Maybe It is because different numbers of SM on 970, but I hardcoded BLOCKS and THREADS number, due to silentarmy algo worksize=NR_ROWS, so in cuda blocks=NR_ROWS/THREAD_PER_BLOCK. I don't have 970 cards, but a have 980, will take a look tomorrow.

UPD. It seems you are running with bad options, because launching cuda_tromp_STUB
Start options are buggy, only works with


UPD2.

Tpruvot added my port to his repo. I think, he fixed launch bugs.

https://github.com/tpruvot/nheqminer/tree/cuda-silentarmy