x264 and Stream Processors

xc3ll · 18th April 2007, 20:54

I was wondering if there was any work being done on x264 to get it working with nvidia's CUDA technology. It would be great to get that working. If there isn't any current development, or it's still in the beta stages, I'd like to participate. I can do testing(have an 8800gts), as well as write some code(I don't know how x264 works--I should take a peak at the source--but I can write code pretty well).

was going to post this in the development thread, but its locked

burfadel · 18th April 2007, 21:01

I'm sure your input would be very welcome! just ask akupenguin on here, he's the x264 author.

akupenguin · 18th April 2007, 22:10

I don't have a good idea of the latency and bandwidth avilable for communicating between cpu and gpu, and without that I can't properly evaluate the chances of different methods. But the most likely so far is: Offload motion search (me.c and half of analyse.c) to the gpu. Run intra search on the cpu at the same time. Then merge those results and do the rest of the encoding of the macroblock in pure cpu.

18th April 2007, 20:54	#1 \| Link
xc3ll Registered User Join Date: Feb 2007 Posts: 32	x264 and Stream Processors I was wondering if there was any work being done on x264 to get it working with nvidia's CUDA technology. It would be great to get that working. If there isn't any current development, or it's still in the beta stages, I'd like to participate. I can do testing(have an 8800gts), as well as write some code(I don't know how x264 works--I should take a peak at the source--but I can write code pretty well). was going to post this in the development thread, but its locked

18th April 2007, 21:01	#2 \| Link
burfadel Registered User Join Date: Aug 2006 Posts: 2,229	I'm sure your input would be very welcome! just ask akupenguin on here, he's the x264 author.

18th April 2007, 22:10	#3 \| Link
akupenguin x264 developer Join Date: Sep 2004 Posts: 2,392	I don't have a good idea of the latency and bandwidth avilable for communicating between cpu and gpu, and without that I can't properly evaluate the chances of different methods. But the most likely so far is: Offload motion search (me.c and half of analyse.c) to the gpu. Run intra search on the cpu at the same time. Then merge those results and do the rest of the encoding of the macroblock in pure cpu.