rdna3

Zig INferenCe Engine — Local LLM inference on AMD GPUs and Apple Silicon

Windows-only version of ComfyUI which uses AMD's official ROCm and PyTorch libraries to get better performance with AMD GPUs. [auto-installation and popular performance enhancing packages like triton * sage-attention * flash-attention * bitsandbytes included ]

The intelligent OptiScaler installer Linux gamers needed. Automates FSR4, XeSS & DLSS configuration with GPU-optimized profiles for RDNA3/4, Arc & RTX cards.

Utilities for accessing AMD's Machine-Readable GPU ISA Specifications.

Installation script for an AI applications using ROCm on Linux.

TRELLIS (Microsoft's Image-to-3D generator) running on AMD GPUs with ROCm. Includes Gaussian splatting, mesh extraction, and GLB export. Tested on RX 7800 XT.

A turnkey, fully-local AI workstation engineered for the AMD Ryzen AI Max+ 395. LLM inference, voice, document parsing, browser automation, agents — all on-device.

Unlock fast, local LLM inference on AMD-powered mini PCs delivering 65-87 t/s for large models without cloud or subscription costs

A honest port of vLLM-ROCm for windows.

PyTorch built from source for AMD RDNA 3.5 (gfx1150) — Radeon 890M/880M GPU acceleration

Multi-GPU tensor/context parallel diffusion on AMD ROCm — with the patch that makes it actually work.

FlashAttention-2 Triton kernel for AMD RDNA3 (gfx1100 / Radeon RX 7900 XT/XTX/GRE): forward, backward, and split-K decode

Working ROCm 7.2.x + PyTorch environment for RDNA3. Built after fighting pipeline rats and wondering, “Lisa Su, girl… what are they doing down there.”

Docker infrastructure for AMD Strix Halo (RDNA 3.5 / gfx1151): PyTorch + ROCm base container and a separate Ollama LLM service. Two folders, two Compose files, one Strix Halo box.

FLUX.1-dev on AMD Radeon consumer GPUs — fast, low-VRAM, and shippable. Backport patches + benchmarks for torchao + diffusers group_offload on ROCm.

Evaluation-backed AMD ROCm port of HunyuanOCR-1.5 with reproducible OmniDocBench v1.6 benchmarks across llama.cpp, vLLM, and Transformers on gfx1100.

Reproducible RDNA reference for Meta Muse-Glimmer-30B — adapting MI-series ROCm recipes to Ryzen AI and Radeon with vLLM, llama.cpp, DFlash and auditable benchmarks.

Multi-GPU vLLM on consumer Radeon (RX 7900 XT/XTX, gfx1100): root cause and fix for the RCCL "operation cannot be performed in the present state" crash, plus 292 benchmark measurements across five model architectures

Local LLM benchmarks on AMD Strix Halo — 26+ models tested across RADV, AMDVLK, and ROCm with llama.cpp

A consolidated RDNA3/Zen3 reference for architecture, ISA, and ML training on consumer Radeon hardware (RX 7700 XT).