Has anyone tried to use a single texture reference across multiple CPU threads accessing the same GPU? I have some global data, that I want to access from multiple CPU threads. I’ve tried many variations to get this to work, but all fail with errors like: kernel launch: unspecified driver error. If I reduce the problem to a single thread, calling the same function, everything works. I have not tried using the low level API, only the high level API. I’ve tried calling cudaBindTexture before calling the threads and tried within the threads without success. If I use two GPUs with a single thread for each, then everything works fine. It’s only when I try to have multiple threads per GPU that I run into problems. If I pass the global device memory pointer to each thread, I can access the data from multiple CPU threads accessing the same GPU. It is only when I try using texture memory that I run into problems.