Future AMD GPUs To Be More Open-Source Friendly?

[phoronix]

Phoronix: Future AMD GPUs To Be More Open-Source Friendly

AMD is on the heels of releasing the next set of GPU programming documentation to aide in the development of the open-source R500/600 drivers (xf86-video-ati and xf86-video-radeonhd). It's already been discussed what this NDA-free documentation release will have, but one of the questions that have repeatedly come up is if/when AMD will release information on accelerated video playback. AMD's John Bridgman has now stated what they plan to release in the video realm as well as a new requirement for their future graphics processors: being open-source friendly while avoiding DRM.

http://www.phoronix.com/vr.php?view=11631

[puntarenas]

So what does this mean for RV670, will there be at least IDCT (Inverse Discrete Cosine Transform) and MC (Motion Compensation) or no hardware-assisted video decoding at all?

[Michael]

So what does this mean for RV670, will there be at least IDCT (Inverse Discrete Cosine Transform) and MC (Motion Compensation) or no hardware-assisted video decoding at all?

RV670 has the UVD, but according to what has been said, the IDCT block is outside of the UVD, so that portion can be opened up.

[Michael]

Update: FYI, it will be about a year before it's concrete on the modular UVD possibility now.

[Mathieu Bonnet]

Hi,

If there are problems open sourcing UVD on RV670 chips, is there any chance on having it supported in the proprietary drivers? (or maybe, as a proprietary module, for the open source drivers? -I wouldn't mind, personally, if there is no other solution).

And what about encoding and transcoding? (in the proprietary drivers, or in the open source ones).

On http://www.sapphiretech.com/us/products/products_overview.php?gpid=198&grp=3, they say "MPEG-2, MPEG-4, DivX, WMV9, VC-1, and H.264/AVC encoding and transcoding".


Thanks in advance for any information, if any are available,

Bye.

[yoshi314]

good to know that they don't only care about the drivers. if they're going to make changes on hardware level to make it opensource-friendly - now that's great news.

While there is no easy way around this for the R600 series, John Bridgman has stated "[a modular version of the UVD is] already on the requirements list for future chipsso i guess that means i'll skip r600 altogether then, when i'll be considering an hardware upgrade. assuming they really will make things easier on next line of chips.

anyway, let's wait and see what happens. it's great that ati folks are still planning to add extra stuff to the specs. i wasn't expecting that.

[Michael]

If there are problems open sourcing UVD on RV670 chips, is there any chance on having it supported in the proprietary drivers?

I think legally they could enable the UVD support if it was in their binary blob, but whether they'll actually do it could be a different story right now.

[d2kx]

I bought my AMD Radeon X1900 XT 512 (R580) the day it was avaible and it's still a great card. I will skip the R600 series and the first R700 parts, but am looking forward to the R780 (R700 refresh) in the end of 2008. But R600 users will not like this new.

[sreyan]

I think that it's great that amd will be releasing the IDCT and motion comp to developers in addition to 3d.

I think it's also quite fair for them to not release anything that might interfere with their legal obligations, especially in the light of their willingness to improve their design. This as a whole helps to level the playing field against intel who has fully open source drivers for their graphics cards, but scrubs out certain portions like macrovision.

I think it's important to point out that this makes fusion processors much more viable for linux use, especially if amd is to compete against intel's menlow and moorestown platforms in the umpc/mid space.

[an0n1m0us]

I'd just like to put my two cents in. I've got an R500 series card and I purchased it for nothing but video decoding. I couldn't give a toss about gaming and hence 3D acceleration.

If there's any chance video acceleration can be prioritised over gaming acceleration, I'm all for it!

[Kano]

Yes, currently even the cheapest NV or onboard Intel gfx card is better using XV. That's why I usally use my box with NV card for videos.

[bridgman]

Is it fair to say that the rendering/scaling part of video acceleration (done by xv or through opengl) is the first priority, not the decoding part (done by xvmc) ?

I'm asking because (a) my understanding is that a modern CPU can do a pretty good job of the decoding part these days, at least for SD, and (b) we use shaders for some of the rendinging work on 5xx and above parts, implying that we need basic 3d functionality in place first so that the shaders can be used for scaling etc...

What I expect is that we will provide info and support to get basic 3d running, then from that point two things will happen in parallel -- using the 3d engine for some video acceleration work, and continuing to evolve the 3d acceleration.

[Dandel]

I'd just like to put my two cents in. I've got an R500 series card and I purchased it for nothing but video decoding. I couldn't give a toss about gaming and hence 3D acceleration.

If there's any chance video acceleration can be prioritised over gaming acceleration, I'm all for it!

I can say that the closed source driver has video acceleration if you enable Xvideo support in the config file. just using the r500 to render will reduce cpu usage on linux by a ton with this only.

[joho]

I just bought an 2600Pro because of the open documentation announcement. I am in the process of building a PVR system using MythTV. And now I read that the one part that I plan to use, video decoding, will not be documented!
My first thought was that they must be joking. But I realize that this is not the case.
What a ($!#@@$@, now I must find a way to return my AMD based motherboard and videocard, and trade those for an Intel G35 based board with on-board graphics. I hope the store will accept that.

[rbmorse]

There's always Ebay...

[Dandel]

I just bought an 2600Pro because of the open documentation announcement. I am in the process of building a PVR system using MythTV. And now I read that the one part that I plan to use, video decoding, will not be documented!
My first thought was that they must be joking. But I realize that this is not the case.
What a ($!#@@$@, now I must find a way to return my AMD based motherboard and videocard, and trade those for an Intel G35 based board with on-board graphics. I hope the store will accept that.

You do not need to completely sell the whole rig... you could look in to switching to either the FGLRX Binary blob at least until the major features used by mytv are set, or get an nvidia graphics card ( and use their binary blob.) ...oh, and on a side note, it usually takes about 2 years for the documentation to usable driver to happen, and that's a safe estimate, because this next release will include initial DRM information for the gpu.

[givemesugarr]

You do not need to completely sell the whole rig... you could look in to switching to either the FGLRX Binary blob at least until the major features used by mytv are set, or get an nvidia graphics card ( and use their binary blob.) ...oh, and on a side note, it usually takes about 2 years for the documentation to usable driver to happen, and that's a safe estimate, because this next release will include initial DRM information for the gpu.

well, in this particular case i expect that the time would be considerably reduced. usually opensource project didn't have people who would work directly on them and this could speed up the process. i personally expect a good driver before the second half of 2008 or in the beginning of it and a good 3d accelerated driver in the first quarter of 2009. we'll just have to wait and see.