Tools for GPU Microbenchmarking

From the site:
Instruction Rate

These are simple instruction throughput tests, so basically your "raw TFLOPS" measurements. Just keep in mind all of them are being measured in operations per second, not FLOPS, which for most of them means nothing special, but for Multiply-Add you will have to multiply the result by 2 if you want "marketable TFLOPS".
 
3070 numbers
 

Attachments

  • 3070 cache memory bandwidth.png
    3070 cache memory bandwidth.png
    28.8 KB · Views: 13
  • 3070 instruction rate.png
    3070 instruction rate.png
    10.2 KB · Views: 13
  • 3070 vector cache + mem latency.png
    3070 vector cache + mem latency.png
    27.5 KB · Views: 11
1060 3GB numbers
 

Attachments

  • 1060 3gb cache memory bandwidth.png
    1060 3gb cache memory bandwidth.png
    32.9 KB · Views: 9
  • 1060 3gb cache memory latency.png
    1060 3gb cache memory latency.png
    28.6 KB · Views: 9
  • 1060 instruction rate.png
    1060 instruction rate.png
    11.8 KB · Views: 12
Back
Top