Forum > GPU crunching
Latest drivers (NVidia and ATI)
MarkJ:
Anything useful in cuda 4.1 for us? Does the optimizing compiler make any difference to earlier ones?
Raistmer:
--- Quote from: MarkJ on 30 Jan 2012, 06:31:34 am ---Anything useful in cuda 4.1 for us? Does the optimizing compiler make any difference to earlier ones?
--- End quote ---
I saw OpenCL 1.1 support on FERMI (not sure it was added only now though).
OpenCL part of distro became wider, separate best practice guide for OpenCL... need to read for more ;)
Jason G:
--- Quote from: MarkJ on 30 Jan 2012, 06:31:34 am ---Anything useful in cuda 4.1 for us? Does the optimizing compiler make any difference to earlier ones?
--- End quote ---
Decent performance gain on Fermis. Older GPUs are seeing a slowdown over Cuda 3.2, which has been reported & apparently reproduced by nVidia. I was hoping they'd fix that by release, as it means will probably have to support 2 builds next installer.
Jason
msattler:
The kitties were rather hoping for another driver release beyond 290.53 to go along with 4.1 too. Hope they haven't given up on it.
Jason G:
--- Quote from: msattler on 30 Jan 2012, 12:56:53 pm ---The kitties were rather hoping for another driver release beyond 290.53 to go along with 4.1 too. Hope they haven't given up on it.
--- End quote ---
I did spot leaked GK107 drivers (295 point something) floating around somewhere, which I think are some mobile chip. experience on the new compiler was pretty good on Fermi's, so I suspect they'll want to iron out any reported issues ( perhaps including ours) before the big release. Just guessing, maybe they've realised they're more or less ready software wise and have all the focus on prepping for a solid launch. They've been pretty quiet about the new architecture, I suppose playing their cards close to their chests. Plenty of rumours about but nothing solid. Certain notorious anti-nvidia guys are suggesting nVidia might well 'win this round', which I presume referred to gaming performance. If that's the case though, it implies a few technical things that should apply to Cuda as well...
Jason
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version