GPU Benchmark Flux DEV fp8 5090 4090 3090

[metamountain]

Lets do a benchmark for flux dev fp8.

Just use the workflow template. Post the time of second or third render , no loading ....

[metamountain]

Nvidia 3090: 26s

[simonlui]

Intel A770 on Fedora Linux. Using Pytorch Nightly.

100%|███████████████████████████████████████████████████████████████████████| 20/20 [00:52<00:00,  2.61s/it]
Prompt executed in 53.26 seconds

If I go back to Pytorch Extension for Pytorch v2.3.110+xpu

100%|███████████████████████████████████████████████████████████████████████| 20/20 [00:46<00:00,  2.33s/it]
Prompt executed in 47.13 seconds