MixBench ~ GPU の精度別性能評価 ~

GitHubで公開されているベンチマークmixbench(https://github.com/ekondis/mixbench)を使用して半精度 (FP16), 単精度 (FP32), 倍精度 (FP64), 整数8bit (INT8) の精度別の GPU の性能評価を継続的に実施します。

mixbench/read-only	FP16	FP32	FP64	INT
	GFLOPS	GFLOPS	GFLOPS	GIOPS
NVIDIA RTX 6000 Ada	80,238.59	75,720.65	1,188.25	37,376.29
NVIDIA RTX 4090	79,668.17	73,406.04	1,085.85	34,803.28
NVIDIA RTX 4080	47,506.80	44,588.82	639.76	20,829.41
NVIDIA A100 40G	56,827.39	14,824.54	7,487.05	13,020.90
NVIDIA GeForce RTX 3090	40,761.49	38,698.40	543.73	17,984.60
NVIDIA RTX A6000	39,911.73	36,306.66	537.56	17,370.73
NVIDIA A40	36,388.94	34,306.09	489.73	15,979.07
NVIDIA TITAN RTX	34,767.29	12,163.79	455.63	15,560.75
NVIDIA RTX A5000	30,128.46	27,930.81	404.05	13,122.55
NVIDIA GeForce RTX 2080 Ti	30,831.05	15,424.95	413.69	14,479.27
NVIDIA GeForce RTX 2080 SUPER	24,395.80	12,179.50	324.75	11,455.82
NVIDIA RTX A4000	22,822.16	17,956.03	305.49	10,029.82
NVIDIA GeForce RTX 2080	21,736.53	10,878.22	289.60	10,215.54
NVIDIA Tesla V100 PCIe 32G	10,757.17	12,243.53	6,133.56	10,268.77
NVIDIA TITAN V	23,827.05	11,904.82	5,998.90	10,020.40
NVIDIA Tesla P100 PCIe 12G	17,585.75	9,083.63	4,606.66	2,747.68
NVIDIA GeForce GTX 1080TI	212.88	12,995.95	425.33	3,449.97
NVIDIA GeForce GTX 1070TI	144.88	8,902.09	289.74	2,136.32
NVIDIA GeForce GTX 1080	147.21	9,075.68	294.20	2,326.99
NVIDIA Tesla k40m	N/A	3,319.34	1,388.02	698.84
NVIDIA Tesla k20Xm	N/A	3,187.65	1,282.44	611.49
AMD Radeon Vega Frontier Edition	23,352.72	11,917.33	810.58	2,506.02