Founded by machine learning expert Sharon Zhou and former Nvidia CUDA software architect Greg ... peak teraflops per second for generic matrix-matrix multiplication (GEMM) and up to 70 percent ...