"Unspecified Launch Failure" Error CUDA and GLSL issues

Hi, All —

I’m getting the following error back from CUDA at run time in a program that uses both CUDA and GLSL to first process and then render particle system data.

[font=“Courier”]“Cuda error in file ‘cudakernel.cu’ - line 214 - Unspecified Launch Failure”[/font]

cudakernel.cu is the module that does the setup and execution of the kernel, but does not contain the kernel proper. Line 214 is the second CUDA_SAFE_CALL() instance below:

[font=“Courier”]extern “C” void
RunCUDA(float Time)
float *posPtr;
float dms;

...{code here}...

// map OpenGL buffer object for writing from CUDA
CUDA_SAFE_CALL(cudaGLMapBufferObject( (void**)&posPtr, cuda_vboPos));

// execute the kernel
dim3 block(CUDA_BLOCK_SIZE, 1, 1);
dim3 grid(numParticles / block.x, 1, 1);
kernel<<< grid, block >>>((float4 *)posPtr, (float4 *)dev_velData, (const float4 *)dev_rndData, Time );

// unmap buffer object
CUDA_SAFE_CALL(cudaGLUnmapBufferObject( cuda_vboPos ));        // <=== UNSPECIFIED LAUNCH FAILURE error on this line


The curious thing is, the error only occurs when I change the value of a #DEFINE constant in the kernel itself from 0.00001 to 0.00001f. I can post the kernel if anyone wants to see it, but it’s a might lengthy.

I am reasonably sure that the error does not happen without the ‘f’, and even with it it only happens about once every minute (~50 frames at 60 fps), after which point the program crashes and Windows freaks out - little flakes of “fuzz” appear all over any redrawn windows and only a reboot fixes it. After raising the error, sometimes the system BSODs with the following code:

[font=“Courier”]The computer has rebooted from a bugcheck. The bugcheck was: 0x100000ea (0x89c8a7a8, 0x8a025d58, 0xba4fbcbc, 0x00000001).
BSOD code 0x100000ea is associated with THREAD_STUCK_IN_DEVICE_DRIVER_M, which basically means the GPU is in an infinite loop (according to this Microsoft KB article)

I’ve tried two different systems, both with 8600’s and CUDA 1.1…no luck

Any advice?