CUDA and hardware reboot execution problem leading to reboot


I’ve been around with a curious and annoying bug for a couple of weeks now. As done in the MatrixMul example, I put the code in my kernels between:

for(tx = threadIdx.x; tx < N; tx += NB_THREAD){

//calculus here


and launch them that way:


dim3 GRID(N / NB_THREAD, 1);

myker <<< GRID, BLOCK>>> (myargs);

and despite that, if the number of cells N of my calculus is not a multiple of the number of threads in one block, the screen freezes for a few seconds and then my computer reboot…

Anyone who got the same problem and find a solution ?

Could be an out-of-bounds memory access. Have you tried running in emulation mode?


As silly as it may sound, no. :">

I’m gonna try and report what’s happening…