AMD Radeon VDPAU Video Performance With Gallium3D

[89c51]

BTW, did you people read my first comment in this thread? VDPAU in radeon doesn't really help that much... until we have complete AVC decoding by HW, and even then, GStreamer doesn't work very well in firefox (it crashes the browser in my system, it's a known bug IIRC), and youtube uses VP8/9.

Youtube uses h246 for most of its videos. I think it falls back to webm if your browser doesn't support h246. And gstreamer with HWA crashes firefox in all systems.


Also there is one more thing with vdpau. If you play a video with gst-play and try to run a vlc video with vdpau enabled it will not play it. +vlc uses more cpu than gst-play.

[dungeon]

What i tried to say is... With currently tested by phoronix-test-suite 30fps video variant, E1-2500 will pass that even with Xv for sure, maybe even E1-2100 (but not exactly sure for that one, because of high peaks Xv can have) . So that is not so good showcase for slowest Kabinis .

With 60fps video using Xv it will need more CPU power then even E1-2500 have, so for sure E1-2100 can't do it too, so both will struggle at that... But using vdpau it work fine and CPU usage is near to non existent .

[dungeon]

I asked as it's the slowest setup with the supported UVD chip thats on the market. How it handles it is a a bit more telling about what it's capable of then a low end GPU with a current gen CPU.

Yup i also tested on second slowest E1-2500 APU equiped laptop few days ago (that is something like 2x1.45GHz)... that worked fine too, less then 10% CPU usage but for the 1080p@60 fps BigBunny variant, that one will use twice CPU power with Xv then Michel tested here (mov/qt variant).

http://distribution.bbb3d.renderfarm...fps_normal.mp4

That is the one, for example will use ~twice CPU power if Xv is used, but for vdpau it is the same , etc... I don't know why Michael does not use that for benchmark it is better quality and better shows an issue with software rendering .

[Kivada]

According to my results with Athlon 5350 i guess that should work fine:

http://openbenchmarking.org/result/1406296-KH-ATH53505338

So Xv is around 10% average and vdpau 2% .

Maybe i can test by disabling 2 cores and downclocking it @1GHz .

I asked as it's the slowest setup with the supported UVD chip thats on the market. How it handles it is a a bit more telling about what it's capable of then a low end GPU with a current gen CPU.

[Kivada]

That defeats the point of passive cooling (i.e. silence). I actually do something similar with a 120 mm fan pointed at a half-height 8400GS, but it's running on 5V and very, very quiet.

Have you seen the size of the heatsinks they put on bottom end GPUs with a built in fan? Their surface area is almost non existant and the fans are always built right into them. This is why I say get the passive version and add in a fan, if the fan dies it'll still be fine passively, but with the fan it'll run much cooler and give you some headroom for overclocking should the GPGPU or gaming itch bite you.

[dungeon]

BTW, did you people read my first comment in this thread? VDPAU in radeon doesn't really help that much... until we have complete AVC decoding by HW, and even then, GStreamer doesn't work very well in firefox (it crashes the browser in my system, it's a known bug IIRC), and youtube uses VP8/9.

I guess most people realise that software decoding is unbeatable , decoding units always lack to supporting some codecs/profiles, etc... but of course accelerating most used ones is good to have .

[asdfblah]

BTW, did you people read my first comment in this thread? VDPAU in radeon doesn't really help that much... until we have complete AVC decoding by HW, and even then, GStreamer doesn't work very well in firefox (it crashes the browser in my system, it's a known bug IIRC), and youtube uses VP8/9.

[bridgman]

And sadly every time you show up to comment on that topic you fail to actually answer the question: Has there been any progress made on UVD for RS780/880 in the last, lets say 9 months? If not, would you still call this low priority ? Or would you call it what it is: Nobody is working on that anymore?

I have answered multiple times, and the answer is yes.

[dungeon]

According to my results with Athlon 5350 i guess that should work fine:

http://openbenchmarking.org/result/1406296-KH-ATH53505338

So Xv is around 10% average and vdpau 2% .

Maybe i can test by disabling 2 cores and downclocking it @1GHz .

Kernel 3.16-rc3 but governor performance and results go down - average 1.26% CPU used .

http://openbenchmarking.org/result/1406296-KH-ATH53576288

That ondemand is somewhat buggy, make high peaks in chart for both xv and vdpau cases .

@Vim_User

Can you do some benchmarks like this (or just tell what CPU is in that laptop) to show how much lack of vdpau is issue for you .

[Vim_User]

But you know that and less than 6 months ago they gave you an answer, to an question in this forum. So they told then the look into it, but even if implemented legal review could forbid them to release it. Still no way to complain about oss driver.

http://www.phoronix.com/forums/showthread.php?94817-support-for-first-generation-UVD-blocks-%28RV6xx-RS780-RS880-and-RV790%29/page3

But in one point you are right, the catalyst legacy hw driver should get at least LTS linux distro release support, if oss support is not in place.

In the thread you linked to bridgman clearly says that a review process has to happen and that this review process can lead to:
- more coding work on the driver
- a release of UVD support
- the bitter end: no UVD for that hardware

Since all this happens AMD internally the only chance we have to get information about this is to actually ask about progress on this, what I have done. And since bridgman constantly fails to tell us if there is progress or not I have to assume that in reality no one works on that, neither legal department nor developers.