Basic question regarding bandwidthTest on Tesla C1060

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi all,

I have a tesla C 1060 and bandwidthTest reports the following:-

Quick Mode
Host to Device Bandwidth for Pageable memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 2252.4

Quick Mode
Device to Host Bandwidth for Pageable memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 2043.6

Quick Mode
Device to Device Bandwidth
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 73715.7

&&&& Test PASSED

Press ENTER to exit…

But, I see that the theoretical peak bandwidth of Tesla C 1060 is actually more than the reported figure - ~102 GBPs

(http://www.tigerdirect.ca/applications/SearchTools/item-details.asp?EdpNo=4259469&CatId=4044)

  1. Can someone tell me why I am not able to acheive full bandWidth using bandwidthTest?
  2. Is it possible for applications to achieve more (memory intensive) bandwidth than what bandWidthTest does? (I understand that bandwidthTest just does a devicetodevice copy, but just curious to know if it is possible)

Thanks!

2

No bus achieves 100% of the theoretical peak.

There’s another thread on the forum where higher than 70% of theoretical bandwidths are being reported (search for recent posts by alex_dubinsky). You can see what kind of tuning he’s doing.

Paulius