I just set up CUDA on my laptop. The graphics card is Quadro NVS 140M. I’m using the nVidia driver 180.29 directly downloaded from nVidia website. I successfully passed the deviceQuery test in the SDK. However, when I ran the bandwidthTest, the program gave me the following message:
Running on…
device 0:Quadro NVS 140M
Quick Mode
Host to Device Bandwidth for Pageable memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 1101.9
Quick Mode
Device to Host Bandwidth for Pageable memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 1175.2
Quick Mode
Device to Device Bandwidth
cudaSafeCall() Runtime API error in file <bandwidthTest.cu>, line 725 : out of memory.
I am wondering whether it is because of a bug in the device or other reasons? Informations are grealty appreciated!
I met the same issue on my thinkpad T61 with graphics card Quadro NVS 140M too. I an running ubuntu 8.10, with the nVidia driver 180.22, and tookit/sdk for ubuntu 8.04. I am able to build them with gcc 4.2 on ubuntu 8.10.
when I ran the bandwidthTest, it gave me:
Running on…
device 0:Quadro NVS 140M
Quick Mode
Host to Device Bandwidth for Pageable memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 1085.4
Quick Mode
Device to Host Bandwidth for Pageable memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 1068.7
Quick Mode
Device to Device Bandwidth
cudaSafeCall() Runtime API error in file <bandwidthTest.cu>, line 725 : out of memory.
In fact, several demo report this ‘out of memory’ error in my environment.
I would like to add that I also have the same problem with the software available in sdk. I am running ubuntu 9.04 x86_64 (development branch) on my 8400M GS. I also use gcc 4.2 to compile them. Today I switched from 180.35 to 180.29 and, strangely enough, codes that used to run out of memory now run a bit longer before they run out of memory.
128MB is probably not enough memory. CUDA contexts take a meaningful amount of memory (to enforce programming semantics), so you probably have 80MB or so total free memory at best.