nVidia likely to remain accelerated video king?

[bridgman]

With respect, your wording is not sufficiently precise to convey the meaning you want, in fact most people interpret it differently.

When you say "VDPAU", do you mean "the NVidia implementation of VDPAU on recent GPUs" or "VDPAU the API" ?

I sincerely doubt that to be true wrt to Nvidia. I can see the difference and the VDPAU documentation clearly refers to being able to offer post processing and compositing support. I see no reason for the driver feature being exposed and then being unused.

The above comment suggests that you are saying *all* VDPAU implementations would include post processing simply because the API allows it. That might be correct if there is only a single VDPAU implementation (NVidia's proprietary driver), but it's unlikely that another implementation would include all of the post processing you get in a proprietary driver.

That was the distinction gbeauche was trying to make, and IMO he was correct to do so.

[bridgman]

FWIW I didn't see any indication that VDPAU required dedicated decode hardware, any more than VA-API or XvBA do. It's probably fair to say that you wouldn't use one of those APIs if you weren't doing *some* degree of decode acceleration, but that could be some shader-friendly subset of the work (eg MC + deblock) rather than the whole decode stack.

That said, for a shader-only decode I still think that adding shader calls (through Gallium3D if possible) to ffmpeg/libavcodec makes more sense than trying to re-implement the rest of the decode stack in a VDPAU / VA-API / XvBA driver (so in that sense I agree with the spirit of what you are saying if not the letter), but I seem to be the only one leaning towards leveraging an existing software decoder

[kazetsukai]

FWIW I didn't see any indication that VDPAU required dedicated decode hardware, any more than VA-API or XvBA do.

VDPAU Requires PureVideo 2 (or above) compliant hardware, present on GeForce 8+.

The 8800GTS and 8800GTX first generation models could NOT play back VC-1/wmv3 because they only had PureVideo 2 hardware, even though their shader / stream processor architecture is identical to the later models.

[bridgman]

Really ? I didn't see anything like that in the VDPAU API spec - in fact NVidia proposed that other hardware vendors use the VDPAU API as well. That wouldn't make sense if the API spec required an NVidia proprietary decoder block. The NVidia proprietary driver's *implementation* of VDPAU may require PureVideo2 or higher, but that's just another argument for distinguishing between the API and the implementation.

[IsawSparks]

Really ? I didn't see anything like that in the VDPAU API spec - in fact NVidia proposed that other hardware vendors use the VDPAU API as well. That wouldn't make sense if the API spec required an NVidia proprietary decoder block. The NVidia proprietary driver's *implementation* of VDPAU may require PureVideo2 or higher, but that's just another argument for distinguishing between the API and the implementation.

If it's so agnostic why didn't AMD use it and not opt to create XvBA?

Oh wait, I know, because it isn't agnostic and neither is XvBA.

All of the this agnosticism is just a matter of keeping the OPEN SOURCE idealogy in focus, it has nothing whatsoever to do with function.

[Hephasteus]

The reason your 2.8 GHz P4 is able to play back two 1080p videos with 3-5% CPU utilization is because you are using GPU hardware decode assist.

Ya there's absolutely NO way my cpu could be decoding with 3-5% cpu utilization. Which is why I said the cpu could DO 2 1080p streams with the OS kicking in 3-5% OVERHEAD. Decoding 2 1080p videos on p4 is a nearly nailed 100 percent hyperthread situation and a nearly nailed 90 to 100 percent cpu usage situation with occasional stuttering on one or both videos.

Reading and comprehension.

[IsawSparks]

Two 1080P streams of what though? Of low bitrate HDTV rips? 40Mbit fullrate profile 4.1 and higher streams?

Sincerely, I'm finding all of this 'my CPU can do it just as well as your GPU' malarkey so tiresome. It's irrelevant to the conversation and basically filled with holes as an argument.

[Hephasteus]

VDPAU Requires PureVideo 2 (or above) compliant hardware, present on GeForce 8+.

The 8800GTS and 8800GTX first generation models could NOT play back VC-1/wmv3 because they only had PureVideo 2 hardware, even though their shader / stream processor architecture is identical to the later models.

http://en.wikipedia.org/wiki/Nvidia_PureVide

Purevideo 2 is fairly useless. Purevideo 3 is useless to me because I always stretch videos outside of what they can decode at anyway. They are nickle and dimeing this decode stuff to death. Which is why I said forget about all the earlier ones and just impliment the latest ones.
VP1's a waste. VP2 cards are bumpcracking themselves into oblivion.
VP3 cards are so underpowered they run full steam decoding this junk anyway. And VP4 cards are all done on pain in the butt process size 40nm and having a hard time getting market share.

[bridgman]

If it's so agnostic why didn't AMD use it and not opt to create XvBA?

Oh wait, I know, because it isn't agnostic and neither is XvBA.

All of the this agnosticism is just a matter of keeping the OPEN SOURCE idealogy in focus, it has nothing whatsoever to do with function.

Actually it's a lot simpler than that - XvBA had already been spec'ed, implemented and early versions were being given to ISVs before VDPAU was announced.

The APIs try to be agnostic in terms of hardware but once you have an implementation they are certainly not agnostic to your implementation.

[LinuxID10T]

Actually it's a lot simpler than that - XvBA had already been spec'ed, implemented and early versions were being given to ISVs before VDPAU was announced.

The APIs try to be agnostic in terms of hardware but once you have an implementation they are certainly not agnostic to your implementation.

Why wasn't the specs given to the public, as well as the ISVs?