# Accelerating Diffusers and xDiT Image Generation with MXFP4 using AMD Quark on AMD Instinct™ MI350 GPUs

**URL:** <https://devcommunity.amd.com/t/accelerating-diffusers-and-xdit-image-generation-with-mxfp4-using-amd-quark-on-amd-instinct-mi350-gpus/709>\
**Category:** Image Gen / Editing\
**Tags:** ai, performance, diffusion-model\
**Created:** [July 14, 2026, 11:49am UTC](https://devcommunity.amd.com/t/accelerating-diffusers-and-xdit-image-generation-with-mxfp4-using-amd-quark-on-amd-instinct-mi350-gpus/709 "2026-07-14T11:49:05Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![nirth\_AMD](https://dub1.discourse-cdn.com/amd/user_avatar/devcommunity.amd.com/nirth_amd/32/623_2.png) [@nirth\_AMD](https://devcommunity.amd.com/u/nirth_AMD)\
**Post date:** [July 14, 2026, 11:49am UTC](https://devcommunity.amd.com/t/accelerating-diffusers-and-xdit-image-generation-with-mxfp4-using-amd-quark-on-amd-instinct-mi350-gpus/709/1 "2026-07-14T11:49:06Z")

</div>

![artificial-intelligence-quark-xdit-images-quark-xdit-thumbnail](https://global.discourse-cdn.com/amd/original/1X/e6c9c388b81c0858540367674a74a79d6904c4a2.webp)

Diffusion models such as Black Forest Labs’ FLUX.1-dev [[1]](https://rocm.blogs.amd.com/artificial-intelligence/quark-xdit/README.html#references) deliver stunning image quality but demand significant compute and memory bandwidth at inference time. To reduce inference cost without sacrificing image quality, precision-aware quantization techniques have become a critical optimization strategy.

In this blog, we demonstrate how **AMD Quark** [[2]](https://rocm.blogs.amd.com/artificial-intelligence/quark-xdit/README.html#references), a high-performance quantization library optimized for AMD Instinct™ MI350 GPUs, enables **MXFP4** quantization for Diffusers [[5]](https://rocm.blogs.amd.com/artificial-intelligence/quark-xdit/README.html#references) and xDiT [[3]](https://rocm.blogs.amd.com/artificial-intelligence/quark-xdit/README.html#references) FLUX.1-dev image generation. Starting from BF16 eager and `torch.compile` baselines, we apply Quark native inference with AITER [[4]](https://rocm.blogs.amd.com/artificial-intelligence/quark-xdit/README.html#references) GEMM kernels and show that MXFP4 ASM with `torch.compile` reaches **1.92× speedup** over BF16 eager and **1.41× speedup** over BF16 `torch.compile`, while preserving CLIP quality.

## Read more: [Accelerating Diffusers and xDiT Image Generation with MXFP4 using AMD Quark on AMD Instinct™ MI350 GPUs — ROCm Blogs](https://rocm.blogs.amd.com/artificial-intelligence/quark-xdit/README.html)