Strange results with pinned memory Pinned memory stopped "working"

Accelerated Computing CUDA CUDA Programming and Performance
1

Greetings,

Without apparent reason using pinned memory (allocated through MallocHost)
stopped giving me the significant benefits that I had experienced before.
Has anyone had a similar experience? If so is there a “fix” for the problem?

I am using a 8800GTS on a PCIEx16 slot. I am running Windows XP professional.

I have included the bandwidthtest results with and without “pinned” below.
I would greatly appreciate any suggestions.

Best regards,

Carlos

------------ PINNED -------------

c:\tools\cuda_sdk\bin\win32\Release>bandwidthTest.exe -memory=pinned
Quick Mode
Host to Device Bandwidth for Pinned memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 1318.8

Quick Mode
Device to Host Bandwidth for Pinned memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 1520.9

Quick Mode
Device to Device Bandwidth
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 45565.7

&&&& Test PASSED

Press ENTER to exit…

------------ PAGEABLE -------------

c:\tools\cuda_sdk\bin\win32\Release>bandwidthTest.exe -memory=pageable
Quick Mode
Host to Device Bandwidth for Pageable memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 1289.3

Quick Mode
Device to Host Bandwidth for Pageable memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 1407.9

Quick Mode
Device to Device Bandwidth
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 46865.9

&&&& Test PASSED

Press ENTER to exit…

2

I think something else must have changed on your system – I don’t think the bandwidth could just change without a hardware or software (or driver) change.

Mark

3

Hi Mark,

Thank you for the reply. You are correct. I added a second card, a FX550 on another PCIx16 slot on the motherboard. I was trying to use the FX550 to drive my monitors while using the 8800GTS for CUDA. I removed the FX550 and I now see some more improvement from pinning. Is there a way to programmaticaly check if the PCIx16 bus is being shared? Will I see the same problem in a SLI setup?

Thanks,

Carlos