Getting cudaErrorLaunchFailure with all Kernel calls disabled.

Gaming and Visualization Technologies APIs OpenGL
1

Hello,

I’m working under the idea that all CUDA kernel calls are denoted by the

function <<<x,y>>>

syntax. With that idea let me try to explain my issue. I am going to leave a lot of detail out because my project is large so my question is general, when using C++/C CUDA OpenSceneGraph (OpenGL) what is considered a kernel call? Could something other than CUDA cause an issue in the kernel resulting in this cuda error?

Thanks,
Zach

For anyone that wants more detail, let me show you my specific case… I’m running a 1080, 64bit Windows 10 on latest drivers and using Cuda 10.1 in VS 2017. In my project, I’ve currently turned off all my kernel calls (denoted by, <<<>>>) but I am still randomly getting the cudaErrorLaunchFailure on the following line of code…

checkCudaErrors(cudaDeviceSynchronize());
checkCudaErrors(cudaMemcpy((char *) device + offset, host, size, cudaMemcpyHostToDevice));
checkCudaErrors(cudaDeviceSynchronize()); // <-cudaErrorLaunchFailure, always this line of code.

My understanding is that an error happens in a kernel call and then it’s caught later down the line like it does above. Now I’ve disabled my

<<<>>>

calls but am still running OpenGL shaders, still running other CUDA memory manipulations so I’m just at a loss at what could be causing the launch failure.

Any help is appreciated.