gtx 1660 ti cutlass 1455mhz:

cutlass_simt_sgemm_128x128_8x2_nt_align1: 4.3 tflops

cutlass_simt_dgemm_128x128_8x2_tt_align1: 0.14 tflops

cutlass_simt_hgemm_128x256_8x2_nt_align1: 7.8 tflops

cutlass_simt_igemm_s8_128x128_32x2_nn_align1: 16.0 tops

cutlass_simt_s8_igemm_s8_128x128_32x2_tt_align1: 15.9 tops

(cublas is garbage)

8k^3