103UJ Size 1200GB (92%used) | GLX Renderer ATI Radeon HD 4350 | GLX Version 2.1.8494 Release | Client Shell | Infobash v3.05


wine 1.1.15 and cat 9-2 has nice looking @ the aquamark3 benchmark first timt for me that a see aquamark on wine looks like on windows XP i think its looks better.

3dmark03 is broken by a wine bug (not cat bug) on wine 1.1.15 for me so i can't test this..

sauerbraten runs fast :-)

for me and for the future 9-3 beta has XvBA lib's so i can't unterstand why AMD don't drop 9-2 (8.582) and jump direktly to 8.592 (9-3beta)...

Yes Yes I'm not the AMD release manager..

[falloutboy]

I guess there's still no tear-free Xv with the 780g igp. I switched the mainboard to one with an 8200 nvidia igp a few weeks back, when they fixed the 2D sloooooowness with the new driver. Now I've got accelerated HD video playback with zero cpu usage in contrast to not being able to watch HD content at all, because I had to use opengl for video playback before, which was using so much cpu-time even on SD content

[oyvind]

Oyvind, how would you summarize the performance difference :

- cpu util is higher so maxed out there ?

- seems like GPU is maxed out; gets slower as I drag the window bigger ?

- seems like something else is slow; CPU not maxed and speed doesn't change when playback window is bigger/smaller ?

My guess is that the problem is more to do with your GPU being relatively small more than its age. AFAIK your X1400 is what we call an RV515 with 4 pixel shaders. The new fglrx video stack probably does more complex video processing than the code in the open source driver (to get better video quality) and the result is that the GPU is maxing out sooner. The corresponding GPU in the 6xx family has roughly 2x the shader power, and the corresponding 7xx has more than 4x.

I'm not sure if there are any options in aticonfig or cccle for disabling video quality options but it would be worth checking.

Thanks for the reply, here are some observations:
* Starting with Catalyst 8.12, I think something changed with XVideo (might be 8.11, cannot remember exactly). I suddenly noticed that when running my GPU with a lower powerstate, XVideo suffered greatly. I could not remember seeing that before. At the same release, power-state switching turned unstable, so I've stopped using it completely.

* Ever since Catalyst 8.12, XVideo has not been flickering in Compiz-environment, but performance is abysmal, especially when going fullscreen. It does *not* peg CPU, and something very weird is going on, so probably hits some kind of GPU-limitations. 2-3 FPS (!) and extreme jerking/tearing with mplayer+XV in fullscreen, a little better with VLC, but still far below 25 FPS (or whatever the FPS of the movie being played back). CPU usage is normal/low while video is jerking and stopping/starting like crazy. Sometimes X crashes (black screen + Xorg 100% CPU) when I run XV in fullscreen in Compiz.

XVideo in non-Compiz environment works much better, and doesn't seem more CPU demanding than before. But probably more GPU demanding (harder to tell really, other than the powerstate-hint).

* Playing back video with OpenGL is suddenly much better than it used to be with fglrx, and mplayer now plays normal movies (not super-duper HD) tear-free in fullscreen with OpenGL (Compiz+unredirect) smoothly. That is with the very same test-videos I use with XV (a 1024x576 25 FPS movie, and a 640x480 60 FPS movie).

* XVideo performance is window size dependent.

* The open source radeon driver is able to play back using XVideo in Compiz without a hitch, and has been so for a while (albeit, you have to enable EXA to get it working). I can see no quality difference between open source driver and fglrx' XV output.

* Even though my GPU is limited (I am aware of that and not expecting magic), something still seems odd/wrong with XVideo. As long as the open source driver can do it on the same hw, I'll continue to complain about it .

[oyvind]

<snip>
I'm not sure if there are any options in aticonfig or cccle for disabling video quality options but it would be worth checking. It would also be useful for someone with a high-end 5xx (same age but more shader power) to see how the recent drivers perform on their GPUs. I suspect that their video performance would be fine -- in which case the simpler shaders on the open source driver might be a better fit for your GPU.

Oh, and please do tell if you know of any options I can try to lower quality of XVideo to gain some performance back (if that turns out to be the issue). I currently don't use any special options in xorg.conf at all. Basically, the only things I've changed is "aticonfig --auto-powerstates=off" and I've lowered mipmapping-quality in CCC. I start with clean amdpcsdb every time I install a new Catalyst-release ...

[bridgman]

* Starting with Catalyst 8.12, I think something changed with XVideo (might be 8.11, cannot remember exactly). I suddenly noticed that when running my GPU with a lower powerstate, XVideo suffered greatly. I could not remember seeing that before. At the same release, power-state switching turned unstable, so I've stopped using it completely.

Yeah, we dropped in the first release of an all new video stack around that time.

* Ever since Catalyst 8.12, XVideo has not been flickering in Compiz-environment, but performance is abysmal, especially when going fullscreen. It does *not* peg CPU, and something very weird is going on, so probably hits some kind of GPU-limitations. 2-3 FPS (!) and extreme jerking/tearing with mplayer+XV in fullscreen, a little better with VLC, but still far below 25 FPS (or whatever the FPS of the movie being played back). CPU usage is normal/low while video is jerking and stopping/starting like crazy. Sometimes X crashes (black screen + Xorg 100% CPU) when I run XV in fullscreen in Compiz.

OK, that makes sense. Here's an interesting question -- did you ever play videos for any length of time under Compiz before ? The obvious reason for asking is that playing video under Compiz involves at least 2x the pixel pushing of video without Compiz. I'm obviously trying to figure out if the slow performance under Compiz is new, or if it was always slow but you never noticed because the flickering chased you away first

XVideo in non-Compiz environment works much better, and doesn't seem more CPU demanding than before. But probably more GPU demanding (harder to tell really, other than the powerstate-hint).

Yep.

* Playing back video with OpenGL is suddenly much better than it used to be with fglrx, and mplayer now plays normal movies (not super-duper HD) tear-free in fullscreen with OpenGL (Compiz+unredirect) smoothly. That is with the very same test-videos I use with XV (a 1024x576 25 FPS movie, and a 640x480 60 FPS movie).

Oh good

* XVideo performance is window size dependent.

OK, so that also points to a maxed-out GPU.

* The open source radeon driver is able to play back using XVideo in Compiz without a hitch, and has been so for a while (albeit, you have to enable EXA to get it working). I can see no quality difference between open source driver and fglrx' XV output.

Makes sense. I suspect that you're seeing the difference between hard-coded bicubic filtering with a helper texture on the open source drivers, vs. a more expensive but more flexible adaptive filter on fglrx. Not sure of that though... if it was the case I would expect you to see a quality difference on moving objects more than on fixed objects, FWIW.

* Even though my GPU is limited (I am aware of that and not expecting magic), something still seems odd/wrong with XVideo. As long as the open source driver can do it on the same hw, I'll continue to complain about it .

I would expect no less. Thanks