> Estimated Credit:
>
> If there are two GPUs with very different characteristics, but both of the
> same type, would they be counted separately?  It is my belief that they
> should be as it is entirely possible that there is a large speed factor
> between 2 GPUs installed in the same system.
>
> jm7

It isn't just the speed factor which is significant here. There's also the 
single precision / double precision question, which means that some projects 
can run on some, but not all, GPUs of ostensibly the same resource class (as 
currently defined), and the ongoing issues caused by varying 'compute 
capability' within the NVidia range.

This latter point is still leading to discussion at GPUGrid - and they 
should know a thing or two on this subject - about issuing different 
app_version tasks to multiple disparate CUDA cards within a single host. As 
I understand it, this would not be possible whilst all GPU devices (from the 
same manufacturing stable) are treated as being in the same resource 
characteristic - another potential reason for counting them separately.

(ref: http://www.gpugrid.net/forum_thread.php?id=2323) 


_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.

Reply via email to