[DivX vs FFdshow] Decoders Benchmarks...

[Kurtnoise]

http://labs.divx.com/node/939

I would like to have some comments about this especially from FFdshow developpers...

:thanks:

[SeeMoreDigital]

Perhaps if DivX was willing to offer their v6.6.1 decoder (to a select few of us), we could conduct our own tests ;)

[GmorG McRoth]

Decoder is free I think.

[check]

Since even the 2004 version of ffdshow there can play both files listed above 30fps, the testing seems like a non event to me.

[_xxl]

Tested samples with timecodec.No pp or other filters used.Amd XP 2000+ 512mb ram.
ftp://mplayerhq.hu/MPlayer/samples/V-codecs/DX50-DivX5/test.b-frames.divx5.avi
ftp://mplayerhq.hu/MPlayer/samples/V-codecs/DX50-DivX5/test.no-b-frames.divx5.avi
ffdshow-tryouts libavcodec mpeg-4 decoder rev 1620
User: 2s, kernel: 0s, total: 2s, real: 2s, fps: 309.6, dfps: 292.6
User: 2s, kernel: 0s, total: 2s, real: 2s, fps: 345.4, dfps: 315.9
DivX 6.7.0.28
User: 3s, kernel: 0s, total: 3s, real: 3s, fps: 243.6, dfps: 226.3
User: 3s, kernel: 0s, total: 3s, real: 3s, fps: 266.6, dfps: 251.5

[honai]

That "benchmark" is useless. Nothing whatsoever gives an indication how well the decoders scale. And if the only viable result is that all tested decoders can render all frames at the nominal framerate you can derive that result from it, nothing more.

It reminds me of the synthetic benchmark re: video card performance when manufacturers ran their latest products at insane resolutions, implying that the architecture would scale like that. Not surprisingly, when the first games actually made use of these resolution the GPUs couldn't actually deliver on the promises.

So the sample of 428fps for DivX 6.6.1 in a synthetic benchmark means absolutely nothing in real-world scenarios.

[check]

honai, your complaint may be true for game benchmarks, but video benchmarks are not synthetic, they are the same as playing back a video.
The only difference in this test from reality is that instead of the renderer only requesting a max of one frame per 1/60secs, it requests each frame as soon as it recieves the previous.

[honai]

No, in reality I would use ffdshow precisely because - gasp! - of its filters. In the "benchmark" above no filters were used, so the "benchmark" is purely synthetic.

Also, how is a fps of >400 for video rated at 23-25fps "real"? It's purely synthetic, and just because a decoder can output raw 400fps it doesn't infer that (1) the decoder would perform well at higher resolutions, or (2) the decoder would perform well with several chained post-processing options activated.

Bring me a benchmark where DivX implements the same filter set as ffdshow, at higher resolution, and actually uses those, then we can talk.