ATI GPU transcoding app?

[slapnutz]

As some may know ATI announced 40nm and Dx11 GPUs at Ceatec in Japen recently. However my question is concerning the following slide...

We know about nVidia's shown this with some 3rd party partner (i forget the application name) but does anyone know what application Amd/ati actually used to illustrate the GPU accelerated transcoding results below?

Cheers.

 

[3dcgi]

Maybe Cyberlink since it's supposed to work with ATI GPUs.
http://www.tgdaily.com/html_tmp/content-view-38117-140.html

 

[madyasiwi]

I believe that particular was referring to PowerDirector.

 

[ShaidarHaran]

Sick increase in perf. Almost 2x > than real-time is pretty damn good for 1080P!

 

[madyasiwi]

Oops! 3dcgi beat me to it

 

[Captain Chickenpants]

That is quite an impressive time. Although I would like to know if the resulting video is of the same quality/bitrate.

 

[slapnutz]

That is quite an impressive time. Although I would like to know if the resulting video is of the same quality/bitrate.

True, also know what the configuration of the 4800 cards would be nice. However this is PR spin so we'll never get the true picture from those particular tests.

EDIT: from the link posted earlier....

According to Shuichi, it will take about 30 minutes to process four full-length movies and compress them into handheld-friendly 200+ MB files.

Geez, getting a standard def movie around 200MB (or even 400MB) with decent qualit is hard enough. I guess were not talking HDTV playback type quality here.

 

[Mariner]

Wow. It's so much faster at transcoding, they even added an extra letter 'c' to the word "Accelerated". Now that is fast.

 

[Nick]

Funnny how they use a Core 2 Duo to make Intel look bad.

I'd really love to see how Phenom and Core 2 Quad compare for this benchmark. Also, does anyone know if this software actually makes good use of the latest SSE instructions?

 

[CarstenS]

I wonder, how many RV770-Chips were used.

 

[Dave Baumann]

That demo is one.

 

[ShaidarHaran]

Funnny how they use a Core 2 Duo to make Intel look bad.

I'd really love to see how Phenom and Core 2 Quad compare for this benchmark.

Sure, they could use a quad core, but it's not like the C2D used in this test is slow. 3.16GHz Wolfdale is a beast at transcoding.

Also you have to consider price. The only quad core faster than the E8500 is the QX9770, and that's a $1500 CPU.

Now, do you really want to see a $300 GPU be 9x faster than a $1500 CPU? That's a 45x price to performance ratio advantage in favor of the GPU. At least with the Wolfdale used in the test it's closer a 10x advantage for the GPU.

Also, does anyone know if this software actually makes good use of the latest SSE instructions?

Unfortunately, Cyberlink doesn't list any use of SSE in Power Director on their website, and Google turns up nothing either so I can't say for sure.

 

[Scali]

Didn't AMD release a transcoding application years ago with the 'Avivo' stuff?
Perhaps they just updated that and/or made a regular interface for it so it could be used in standard applications?

 

[Scali]

Sure, they could use a quad core, but it's not like the C2D used in this test is slow. 3.16GHz Wolfdale is a beast at transcoding.

Also you have to consider price. The only quad core faster than the E8500 is the QX9770, and that's a $1500 CPU.

Now, do you really want to see a $300 GPU be 9x faster than a $1500 CPU? That's a 45x price to performance ratio advantage in favor of the GPU. At least with the Wolfdale used in the test it's closer a 10x advantage for the GPU.

I think his point was rather that video transcoding generally scales well to 4 cores, so a cheap quadcore like the Q6600 might actually deliver better transcoding than the C2D they used.

 

[madyasiwi]

From MainConcept H.264 and VC-1 codecs feature list. They don't mention anything about SSE-2/3 except in the decoding part. So...

 

[madyasiwi]

Didn't AMD release a transcoding application years ago with the 'Avivo' stuff?
Perhaps they just updated that and/or made a regular interface for it so it could be used in standard applications?

The one which is later found out to only use the CPU, even though it was initially advertised as 'requires Radeon X1XXX' or something like that?

 

[ShaidarHaran]

I think his point was rather that video transcoding generally scales well to 4 cores, so a cheap quadcore like the Q6600 might actually deliver better transcoding than the C2D they used.

And still be much much slower than the GPU in question. 3.16GHz Woldale -> 2.4GHz Kentsfield yields ~70-80% perf. difference in transcoding apps (without SSE4).

My point still stands in that all but the fastest quad cores in existence will be an order of magnitude slower than a 4800 in this test.

 

[Davros]

the prograam they use can you buy it ?

 

[Scali]

the prograam they use can you buy it ?

Yes, it's available from the same source as the GPU-accelerated Havok libraries.

 

[Dave Baumann]

The one which is later found out to only use the CPU, even though it was initially advertised as 'requires Radeon X1XXX' or something like that?

Later found out? It was indicated as such at launch:

http://www.beyond3d.com/content/reviews/25/

As surprising as it may be, though, the application is also software based itself and is not using using any graphics hardware acceleration yet. Evidently ATI utilised their knowledge of video encoding, taken from the consumer video side of their business, to create the transcoding routines and speed them up in relation to other software solutions.