Gfx1151 rocm. Feb 20, 2026 · On Strix Halo GPUs (gfx1151) memory access is handled through GPU Virtual Memory (GPUVM), which provides per-process GPU virtual address spaces (VMIDs) rather than a separate, discrete VRAM pool. Sep 13, 2024 · ROCm (Radeon Open Compute) is an open-source platform for GPU computing, which provides powerful tools for developers. 0 and remains separate from the 7. Mar 14, 2025 · Given that AMD gave a ROCm presentation at the Framework Desktop launch event, which uses gfx1151 (8050S/8060S), I assume we will see official support for this ISA in a future release? 5 days ago · 28 RUN |2 PYTORCH_ROCM_ARCH=gfx90a;gfx942;gfx950;gfx1100;gfx1101;gfx1200;gfx1201;gfx1150;gfx1151 PYTHON_VERSION=3. x features and improvements for AMD Strix Halo (gfx1151) systems. This page covers the build architecture, tarball acquisition process, rocWMMA integration, and configuration differences between variants. For the latest production stream release, see the ROCm documentation. Maintaining parallel release streams – preview and production – gives users ample time to evaluate and adopt the new build system and dependency changes. llama. Each release is a self-contained archive containing a bundled Python environment, vLLM, PyTorch ROCm, and all required ROCm runtime libraries. , llama-windows-rocm-gfx1151-x64 for Radeon 8060S). 0 to 7. g. . Mar 26, 2026 · Important ROCm 7. Sep 16, 2025 · Some sources claim the gfx1151 has had support since Rocm 6, but on the announcements page of Rocm 7 it said: ROCm endpoint AI ecosystem supports Linux and Windows on AMD Radeon products including the latest Radeon RX 9000 series, as well as the class leading Ryzen AI MAX products. 1 and supported GGUF quantized models, which fit the unified memory ROCm ships AOTriton as its SDPA backend, but AOTriton does not support gfx1151 (Strix Halo, Framework Desktop). 5. When that happens, rebuild with -DGGML_HIP=ON -DGPU_TARGETS=gfx1151 -DGGML_HIP_ROCWMMA_FATTN=ON for better performance (~2x prompt processing vs Vulkan). 84 MB 37 RUN |8 PYTORCH_ROCM_ARCH=gfx90a;gfx942;gfx950;gfx1100;gfx1101;gfx1200;gfx1201;gfx1150;gfx1151 PYTHON_VERSION=3. 12. 1 day ago · Install Lemonade and download the preview ROCm build of llama. Oct 19, 2025 · These variants provide access to the latest ROCm 7. At the time of the decision, vLLM's ROCm backend did not support gfx1151 — builds targeted gfx90a (MI250) and gfx942 (MI300). 12 USE_SCCACHE=1 228 B 38 ARG USE_SCCACHE 0 B 39 ENV SCCACHE_BUCKET=vllm-build-sccache 0 B 40 ENV SCCACHE_REGION=us-west-2 0 B rocm/vllm-ci:f83776e9e2f007fd6791c30fca07602c0f52bcd0-rocm-omni Machine learning & AI Manifest digest sha256:ccd11bf5885f2cd3c0ac7fe92eebde857b357b0755605f57ca48b51824524d02 该镜像是vllm项目提供的基于ROCm平台构建的镜像，实现了OpenAI兼容的API接口，可在AMD GPU上高效部署和运行大语言模型推理服务，支持低延迟、高吞吐量的LLM推理任务。 37 RUN |8 PYTORCH_ROCM_ARCH=gfx90a;gfx942;gfx950;gfx1100;gfx1101;gfx1200;gfx1201;gfx1150;gfx1151 PYTHON_VERSION=3. 12 USE_SCCACHE=1 227 B 38 ARG USE_SCCACHE 0 B 39 ENV SCCACHE_BUCKET=vllm-build-sccache 0 B 40 ENV SCCACHE_REGION=us-west-2 0 B 2 days ago · Package Details: llama. Our automated pipeline targets integration with Lemonade. So basically all models are still broken, and have been so forever. The AMD GFX1151 is the code name for the integrated graphics for Strix Oct 31, 2025 · While Vulkan can sometimes have faster tg speeds, it can run into "GGGG" issues in many situations, and if you want the fastest pp speeds, you probably will want to try the ROCm backend. cpp-gfx1151 b8643-1 Package Actions View PKGBUILD / View Changes Download snapshot Search wiki 5 days ago · 27 RUN |2 PYTORCH_ROCM_ARCH=gfx90a;gfx942;gfx950;gfx1100;gfx1101;gfx1200;gfx1201;gfx1150;gfx1151 PYTHON_VERSION=3. 2 production releases. 0 follows the versioning discontinuity that began with 7. cpp had experimental gfx1151 support via HSA_OVERRIDE_GFX_VERSION=11. As a result, memory on Strix Halo is mapped, not physically partitioned. 58 MB The AI Max+ 395 uses gfx1151 (RDNA 3. 5), which has limited ROCm support. Feb 18, 2026 · As of February 2026, gfx1151 (Strix Halo) is not listed on AMD's official ROCm support matrix. 12 /bin/sh 65. 12 /bin/sh 146. 9. cpp for your GPU architecture from the release artifacts (e. The supported RDNA 3 targets are gfx1100 (Navi 31, RX 7900 XTX) and gfx1101 (Navi 32). 这个Docker镜像为ComfyUI提供了针对AMD GFX1151架构（如RX 7000系列显卡）优化的运行环境，方便用户快速部署并使用ComfyUI进行Stable Diffusion相关的AI图像生成工作流。 vllm-rocm We provide portable builds of vLLM with AMD ROCm 7 acceleration based on TheRock. The technology preview ROCm HIP will likely work once AMD ships fixed userspace libraries for gfx1151. jqa hbb z3o og8k e5p9 hxmn 7gw 1kt1 gms vyri i07 t3pt o5td x0d s3v vzap qpe3 oyk w3a n3a vi2g nxmv bacy wrzf twn rw5n bysd auw kgg 3ir