Benchmark results

[noctrex]

Hey, just found out about your project over here.
You said in your readme that you were interested in submitting benchmark results, so here's mine.
I'm running a rig with an AMD 7900XTX, So it does OOM. I just wanted to see what would happen.
Of course me running it on ROCm, it does not support torch.compile yet, so I ran without it.
I hope you will have the option in the future to also load the fp8 version of the model, so that it could run also in smaller cards.

####### Benchmark Report #######

Configuration:

{
"default_prompt": "Turn this into art.",
"default_steps": 2,
"default_seed": 52,
"models_path": "FLUX.2-klein-4B",
"resolution": {
"height": 320,
"width": 576
},
"compile_models": false,
"enable_spatial_cache": true,
"target_fps": null,
"interpolation_exp": 2,
"use_reference_image": false,
"logging": false
}

Hardware Information:

{
"platform": "Windows-11-10.0.26200-SP0",
"python": "3.12.10",
"cpu": "",
"cpu_cores_logical": 16,
"gpu": [
{
"name": "AMD Radeon RX 7900 XTX",
"vram_gb": 23.98,
"cc": "11.0"
}
]
}

Results:

---

Dynamic Area Processing Time (s) FPS

         0%                1.2440        3.22
        10%                1.3130        3.06
        25%                1.3623        2.95
        50%                1.4324        2.82
        75%                1.4365        2.82
        90%                1.5762        2.57
       100%                1.5699        2.62

---

End-to-end latency: 0.0000 seconds

End of Report

[noctrex]

I'm sorry about the duplicate issues created. My browser hanged, and it was submitted multiple times erroneously. Please delete the duplicate ones.

[tensorforger]

no worries, removed

[tensorforger]

Thanks a lot for this report. At least now I know that it works normally on windows too.
For now you can try to decrease resolution aggressively, e.g. to 256x256. I will try to reduce vram usage in next updates.

[saintbrodie]

After installing triton-windows:

####### Benchmark Report #######

Configuration:

{
"default_prompt": "Turn this into art.",
"default_steps": 2,
"default_seed": 52,
"models_path": "FLUX.2-klein-4B",
"resolution": {
"height": 320,
"width": 576
},
"compile_models": true,
"enable_spatial_cache": true,
"target_fps": null,
"interpolation_exp": 2,
"use_reference_image": false,
"logging": false
}

Hardware Information:

{
"platform": "Windows-11-10.0.26200-SP0",
"python": "3.12.9",
"cpu": "",
"cpu_cores_logical": 24,
"gpu": [
{
"name": "NVIDIA GeForce RTX 4090",
"vram_gb": 23.99,
"cc": "8.9"
}
]
}

Results:

---

Dynamic Area Processing Time (s) FPS

         0%                0.1450       27.79
        10%                0.1970       21.77
        25%                0.2069       19.39
        50%                0.2378       16.86
        75%                0.2746       14.59
        90%                0.2949       13.58
       100%                0.3069       13.06

---

End-to-end latency: 0.0010 seconds

End of Report

[tensorforger]

Thanks for sharing!