Segmentation fault on cudaMalloc even in SDK example

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

we have a PC with Tesla C2070 inside. It was working with no problems. Recently I started getting segmentation faults on cudaMalloc function. We have found, that all users are experiencing the same problem. I have updated the system from CUDA 3.2 to CUDA 4.0 and we are still getting the same message even on simple SDK examples like bandwidthTest. We are running 64-bit Ubuntu 10.04. Do you have any idea what to do? I hope the card is still ok.

Thanks, Tomas.

2

So I have solved it. We had to plug a monitor. The same think happened to me already with one mutli-gpu system so I thought that it was problem with detection of a primary card. Now I see it happens even on single gpu systems. It must be a problem of linux. Do you have any idea what to do with it? It is stupid to plug monitors to computers in a server room.

3

So I have solved it. We had to plug a monitor. The same think happened to me already with one mutli-gpu system so I thought that it was problem with detection of a primary card. Now I see it happens even on single gpu systems. It must be a problem of linux. Do you have any idea what to do with it? It is stupid to plug monitors to computers in a server room.

4

Are you using the script in the release notes to load the driver and create the /dev/nvidia entries?

5

Are you using the script in the release notes to load the driver and create the /dev/nvidia entries?

6

No, I am not:-). I think this might be the reason. I will try it.

Thanks a lot, Tomas.

7

No, I am not:-). I think this might be the reason. I will try it.

Thanks a lot, Tomas.