cudaSafeCall initialization error on Win XP 64 bit with 64bit SDK

I am not having a lot of success with CUDA so far unfortunately.

My setup is as follows:

AMD Athlon X2
Geforce 8600GT
Windows XP 64bit

As suggested by the website I installed the following:

181.20_geforce_winxp_64bit_english_whql
CudaSetup-2.1-win64
NVIDIA_SDK10_CUDA_x64_2.10.1215.2015

My graphics card driver was reported correctly by my system device manager and all packages appeared to install without a glitch so everything looked quite promising.

After some playing around to make sure the VS8 build configurations correctly used x64 platform I was able to build all the samples (except for 5 that needed D3DX9 - DirectX9 3D? I imagine I need Direct X toolkit but will leave that for now)

However when I try to run deviceQuery I get:

cudaSafeCall() Runtime API error in file <deviceQuery.cu>, line 59 : initialization error.

(btw, there looks to be similar post to this except in that case the person was using 32bit SDK)

A similar error message is produced for all the samples that did build.

I noticed that a new version of the drivers was available:

182.06_geforce_winxp_64bit_english_whql

so I installed these and again the device manager seemed to be reporting a healthy Geforce 8600GT card. Unfortunately the error is the same and I am more or less at a dead end. I found various references to posting a bug report but could not find out how to do so. If my solution is not obvious to people then I am happy to post a bug report if someone provides a summary of how to generate it.

Also I did run depends.exe on the deviceQuery.exe binary and didn’t see anything out of the ordinary, in particular no dependency on 32bit shared libraries exists.

Thanks in advance.

For those interested my problems were down to two things:

  1. not completely removing my virus checker from the machine before installing the CUDA components

  2. trying to run the samples on a server machine (with the GPU installed) via remote desktop - it appears that the windows/nvidia driver treatment of attached displays is a bit weird and it fails to think it has anything attached when using remote desktop - I would never of thought of this and stumbled onto it by accident