ATI Avivo XCode released!

[bond]

i would be really surprised if ati's encoders would perform any good qualitywise

after all they are meant for maxing out the speed, so why offer high quality bringing the speed down?

[videomixer9]

I think their goal is too use the GPU to max the speed while keeping the quality as you know it from other encoders. Speed says nothing about quality, just because it's damn fast doesn't mean it's bad quality, it can also mean it is better written or makes better use of processors that are meant for graphic processing ... I think nvidia and ati both showed us that it is their goal to make more use of their graphic processors to accelerate video.

[bond]

you are right, and i said "i would be surprised", which means i dont know yet how things are handled

still before we can do valid speed and quality comparisons we need to know what codec settings ati uses, so gimme some damn samples!

[breez]

Note: at this point the encoding is done entirely in software, no help from GPU.

[bond]

Originally Posted by breez

Note: at this point the encoding is done entirely in software, no help from GPU.

what tells you this?

[slavickas]

Originally Posted by bond

what tells you this?

article at anandtech states so, although maybe their verson was older

[alcuin]

Originally Posted by bond

what tells you this?

http://www.anandtech.com/video/showdoc.aspx?i=2645&p=3

The other item of interest that ATI briefed us on is the ATI Avivo Video Converter. We mentioned in our previous coverage that ATI was working on GPU accelerated video transcode, to speed up the conversion of videos from one format to another (e.g. MPEG-2 to H.264). Unfortunately, the GPU accelerated transcode isn't yet ready for debut, but what ATI is making available is the software front end for it. .... However, keep in mind that despite ATI's release of this tool, the video conversion itself is done entirely on the host CPU and not on the GPU. So why bother? Well, thanks to ATI's experience in dealing with video, they have optimized a number of the transcoding algorithms so that conversion using the utility is actually faster than on other software solutions.

et al.

regards,
-e

[bond]

Originally Posted by alcuin

However, keep in mind that despite ATI's release of this tool, the video conversion itself is done entirely on the host CPU and not on the GPU. So why bother?

wtf?

[LiFe]

Originally Posted by bond

wtf?

The article actually has no usefull info. No point in quoting speed figures (even with pretty graphs) if you don't show features and quality.

[Sharktooth]

So why the hell it does work on 1xxx series only?!?

[Sulik]

Apparently, ATI's H.264 encoder is currently baseline profile only (No B-frames or CABAC), which definitely would limit its potential in quality (Would also explain the results at MSU when compared with Main profile codecs).

It would be interesting to compare the quality when using CAVLC in x264.

[CeeJay.dk]

Originally Posted by Sharktooth

So why the hell it does work on 1xxx series only?!?

Because the app check for the presence of a 1xxx series card and denies running if it is not found.
I read on another site that ATI did this because they wanted to add hardware-acceleration later and that it would only be possible to do on a 1xxx card.

I however believe it has more with ATI wanting people to upgrade rather than giving all their customers (or worse .. their competitors customers) something to enjoy for christmas.

Since it is done entirely on the CPU, it should be possible to either do as Yong did , or use a program to fake the make and model of the card so the driver detects a 1xxx card, or modify the driver or the program to remove this check or make it so it always passes it.

ATI is also busy at licencing this new ultrafast piece of code to thirdparty developers so versions of it might appear in other encoding applications , that do not have the 1xxx restriction.

I'm fully convinced that its only a matter of time before someone develops a hack or even reverseengineers the program.

UPDATE : That other site I remembered seeing the news at , was Elite Bastards dot com:
http://www.elitebastards.com/page.php?pageid=13022

[Doom9]

Yong, what's your filter chain to write the output? Have you managed to get a raw stream from the encoder instead? I guess that would be more useful, so we can mux using mp4box, which we know does as it should.
I ran the installation routine as well and the filters at least show up in graphedit.

[Sulik]

I can get a raw stream from the encoder:
DShowSource-> "ATI Video Scaler" -> "ATI MPEG Video Encoder" -> File Dump filter

[bond]

Originally Posted by Sulik

I can get a raw stream from the encoder:
DShowSource-> "ATI Video Scaler" -> "ATI MPEG Video Encoder" -> File Dump filter

which filters does the tool include?

[Sulik]

It includes video scaler, video encoder, audio encoder, mpeg decoder, multiplexer and file writer.

[Doom9]

a lot of filters, all having a name starting with ATI.. just go ahead and try it.. it'll work on your box as well there's also an audio encoder, playback filters, resizing filter, capturing filters.. quite a bunch.

[bond]

yeah just found it out

did anyone get things to work in graphedit? i till now was only able to connect divx5 .avi files via the m$ splitter to the ati decoder and than to the encoder

but when encoding it never finishes but produces a bigger and bigger getting file

[Doom9]

I did a short test run. AviSynth script (the same used for the codec comparison qualification) -> ati mpeg video encoder -> dump filter (from the graphedit package on my site). I trimmed the script to 2000 frames, encoded, muxed to mp4 and it plays just fine. I haven't found out how to directly create an mp4 so far. I saw no moving of the progress slider in Graphedit so that's kinda unnerving.

I can't say anything about speed because I'm currently encoding VC-1 and that occupies both cores (total load about 90%).

@edit: hmm, I actually managed to write an mp4. I connected the video encoder to the ati mpeg multiplexer, set it to output mp4 generic, and connected its output to the ati mpeg file writer, but I'm afraid the resulting mp4 will just come up with a black screen in MPC.

[bond]

here my findings with asp and avc in mp4:

ASP:

- seems to be simple profile + max 2 non adaptive b-frames
- no qpel, no gmc
- h263 quant
- places the vol together with every keyframe in mp4, which is not allowed in mp4 (seems to be a leftover of the avi workaround doing the same)
- the mandatory ctts atom carrying the b-frame timestamps is missing in mp4
- wanted filesize isnt matched correctly
- i see visual artefacts here and there when decoding
- signals a strange aspect ratio
- seems to place keyframes every ~50 frames but not exactly
- signals the average bitrate as 0
- mp4 seems to play fine in mplayer, vlc and with gabest+ffdshow
- quality is clearly not good

AVC:

- seems to be baseline profile + max 2 non adaptive b-frames
- max 1 reference frame
- no cabac, no weighted (bi)prediction, no multiple slices
- loop uses 0:0, but was never enabled in the encode
- again the strange sar signalled
- the mandatory ctts atom carrying the b-frame timestamps is missing in mp4
- wanted filesize isnt matched correctly by far
- i am unable to play the file with nero, haali and gabest mp4 splitter (jerky in qt7), mplayer and vlc handle it fine
- quality is clearly not good