I made this change.
-- David
On 24-Feb-2014 5:33 PM, Richard Haselgrove wrote:
With the release of the GTX 750/750Ti 'Maxwell' GPUs utilising the GM107 chip,
could somebody please update /lib/coproc.cpp so that the peak_flops value for
this architecture is calculated correctly?
Referring to
http://international.download.nvidia.com/geforce-com/international/pdfs/GeForce-GTX-750-Ti-Whitepaper.pdf
"Each SM is now partitioned into four separate processing blocks, each with its
own instruction buffer, scheduler and 32 CUDA cores."
So that's 128 cores per SM, and the cards being seen at SETI report their
Compute
Capability as 5.0
The current default handler at
486 default: 487 flops_per_clock = 2; 488
cores_per_proc = 192; 489 break; 490 }
overstates peak_flops by 50%
Thanks. _______________________________________________ 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.
_______________________________________________
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.
