[FFmpeg-devel] transcoding on nvidia tesla
Fri Feb 1 00:17:42 CET 2008
I guess alternatively, if it's too difficult or crazy to rewrite parts of
ffmpeg for the tesla, how hard would it be to offload the computations to a
high-performance gpu that's already supported by standard linux os's?
Those GPU's and/or chips optimized for video transcoding do magic.
Essentially anyone on the planet would be able to offload their conversion
to their gpu, and a 6gig dvd could go from dvd to anything, several thousand
times faster. Or producers could convert to dvd/hddvd several thousand times
Also, a company called www.ripcode.com does something similar, but
leveraging Texas Instruments DaVinci chip
(www.ti.com/corp/docs/landing/davinci/index.html) which is an optimized hd
This enables them to do transcoding on the fly, but nowhere near the speed
of tesla or high-throughput gpu.
I appreciate any feedback about this,
From: ffmpeg-devel-bounces at mplayerhq.hu
[mailto:ffmpeg-devel-bounces at mplayerhq.hu] On Behalf Of Reimar D?ffinger
Sent: Thursday, January 31, 2008 3:06 PM
To: FFmpeg development discussions and patches
Subject: Re: [FFmpeg-devel] transcoding on nvidia tesla
On Thu, Jan 31, 2008 at 01:51:08PM -0800, Dan Khasis wrote:
> How would one go about porting or using the ffmpeg application on nvidias
I'd say firstly you'd have to learn to use/program with CUDA, then next
you'd have to find out how you call CUDA-functions from normal C
programs (the examples only show it the other way round, but that does
not seem like a feasible way to me for FFmpeg), and lastly you'd have to
find a place in FFmpeg that has enough data to process in one go to make
this efficient. Here probably the xvmc support is a good starting point,
everything beyond will probably require a few "architectural"
ffmpeg-devel mailing list
ffmpeg-devel at mplayerhq.hu
More information about the ffmpeg-devel