ATI Universal Video Decoder (UVD)

Geo

Mostly Harmless
Legend
http://www.hkepc.com/bbs/news.php?tid=641493

Well, December for this particular part would clearly seem to be aimed at the Vista mainstream market. From that pov, the good news is it is certified for full 1080p.

What I can't quite figure out is if they've actually done something different, and if so, what? Isn't the Xilleon core supposed to be embedded in all X1K already as part of Avivo?

So what's this ATI UVD thingy? A new branding of a subset of Avivo? Or something actually new on the hardware side?
 
BRiT said:
Smells like marketting...

I was thinking the same.

First, both ATI and NVidia convince the PC-world the that Pixel Shader 3.0 pipelines will solve all display-tasks, now and in the future. But it seems ATI and NVidia have quietly admitted "jack of all trades == master of none!" Pixel-shading 3D-pipelines (at least those in existing hardware) don't deliver sufficient pixel-throughput to fully offload 1080p HD-video playback.

A purpose-designed 2D-video block will deliver better efficiency, power-consumption, and final throughput. But I suppose the next question is whether that 2D-processor should be micro-architected as programmable or fixed-function -- i.e. what does Xilleon add to the R5xx/RV5xx AVIVO-pipeline?
 
Well, ATI has been talking about a new version of Xilleon was on the way. So it occurs to me to wonder if that might be what we're seeing here --the new version of Xilleon replacing the old one.
 
This sort of card would actually be pretty much ideal for my requirements. I'm thinking of building a small (and possibly passively-cooled) media PC. Don't need any PVR or TV functionality but I would like it to have the ability to play 720p (and possibly later 1080p) h.264 video, possibly to use it as a Blu-Ray/HD-DVD player in the future. At present, this rules out all but the fastest CPUs which bumps up the price somewhat.

A low-end card which included this UVD functionality would mean I could build a PC with a cheap Celeron M or Semperon yet still have the h.264 functionality. Quite an attractive option if you ask me. I suppose it's too much to hope that ATI will include this in future IGPs? This could potentially cut down the cost even further.
 
Mariner said:
I suppose it's too much to hope that ATI will include this in future IGPs? This could potentially cut down the cost even further.

I don't think it's too much to hope at all, just a question of when. If memory serves, they have a new IGP coming middle of next year. What would be lovely is if they said this was part of RS690 as well, but I haven't heard that.

Of course I'd also be utterly unsurprised if future IGPs are AMD-only. . .
 
Mariner said:
I suppose it's too much to hope that ATI will include this in future IGPs? This could potentially cut down the cost even further.
Since R6XX ASICs will also include it, it should be only a matter of time until an IGP becomes that sophisticated. At 65nm and further down the road, the cost would also be pretty negligible.

Broadcom already has a very powerful dedicated ASIC, which is built in Toshiba´s current HD-DVD players and since ATi also targets (some SKUs specifically were built for that purpose in mind) the video market, such enhancements could mean everything to a potential customer in the future.

So, it´s actually a pretty smart choice by ATi. (Now, if they could not only architect this into the drivers, but also have DXVA-support for it, a potential customer using 3rd-party multimedia-players would be even happier.)
 
My problem is that I was hoping to build my new Media PC sooner rather than later. Looks as though it would be best for me to wait a little longer and see what develops. Grrr. :mad:
 
asicnewbie said:
I was thinking the same.

First, both ATI and NVidia convince the PC-world the that Pixel Shader 3.0 pipelines will solve all display-tasks, now and in the future. But it seems ATI and NVidia have quietly admitted "jack of all trades == master of none!" Pixel-shading 3D-pipelines (at least those in existing hardware) don't deliver sufficient pixel-throughput to fully offload 1080p HD-video playback.

Actually, NVidia admitted the opposite, which is why they included the video processor (VP) engine in the NV4x, otherwise, they could have left it out and just used pixel shaders. My guess is that the UVD is ATI's version of a VP, a highspeed scalar processor with hooks for codec operations that can handle VC-1, H.264, and MPEG-2. The entropy coding portion of the pipeline in particular, CABAC especially, requires acceleration. They'll claim HD-DVD, BluRay, and DVD playback in marketing literature == Universal. (Wha? No On2 support for YouTube!?! :) )
 
BRiT said:
Smells like marketting...
Yeah, NV has PureVideo and Intel just went with Clear Video. ATI probably wanted to make the Video part clearer, and Universal ties in with AVIVO's transcoding, at least in my mind.

Edit: To be clear, it would seem there's some hardware improvement as well. I think they reacted a bit to the other names and took advantage to rebrand the VO part of what's probably AVIVO 2.

Rys said:
And what better excuse for a renewed marketing push than a genuine improvement?
Better than the reverse, anyway. :)
 
Last edited by a moderator:
Smells like both to me. If it's really RV516-based, it needs to be new silicon to do what they claim, in my opinion. And what better excuse for a renewed marketing push than a genuine improvement?
 
That seems likely. The capability claimed is clearly a step up from RV515 and even RV530, as the last round of testing showed.

So there is almost certainly some new hardware capability involved, tho it is a bit frustrating that Xilleon got cited last time and Xilleon gets cited this time, and yet clearly something "more" is there this time.
 
DemoCoder said:
Actually, NVidia admitted the opposite, which is why they included the video processor (VP) engine in the NV4x, otherwise, they could have left it out and just used pixel shaders. My guess is that the UVD is ATI's version of a VP, a highspeed scalar processor with hooks for codec operations that can handle VC-1, H.264, and MPEG-2. The entropy coding portion of the pipeline in particular, CABAC especially, requires acceleration.

I didn't know that, thanks for the correction! And wouldn't the VP add some integer units for fast pixel-level motion-compensation? For example, a 3D-pipeline with a raster-output of 4-pixels/clock is going to struggle with complex 2D-motion @ 1080p. (Or does the 3D-pipeline already already have a special integer packed-pixel format for 2D-video?)

They'll claim HD-DVD, BluRay, and DVD playback in marketing literature == Universal. (Wha? No On2 support for YouTube!?! :) )

And what about Realvideo10?
 
Well, motion-comp is usually a fixed-function block. VP has scalar/integer ops I think, haven't looked at it in while. Obviously, it could still be generalized more since more of the codec could be moved from CPU to GPU than is currently supported.
 
One of the slides says "complements mainstream cpus", which certainly would suggest they've offloaded more.
 
Back
Top