If I recompile my code with the nvcc -arch sm_11 flag, I get a runtime error saying “invalid device symbol.” when trying to use cudaMemcpyToSymbol. Everything works normally if I don’t use the arch sm_11 flag.
I just installed the 2.0 toolkit and SDK, and I am still getting this error. Has anyone been able to use atomic operations (sm_11) with cudaMemcpyToSymbol? Thanks
I think this was caused by running the code compiled for a compute 1.1 architecture on a 1.0 card.