nheqminer Only CUDA devices with SM 5.0 and higher are supported

I got this error "Only CUDA devices with SM 5.0 and higher are supported" when running the command with option for GPU. My GPU is Gigabyte GTX 650 Ti boost which has shader model 5.0 according to the specs at https://www.techpowerup.com/gpudb/b1694/gigabyte-gtx-650-ti-boost-windforce-2x-2-gb But when running the command with -ci option its output is "#0 GeForce GTX 650 Ti BOOST | SM version: 3.0 | SM count: 4" Is that strange or not ???

May 19 '17 21:05 anselal

I have the same issue (using a GTX 650 Ti, but non-boost). What I found out is that naming it "SM 5.0" is misleading for the user, because in fact the program is comparing the Compute capability of the card.

The line which contains std::runtime_error("Only CUDA devices with SM 5.0 and higher are supported."); is not stating an accurate error, therefore leading to the confusion.

The source code has calls to cudaGetDeviceProperties() to compare which returns the cudaDeviceProp structure. Source files cuda_djezo.cpp and cuda_tromp.cpp have it.

According to CUDA Toolkit docs, cudaDeviceProp.major returns the major version of the card's Compute capability, which is distinct from Shader Model support. And according to this official CUDA list, our GTX 650 Ti's have Compute 3.0 capability, so yes it is below 5.0, but not in the Shader Model! Looks like Compute 5.0 is good for Maxwell and up.

So, either the code must be updated to support Compute 3.0 or change the wording of the command line output so that it is clear that it's checking Compute capability and not Shader Model support.