NVIDIA VDPAU Reference Benchmarks

[phoronix]

Phoronix: NVIDIA VDPAU Reference Benchmarks

At the request of Phoronix readers curious about the NVIDIA VDPAU performance between different GeForce graphics cards, here are some results.

http://www.phoronix.com/vr.php?view=17445

[h**2]

What exactly is mp4-decoding compared to h264 decoding?

[oblivious_maximus]

What exactly is mp4-decoding compared to h264 decoding?

Most likely it's meant to represent mpeg 4 part 2, or DivX/XviD. As opposed to mpeg 4 part 10, which h264 falls under.

[FireBurn]

Any benchmarks using the VDPAU state tracker with mesa on r300, r600 & nouveue?

Would be interesting to see how this is progressing - even if it's only mpeg2

[tomm3h]

Can anyone offer any incite as to why VC-1 takes such a dive on newer hardware?

That would seem like a pretty serious regression to me, given the difference over other decoding tests.

Is it failing to decode via VDPAU and falling back to the CPU?

It would have been nice to spend a little more time looking into the reason behind it, Michael.

[mczak]

Can anyone offer any incite as to why VC-1 takes such a dive on newer hardware?

Those 3 cards which are faster use VP2. This version could not fully accelerate VC-1 on the gpu, in particular the bitstream decode is done on the cpu. I bet that's the reason, give it a slow cpu and things won't look so rosy :-).
Plus I'm not sure it would make even sense to have faster VC-1 acceleration, no idea if higher resolutions / multistream etc. are needed somewhere (this might be the reason why the gt2xx are a bit faster there than the later ones).

[willmore]

4K?

The Keppler generation is supposed to accelerate 4K video. Any chance you could test that? From the speed of the X.264 1080p decode, it looks like playing 2160p30 video might be 'iffy'. Then again, not everything needs to scale up 4x.

I might need to go test a few different members of my nVidia card collection to see what there is to see. I have a 1x G210 that might be interesting to compare to a 16x G210. Hmmm, I think I have an identical 16x G210 to the one I cut down to make the 1x.

[Kano]

The VC1 speed is faster with older cards because the cpu is very fast. Do the same benchmark with an atom cpu and things would be reversed. Due to a bug in the chip partly VC1 is software rendered. One other thing you don't see in those benchmarks is that accelleration depends on the video size, the older chips can not accellerate all res which are used in the web using flash.

[jyavenard]

Doesn't contain any interesting info

Unfortunately, this test contains no information that are actually of interest for what VDPAU matters: HTPC users.

All those figures are useless. They only show that all cards are fast enough to decode data. No one really care that it can decode data a 400 frame per second, when all you will ever need at the most is 60 fps.

As far as VDPAU is concerned , any VDPAU cards ever released, including the slowest 8200, will be fast enough to decode data, including the highest quality bluray content.

What matters however is how fast they will perform for deinterlacing content.
You would find that the 520 is much slower than the GT220 and can't be used at the highest quality settings.

[Evgeny]

Unfortunately, this test contains no information that are actually of interest for what VDPAU matters: HTPC users.

All those figures are useless. They only show that all cards are fast enough to decode data. No one really care that it can decode data a 400 frame per second, when all you will ever need at the most is 60 fps.

As far as VDPAU is concerned , any VDPAU cards ever released, including the slowest 8200, will be fast enough to decode data, including the highest quality bluray content.

Agree. This test on a new hardware is meaningless. It should contain data for old laptops and htpc with modest hardware.
Also as far as I know vdpau runs from 8400gs not 8200. And the slowest one should be 8600m gpu (mobile gpu for laptops).

But if anyone is intrested may be we could post here test results with a relevant data.