中文版本

katago-benchmark

nnEvals/s: the bigger the better

1.9.1 cuda b40-s985(cuda11.0)

./katago benchmark -config ../default_gtp.cfg  -model ../../weights/kata1-b40c256-s9854456576-d2405111631.bin.gz -v 5000 -t 32,48,64,80,96,112,128

	A100PCIE	A30	GTX1080TI	RTX2080TI	RTX3080	RTX3090	V100	V100S	A40	A5000	3060	3060TI
nnEvals/s	3558	2212	425	1225	1376	1607	1784	1773	2473	1548	635	840

1.10.0 tensorrt b40-s1080(cuda 11.0 tensorrt 8.0.0.3)

./katago benchmark -config ../test.cfg  -model ../kata1-b40c256-s10800760064-d2633359377.bin.gz -v 5000 -t 32,48,64,80,96,112,128

	A100PCIE	A30	GTX1080TI	RTX2080TI	RTX3080	RTX3090	V100	V100S	A40	A5000	3060	3060TI
nnEvals/s	3986	2533	285	1975	2296	2590	2084	2070	2920	2323	1097	1457

1.10.0 opencl b40-s1080

./katago benchmark -config ../test.cfg  -model ../kata1-b40c256-s10800760064-d2633359377.bin.gz -v 5000 -t 32,48,64,80,96,112,128

	A100PCIE	GTX1080TI	RTX2080TI	RTX3090	V100	V100S	Mac Air M1	A40	A5000
nnEvals/s	2230	315	1086	1959	1509	1634	93	1788	1500