tobyc11 / ffmabench

CUDA benchmark that shows your GPU's TFLOPS

ffmabench

CUDA benchmark that shows your GPU's TFLOPS

Example Output

This is on a GeForce RTX 2060 SUPER. It is 7.2 TFLOPs on paper.

Running 16000 ops/thread * 1024 threads/block * 1048576 blocks
Launch latency: 0.000040 s
Kernel duration: 2.100975 s
Your GPU's TFLOPS is 8.177092

And this is on a Quadro GP100. 10.3 TFLOPs on paper.

Running 16000 ops/thread * 1024 threads/block * 1048576 blocks
Launch latency: 0.000051 s
Kernel duration: 1.556323 s
Your GPU's TFLOPS is 11.038755

MacBook Air M1

I ran a similar benchmark using Metal on the M1 air. For whatever reason, I got nowhere near the TFLOPS metric on paper (2.6 TFLOPS).

Benchmarking pixel shader
Finished in 1.661727 s
  0.807700 TFLOPS

Benchmarking compute shader
Finished in 2.701267 s
  0.794991 TFLOPS

ffmabench's Issues

Does it run in Ubuntu with multi 4090 ?