Arrivederci Radeon

[amphigory]

After a couple of years waiting for accelerated video decode and rendering on GPU I've finally run out of patience. My R600 based card did a decent job with non-HD source material but 1080i deinterlace and rendering is less than optimal.

I've ordered up a NVIDIA GT520 card and will be installing that in the next couple of days. I hate this. The Gallium driver architecture is the first sane approach to video on Linux I've seen. Unfortunately, the performance is sucky for HTPC applications. Bleh. Back to the blob.

Has anyone heard of any pending work for VDPAU/pipe-video? I've been building Mesa from git for the last couple of years and have not seen any new commits for video in quite some time. Also, is there any news on AMD opening up UVD2?

[ChrisXY]

video acceleration in gallium is working on r600g but currently limited to mpeg1 and mpeg2 I think.

There is work being done for webm support and I am sure somewhere somewhere is trying to implement it for h.264, but that is not ready yet.

I am curious: What performance exactly do you need? Only video playing?

Recently I was surprised that portal 2 in wine was running very well on my HD 6550 mobile card on mesa git. Needs only minor performance tuning now to be perfect so it is indeed quickly progressing...

[pingufunkybeat]

In terms of 3d functionality, the drivers are quite good -- they can run most stuff, including Unigine benchmarks, and GL 3 should be here by the end of the year.

But it's been a while since I noticed major commits in mesa. The pace has slowed down considerably.

[amphigory]

I'm fairly satisfied with Gallium OpenGL performance as I don't use Linux for games. Standard definition video with H.264 is fine. The big issue right now is with deinterlace for MPEG2 1080i. I'm using XBMC + MythTV backend + Ceton InfiniTV tuner (Ceton tuner is AWESOME by the way).

All is fine with 720 content but 1080i sucks. Since there appears to be no provision for GPU deinterlace processing on Radeon it is done on CPU. This would be fine if XBMC was using ffmpeg-mt but all decode and rendering seems to be done on one core. Any filter aside from 'Weave' takes core 0 utilization to 100% (the other three are idle) and video gets a bit too choppy for my taste.

I can get OK results with Weave deinterlace on 1080i content but that borks other content. Having to switch deinterlacers manually depending on content is not really going to be an option for my wife. She already uses intemperate language regarding our HTPC. :-)

I've spent quite a bit of time trying to optimized things to no avail. I suppose I could try building XBMC from source and link to ffmpeg-mt but I've already burned enough time jacking with it. If $50 buys me good video with NVIDIA GT520 then it is money well spent. I'm just kinda sad that I'll be back to using binary blob.

[amphigory]

As a side note regarding Ceton tuner -- I had to build MythTV from source with patches from Ron Frazier. The CableCard provisioning process with Comcast was an unholy mess (took nine calls over two days). However, despite the work required to get it up and running, I can say that it is the best tuner card I have ever used. It will record four simultaneous MPEG2 streams -- and after recording over 100 shows I have yet to see a single recording failure. Freaking awesome gear.

Caveat: Comcast Seattle sets all 'non-premium' content to CCI 0x00. I am lucky. Some providers put restricted 5C/CCI flags on standard channels.
For more on 5C/CCI:
http://en.wikipedia.org/wiki/Copy_Control_Information

http://www.cetoncorp.com/index.php

[amphigory]

NVIDIA GT520 arrived and installed in HTPC. CPU utilization rendering 1080i (w/deinterlace) down from 100% to 6%!

[duby229]

Is this really a fair comparison though? Open source ATi drivers vs proprietary nVidia drivers...

I understand that ATi's proprietary driver sucks ass though, so if your going to use a proprietary driver then there is no better choice than nVidia. Still a better comparison would then be nouveau vs r600g.

I'm not necessarily complaining per se, but r600g is soooo close now the feature you need is literally only right around the corner... What I mean is, we've all been waiting for so long already that the time left until it gets completed is almost like a blink of an eye. I think by giving in at this point you've given legitimacy to proprietary ideology.

[amphigory]

Is this really a fair comparison though? Open source ATi drivers vs proprietary nVidia drivers...

I understand that ATi's proprietary driver sucks ass though, so if your going to use a proprietary driver then there is no better choice than nVidia. Still a better comparison would then be nouveau vs r600g.

I'm not necessarily complaining per se, but r600g is soooo close now the feature you need is literally only right around the corner... What I mean is, we've all been waiting for so long already that the time left until it gets completed is almost like a blink of an eye. I think by giving in at this point you've given legitimacy to proprietary ideology.

Oh... I agree that it's not fair at all. However, that is the state of affairs. Gallium drivers at this stage of the game cannot compete with binary blobs. I'd guess that another year will pass before we will see advanced VDPAU capabilities in Gallium based on the number of commits I've seen in Mesa master.

I won't wait that long. As much as I'd like to use the open source drivers they just do not do what I need. I'll be happy to try reinstalling my R600 card when drivers have the required capabilities. I look forward to that day.

[Kjella]

Is this really a fair comparison though? Open source ATi drivers vs proprietary nVidia drivers...

Of course not. But ideology can only go so far in making up for deficiencies in the result.

I'm not necessarily complaining per se, but r600g is soooo close now the feature you need is literally only right around the corner... What I mean is, we've all been waiting for so long already that the time left until it gets completed is almost like a blink of an eye.

How long have you had this feeling? AMD first came up with their open source strategy in 2007. My card (Evergreen) was released in 2009. In 2011 there's still no accelerated video or HDMI audio. According to AMD it's highly unlikely that UVD will ever be released and the shader implementation seems to have stalled. The 3D engine is better now than a year ago but in other areas I feel it's completely standing still. Not that this is a complaint, but thinking the pace will be wildly different in the future seems unlikely.

I think by giving in at this point you've given legitimacy to proprietary ideology.

Ideology is to care more about how the product is made than what it actually delivers. Not caring whether it's open or closed source is not an ideology.