DaveT wrote:

OK. I've now got a card with an Nvidia GeFoce 8400 engine and 512 MB of vDram. I have the latest
supposedly CUDA-enabled version 6.4.7 of BOINC. The start-up log says "CUDA devices found". How do I
determine if it's actually using this capability? So far it's processing Astropulse at the rate of
about 0.1% per 10 minutes. That sounds like about 166.66 hours to completion, versus an estimate of
194 hours, which is about what it used to take. Is this roughly 16% improvement the best I can hope
for? That would be about 7 days. By the way, for what it's woth, it rates my 2 CPUs at 2594 floating
point MIPS and 4146 integer MIPS each.
Thanks
Dave

The seti@home website is down right now so I can't confirm this, but I think that the stock version of BOINC will only do regular workunits on the Nvidia GPU, not Astropulse.

Also, from what I've read, on a multi-core system one core will not run a workunit but will be "feeding" the CUDA GPU. Probably the same on a single core system; the CPU will be feeding the GPU.

I'm still in the planning stages trying to figure out what I'll need to run CUDA units. So far it looks like a major upgrade.



--   
</Tazz>