AMD Ontario / Bobcat / Llano and H.264

[highlandsun]

The new Ontario APU looks to be an amazingly good part with only 9W TDP. Is its GPU closely enough related to existing GPUs that support for it will arrive quickly in the open source radeon driver? Chips like this will obviously be great candidates for nettops and HTPCs, but only if we have decent codec acceleration available. Will that happen in the timeframe of these product releases? Are we going to still have to use Catalyst and XvBA?

[darkbasic]

Forget about gpu acceleration with open drivers.

[droidhacker]

Forget about gpu acceleration with open drivers.

That statement is absurd.
GPU acceleration WILL arrive. Not necessarily via UVD, which is still a legal question to be resolved by AMD, but DEFINITELY via shaders.... which will be more than sufficient for anything but ancient CPUs with GPUs that don't even have UVD.

[darkbasic]

It will arrive but NOBODY is working at it and it is NOT a priority.
Intel is working on h.264 gpu accelleration from years and there is still NO gpu acceleration for my gma45. If someone need gpu acceleration and if we want to be realistic the reality is there isn't anything and there will be nothing in an acceptably near future.
Having gpu acceleration in two or three years is useless for someone who plans to buy an hardware to effectively USE it.

[Schmaker]

Well thats the problem of AMD policy. They have two drivers for their cards, but none of them can use full potential of GPU, while nVidia has one closed source with 100% features included.
I guess this wont be fixed in acceptable time (<1 year), so Radeon users are just screwed. And that sucks, because their cards are doing very good job when you look on pricing, performance and effecienty.

[rohcQaH]

nVidia has one closed source with 100% features included.

Yeah, 100%. It offers out-of-the-box support, high-res boot screens and VT consoles, fast VT switching, proper XRandR, full kernel support, multiple X servers on a single card and usable backtraces. Oh, wait..

You cannot have most of these things in a closed driver, but you can't have awesome 3D performance and hardware video decode in an open one. 3d performance is a matter of resources (the majority of users doesn't use OSS drivers so nobody invests the needed amount of work). Video decode is a matter of not endangering contracts about copy protection with the movie industry.
AMD said they'd try to decouple video decode from copy protection hardware, but that'll only be done if it's cheap. Obviously it mustn't degrade performance, enlarge the die area or draw more power.
In other words: there's hope for future GPU generations, but nothing we could expect anytime soon.


IIRC there is work being done on a shader-based video acceleration for g3d, but it's an unfunded hobby project and I'm not sure about it's status.

[bridgman]

AFAIK the first step to decoding video on Gallium3D drivers is actually *having* Gallium3D drivers, and that is where the work is currently focused.

[droidhacker]

you can't have awesome 3D performance and hardware video decode in an open one.

I don't know if I'd use the word "can't".

3d performance is a matter of resources (the majority of users doesn't use OSS drivers so nobody invests the needed amount of work).

Where do you get this statistic?
Is across all GPU vendors? Or specific ones where there are options? I.e. including nvidia in the calculation will create a really ugly skew in the results since virtually ALL nvidia users use the blob. What if you focus strictly on AMD hardware? If more AMD GPU users use the blob driver, then I DOUBT that it is a particularly conclusive majority.

Video decode is a matter of not endangering contracts about copy protection with the movie industry.
AMD said they'd try to decouple video decode from copy protection hardware, but that'll only be done if it's cheap. Obviously it mustn't degrade performance, enlarge the die area or draw more power.
In other words: there's hope for future GPU generations, but nothing we could expect anytime soon.

Physical hardware changes may not be required to make UVD on Linux a reality. Just for them to think up ways to use UVD in a manner that doesn't give up ways to exploit BAD_DRM. Of course this is easier said than done and may never amount to anything.

It would be really nice if the various hardware makers could grow some brass ones and just tell hollywood to shove it. Its called boycotting. If all the major hardware vendors refused to support BAD_DRM, then the content providers would be forced to provide their content without DRM. After all, what use would it be to sell content that nobody can use?

Note that this line of thinking boils down to the fact that the majority of AMERICANS are really really STUPID and don't even respect their own rights.

[deanjo]

What if you focus strictly on AMD hardware? If more AMD GPU users use the blob driver, then I DOUBT that it is a particularly conclusive majority.

Well even those results are skewed now with AMD forcing people to utilize the open source drivers on older hardware with their killing off of blob support for R300-500 devices.

[deanjo]

high-res boot screens

Hmmm, I've been booting @ 1920x1200 for years with nvidia hardware. Adding vga=0x37d to the grub line works well enough.