OpenCL pitfalls with Windows SYSTEM user

I’m a C programmer, but I’m not an OpenCL developer, and so I don’t have a knowledge of GPU architecture or the OpenCL system calls, but I’m hoping these won’t be requisites for resolving this issue.

The dnetc opencl client successfully locates OpenCL compatible processors (Intel graphics card) and successfully executes a kernel when the client is running as a Windows user or administrator, but when run as a service as SYSTEM the client shutsdown at the stage when a “cruncher” - which I assume is a thread which executes an OpenCL kernel - should start.

This problem also exists with the stream client, but I have successfully run the CUDA client as a service, so I’m wondering whether SYSTEM doesn’t have permission too use a system call, which is differently implemented in the CUDA version.

Can anyone recommend a method for debugging the issue?
Or if anyone has overcome this issue in the past, please would you advise?

Many thanks

Only thing I can think of, replace clMapBuffer with clWrite\ReadBuffer. This could be WDDM related.