RapiHD: CUDA and H.264

that is AWESOME

i can't WAIT to buy this
 
Really slick. NVIDIA's also sponsoring a competition for the implementation of the LAME "MP3" encoder on CUDA, but nobody's really given it a good go yet (multithreading psy models is tricky shit).

Good to see CUDA's starting to get used for something. I dig their little applet too.
 
its stuff like this and physx that will keep me buying nvidia over ati.
 
its stuff like this and physx that will keep me buying nvidia over ati.

ya, same here. I was looking over all those 4850 specs, but ATI does not have anything like this at all

Also, looks like we wont be seeing this for a couple months

" Sound too good to be true? It's not. Elemental Technologies’ RapiHD™ Accelerator for Adobe Premiere® Pro will be available in the fall of 2008."

http://www.elementaltechnologies.com/products.php?id=5
 
but ATI does not have anything like this at all

Yes it does, you can use CAL on ati based cards
http://ati.amd.com/technology/streamcomputing/sdkdwnld.html

The RapidHD solution is currently only a baseline H264/AVC encoder, but will probably end up being the first decent GPU accelerated encoder available. The one im really looking forward to is Availmedia/x264 GPU encoder, which will be open source and fit in with all the other x264 encoder GUI's that most consumers use.
 
but what products are around that use it?

nVidia is pushing CUDA with physx and now this, something incredibly useful for just normal people; why doesn't ATI back CAL as much?

My guess would be nvidia has more money to spend where in new markets where the return on investment is unknown. And because of this NV is ahead of the curve, they both were working on GPGPU programing software at the same time, but because NV has more money to reinvest into itself, they are ahead currently.

It is all a work in progress for both companies, and this is one of the reasons why nvidia GPUs have been a better buy for the past few years, atleast IMO. The only thing NV is really lacking right now is folding@home support, and I believe their has been a big push from NV to get this up and running too.
 

CAL is nothing like CUDA though. CUDA is basically C with a few extensions to support Nvidia's stream processing data structures. The hard part is coming up with the algorithm...not dealing with learning new syntax or writing assembly.

CAL is basically assembly code and is way lower in the software stack. The closest thing to CUDA on the ATI side is Brook but that's an academic project without the corporate backing and resources that CUDA has from Nvidia.

 
the nVidia 9-series can take advantage of CUDA right? If so, I'm very interested in this, as fast as quad cores are for video encoding, it does get annoying after a while.
 
Yeah cool but how is the quality of the video afterwards?

whatever quality you pick it to be?

TerraPhantm said:
the nVidia 9-series can take advantage of CUDA right? If so, I'm very interested in this, as fast as quad cores are for video encoding, it does get annoying after a while.

yeah anything 8-series and newer
 
Yeah cool but how is the quality of the video afterwards?

The current RapiHD solution is a H264 baseline encoder only. So quality will not be as good as Ateme(Nero)/x264 solutions currently available, due to the lack of available encoding options. So if your encoding videos for there highest quality then CPU encoding will still be the way to go.

Once RapiHD/Avail media get there GPU H264 high profile encoders out, then the quality will be the same as current cpu based solutions. The wiki here lists the differences between the profiles and what there intended profiles are for such as

Baseline - this profile is used widely in videoconferencing and mobile applications
High - primary profile for broadcast and disc storage applications, particularly for high-definition television applications

http://en.wikipedia.org/wiki/H.264/MPEG-4_AVC
 
Anyone here know how CUDA works with SLI? For example could i keep an old 8800 and stick in a newer card (e.g. GTX 280) with it and have them work together for things like physx or video encoding?
 
So this is basically an encoder using the GPU on the video card to enhance the number crunching, right? What's the big deal... ATI has been doing this for what, 4 years now with their Avivo converter - I used to use that on an old X1300 card to give me some pretty serious bumps in encoding speed for h.264 clips/movie crunching. And ATI still offers the capability even with improved versions of the encoder - meaning the old cards still work, they're not requiring you to own the latest and greatest hardware just to get a speedboost in encoding like this Nvidia encoder seems to be doing (requires an 8xxx series card or better).

Wonder why Nvidia seems to be so late to this party, and the recently announced Folding@Home client for Nvidia GPUs too - ATI has been way ahead on both counts for years. It's nice to see 'em in the game now considering Nvidia has that old saying "The way it's meant to be played." They're just late starters I suppose. :)
 
I can't wait for this! :cool:

I hope this will work with my 8600GT.

BTW, which graphics cards will this work for? Obviously the new ones coming, but any from the 7, 8, or 9 series?
 
Wonder why Nvidia seems to be so late to this party

The ATI encoder didnt actually use the GPU at all, hacked versions of the encoder that allowed it to run on Nvidia and even Intel intergrated graphics chips showed exactly the same speed.

The reason it was so fast, was because of the absolute crap quality it delivered, if you encoded the same file at the same bitrate with either Nero (Ateme) or x264 then you would be astounded at the differences. There was some graphs in a video codec comparison on doom9 sometime ago and the ATI encoder was light years away quality wise from any of the H264/AVC encoders out there
 
Anyone here know how CUDA works with SLI? For example could i keep an old 8800 and stick in a newer card (e.g. GTX 280) with it and have them work together for things like physx or video encoding?

It's tough to answer your question until the new cards come out, CUDA is updated to support them, and NVIDIA releases their CUDA based PhysX. I do know it's possible to run CUDA on any card (or combination of cards) in a multi-GPU system, so I would guess it would work.
 
So this is basically an encoder using the GPU on the video card to enhance the number crunching, right? What's the big deal... ATI has been doing this for what, 4 years now with their Avivo converter - I used to use that on an old X1300 card to give me some pretty serious bumps in encoding speed for h.264 clips/movie crunching. And ATI still offers the capability even with improved versions of the encoder - meaning the old cards still work, they're not requiring you to own the latest and greatest hardware just to get a speedboost in encoding like this Nvidia encoder seems to be doing (requires an 8xxx series card or better).

Wonder why Nvidia seems to be so late to this party, and the recently announced Folding@Home client for Nvidia GPUs too - ATI has been way ahead on both counts for years. It's nice to see 'em in the game now considering Nvidia has that old saying "The way it's meant to be played." They're just late starters I suppose. :)

I'll start with the Folding@Home client. Why do you think Stanford only supported AMD's GPUs, when NVIDIA has the fastest GPUs in the planet, for almost two years now ?
It's definitely not because NVIDIA is a "late starter" and I'll give you a hint: money was involved.

As for CUDA and H.264 and your Avivo comparison, one has nothing to do with the other. CUDA allows developers to access the computing power of a GPU directly. That means Stream Processors are being used for all the calculations necessary for encoding/transcoding videos and the advantages are clear, since a modern GPU is much more powerful than any high-end Quad-Core CPU in the market. Avivo is a piece of hardware dedicated to one task and one task alone. It is nothing like CUDA.
 
I'll start with the Folding@Home client. Why do you think Stanford only supported AMD's GPUs, when NVIDIA has the fastest GPUs in the planet, for almost two years now ?
It's definitely not because NVIDIA is a "late starter" and I'll give you a hint: money was involved.

As for CUDA and H.264 and your Avivo comparison, one has nothing to do with the other. CUDA allows developers to access the computing power of a GPU directly. That means Stream Processors are being used for all the calculations necessary for encoding/transcoding videos and the advantages are clear, since a modern GPU is much more powerful than any high-end Quad-Core CPU in the market. Avivo is a piece of hardware dedicated to one task and one task alone. It is nothing like CUDA.

AVIVO = Purevideo (MPEG4, etc.)
AVIVO UVD = PurevideoHD (VC-1, H.264, etc.)

They both are only used for video decoding (taking the load off of the CPU).

In fact, ATI's AVIVO UVD is actually more powerful than NVIDIA's PurevideoHD, mainly because it is able to take more of the processing load off of the CPU than NVIDIA's cards can.
 
AVIVO = Purevideo (MPEG4, etc.)
AVIVO UVD = PurevideoHD (VC-1, H.264, etc.)

They both are only used for video decoding (taking the load off of the CPU).

In fact, ATI's AVIVO UVD is actually more powerful than NVIDIA's PurevideoHD, mainly because it is able to take more of the processing load off of the CPU than NVIDIA's cards can.

Yes, UVD is a little better than PureVideo, but one thing has nothing to do with the other. Hardware decoding capabilities is not the same as writing a CUDA app that uses Stream Processors to decode/transcode a video or calculate physics equations, etc. With CUDA, you can code an application that harnesses the GPU's processing power for anything you want. Hardware video decoding capabilities only have one purpose.
 
Yes, UVD is a little better than PureVideo, but one thing has nothing to do with the other. Hardware decoding capabilities is not the same as writing a CUDA app that uses Stream Processors to decode/transcode a video or calculate physics equations, etc. With CUDA, you can code an application that harnesses the GPU's processing power for anything you want. Hardware video decoding capabilities only have one purpose.

Um, I never said that it was the same as a CUDA app. I was simply pointing out that the AVIVO UVD was more powerful than PurevideoHD, only in terms of decoding HD codecs/formats. It may only be one purpose but it is very important for HTPCs.

I hope they will be able to code CUDA with the LameMP3 encoder/decoder. Using my GPU for video AND music would be great.
 
Um, I never said that it was the same as a CUDA app. I was simply pointing out that the AVIVO UVD was more powerful than PurevideoHD, only in terms of decoding HD codecs/formats. It may only be one purpose but it is very important for HTPCs.

I hope they will be able to code CUDA with the LameMP3 encoder/decoder. Using my GPU for video AND music would be great.

I'm unaware of any feature on AVIVO UVD that is not supported by PureVideoHD. The latest video process core from nvidia supports full stream H.264 decode, the features are pretty much the same although there may be quality differences. I have found nvidias noise reduction to be a bit more subtle than the sledgehammer employed by AMD (and seemingly cribbed from ffdshows temporal denoiser - they artifact in the exact same manner).

The advantage PVHD has is that quality mangling feature likes noise reduction and edge enhancement are not cranked to 11 by default as they are with AMD solutions, but at least you can turn them off now (you couldn't in the past).

FWIW the video processing features of both brands of GPUs are provided by dedicated silicon on the GPU. For this reason the lowly 8400GS rocks the latest video bits while uber cards do not.

GPU based trasncoding is coiol, but Adober Premier? Bleh. Gimme Vegas Pro any day, or a stand alone app.
 
I'm unaware of any feature on AVIVO UVD that is not supported by PureVideoHD. The latest video process core from nvidia supports full stream H.264 decode, the features are pretty much the same although there may be quality differences. I have found nvidias noise reduction to be a bit more subtle than the sledgehammer employed by AMD (and seemingly cribbed from ffdshows temporal denoiser - they artifact in the exact same manner).

The advantage PVHD has is that quality mangling feature likes noise reduction and edge enhancement are not cranked to 11 by default as they are with AMD solutions, but at least you can turn them off now (you couldn't in the past).

FWIW the video processing features of both brands of GPUs are provided by dedicated silicon on the GPU. For this reason the lowly 8400GS rocks the latest video bits while uber cards do not.

GPU based trasncoding is coiol, but Adober Premier? Bleh. Gimme Vegas Pro any day, or a stand alone app.

I didn't say it had more features. AVIVO UVD is more POWERFUL than PurevideoHD, i.e. it can take more load off of the CPU. That's it! Nothing else.

The uber cards do have PurevideoHD and are far more powerful than the 8400GS in HD decoding. The 8800GTS 320/640 and the 8800GTX/Ultra only have Purevideo, making them weaker/incapable in that department. However, the 8800GT/GTS(G92) and 9600GT/9800GTX/GX2 all have PurevideoHD and take far more load off of the CPU than the 8400GS ever could (it being the weakest card with PurevideoHD). Also, unless you get an 8400GS with dual link DVI, you won't even be able to reach resolutions beyond 1920x1200, let alone 2560x1600 like the 8500GT cards (and above) are capable of.

Another thing, Adobe Premiere and Vegas do not use the GPU to do anything or any processing (unless CUDA is counted, but that's another story).

The only program I am aware of that can use the GPU to help render video is Adobe After Effects CS2 and CS3. It also uses OpenGL as the API.

http://www.adobe.com/products/aftereffects/opengl.html
 
In a way you are both right.

I think it may help the discussion if the more detailed terms AVIVO+UVD and PureVideo+VPx were used. AVIVO and PureVideo/PureVideoHD are the technology umbrellas, whereas UVD and VP1/VP2/VP3 are the hardware optimisations that enable elements of AVIVO and PureVideo to use the GPU instead of the CPU.

So, for example, the 2900XT and 8800GTx used versions of UVD & VPx that did half the video decode (so the other half had to be done by the CPU). Post 2900XT, ATI GPUs used UVD with full video decode, but post 8800GTx, Nvida GPUs used VP2 which does 3/4 of the video decode (not sure about VP3).
 
And this has nothing to do with what I posted in the OP. It's CUDA being used to encode/transcode HD video. It's direct manipulation of a GPU's processing power, while UVD and PureVideo are a totally different "beasts".
 
I'm wondering how well this would work for FEM/FEA analysis. I may have to write some code this weekend and find out. XD
 
wait the youtube is converting to iPOD or iPHONE.... thats not HD.. just HD-source... wait.. i'll watch it again before sounding really dumb....
 
Does CUDA work on Vista yet? I wanted to play with it a while back but it only supported XP, which is kind of strange since nVidia is such a pusher of DX10 tech.
 
Does CUDA work on Vista yet? I wanted to play with it a while back but it only supported XP, which is kind of strange since nVidia is such a pusher of DX10 tech.

I'm prety sure it does.
 
Does CUDA work on Vista yet? I wanted to play with it a while back but it only supported XP, which is kind of strange since nVidia is such a pusher of DX10 tech.

The beta of CUDA 2.0 does.
 
Back
Top