[ Date Index ] [ Thread Index ] [ <= Previous by date / thread ] [ Next by date / thread => ]
|
On 14/02/18 10:49, Andy Wills wrote:
Hi Tom, I think it probably depends a bit on the specific applications you're using -- it's possible that there is some demand on the CPU and I/O also, even if the GPU is the workhorse? Very interested to hear how you get on either way - I just built a machine with a GTX 1060 and CUDA-based stuff was one of the things I was planning to try out when I got a chance :-) All the best Andy On Tue, Feb 13, 2018 at 10:37:11AM +0000, Tom via list wrote:I've got another book on Neural Networks and it has some detailed CUDA stuff in it. I've got an old Geforce GT720 which has a performance of 240 or so compared with 10,000 for higher end cards. So I thought a low level desktop that can contain a high end card might be more useful for playing with this stuff. Anyone got any thoughts on that? Tom te tom te tom -- The Mailing List for the Devon & Cornwall LUG https://mailman.dclug.org.uk/listinfo/list FAQ: http://www.dcglug.org.uk/listfaq Most of the data crunching using GPUs is just almost completely GPU based - the io is small %s of activity so loading the problem onto the cpu is going to take seconds of runs of minutes from what I can gather. The host cpu is merely going to get data from the internet or disk and then feed it to the GPU and drum its fingers. Your GPU is around 4,375 GFlops. My screamingly fast 8 core machine is 100th that. I think all I need is a 2 core 64bit atom with some big IO on it! Tom te tom te tom |
-- The Mailing List for the Devon & Cornwall LUG https://mailman.dclug.org.uk/listinfo/list FAQ: http://www.dcglug.org.uk/listfaq