Any hard reset tools available? gpu is hung and remote desktop reboot fails

Accelerated Computing CUDA CUDA Programming and Performance
1

I am running the gpugrid.net application. The 9800gtx+ is hanging periodically and in debugging the problem I moved it from an AMD64 platform to an Intel64 platform. It is still hanging even after underclocking. The symptom is that after finishing a work unit, it periodically fails to iniitiaize. When this happens I need to turn the power off and on. A Windows (vista 64) restart does not work and leaves the GPU still hung.

Nvidia “performance” and “monitor” show no gpu and most objects are missing from their menus. PowerStrip show “-” for clock speeds and clicking “apply” for factory settings does not work as the clocks go back to “-” after the apply.

Anyway, I have a remote desktop to the system and was wondering if there is some way or tool I can use to cause a hard reset to the GPU. I have downloaded the toolkit, SDK, driver 178.08 (i have 178.24 maybe should use 08??) I am a Visual Studio 2008 type developer and not into CUDA yet, but possibly there is some tool I can use to issue a reset?

Both mombos had onboard video, ATI XL and Intel Pineview. The graphics board (SVGA) display works fine, but the CUDA application is hung. I have no games to try out but dxdiag show no acceleration present, but otherwise no problem (no errors).

…TIA…

2

I’ve seen similar behavior in linux, but only from buggy CUDA programs that randomly write to unallocated memory regions (hard power down/boot after every test makes for fun debugging :) ). I attributed the problem to some memory needed by the driver being overwritten by the CUDA prog: memory that is apparently not cleared/initialized on a soft boot and can only be fixed by a hard power down.

I’m not aware of any tools to perform such a hard reset without powering down. I haven’t seen this behavior since I fixed the bugs in my code that led to the out of bounds memory writes (at least 8 months ago).