8600GTS 2.1 bandwidthTest failure

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi all,
Wondering if anyone can give me some pointers. I just installed CUDA 2.1 on 64-bit Ubuntu on my desktop at home using an 8600GTS. I was able to successfully build the SDK, and most of the demos run without problem. However, bandwidthTest fails with a Runtime API error, “out of memory”, usually very quickly. I have tried with --mode=pinned, and also changing the start and end and increment sizes, all to no avail. Certain combinations run for slightly longer, but all crash/fail with the above error message. Is there an issue with this card? It reports 256MB of memory, so the test should not be failing at 100KB…

Thank you

2

A little more information. I was able to run a test program to allocate powers of 2 up to 16MB, then it fails attempting 32MB. On a card with 256MB, what’s going on? I assume some must be allocated for the display ( this card is also running my main desktop), but how to find out how much? I cant’ seem to find this in the programming guide…any pointers?
Thanks!