This is a custom drop-in CUDA kernel designed to bring older Pascal GPUs back to life when running heavy FP8 models in ForgeUI.Since the Pascal architecture lacks Tensor Cores, PyTorch defaults to a painfully slow fallback path when handling FP8 weights. This mod intercepts that process, converting FP8 to INT8 on the fly and computing it using native <code>__dp4a</code> instructions.The Result: Roughly 2x faster generation speed with zero visual quality loss.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/a0164f55-f2b2-44cf-a56b-b87374fed4c5/original=true/a0164f55-f2b2-44cf-a56b-b87374fed4c5.jpeg" />And Pytorch profiler : FP8 Stock <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/2b60aef7-811f-4233-b5cc-6eedb9575327/original=true/2b60aef7-811f-4233-b5cc-6eedb9575327.jpeg" /> INT8 <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/b88810dc-5e4d-4f34-af31-9219a4f34d8c/original=true/b88810dc-5e4d-4f34-af31-9219a4f34d8c.jpeg" /><h3 id="tested-setup"> Tested Setup</h3><ul><li>Hardware: Titan X Pascal (<code>sm_61</code>)</li><li>Models tested: Z Image Turbo (ZiT) and Flux 2 Klein 9B (both in FP8)</li><li>UI: ForgeUI (Neo branch). It might work on the original main branch if it supports these models, but Neo is fully tested.</li></ul><h3 id="important-warning"> IMPORTANT WARNING</h3>Do NOT use this mod if you are on a Turing, Ampere, Ada, or newer GPU (RTX 20xx / 30xx / 40xx). Starting with Turing, NVIDIA introduced Tensor Cores, meaning your native hardware path is already significantly faster than this implementation. This kernel is strictly a rescue patch for Pascal (sm_61) architecture!<h3 id="installation-instructions"> Installation Instructions</h3>No compilation required. Just follow these steps:<ol><li>Go to your ForgeUI <code>backend</code> folder.</li><li>Find <code>operations.py</code> and make a backup of it (just in case you want to revert later).</li><li>Replace the original <code>operations.py</code> with my modified version.</li><li>Inside the <code>backend</code> folder, create a new folder named <code>ext</code>.</li><li>Inside <code>ext</code>, create another folder named <code>zimage_ext</code> (The path should look like this: <code>backend/ext/zimage_ext/</code>).</li><li>Drop the provided library file (<code>.pyd</code>) into the <code>zimage_ext</code> folder.</li></ol>That's it! Restart your ForgeUI. Any FP8 models will now automatically convert and compute in INT8, giving your 10-series card a massive speed boost.

Files with SHA256: 0019aee16223dfac13d621775bfdb4e27539b1d312ef0516a3b507af5227e6b7