From cb1a52aee3b31c692a6629ebd7ff5503b3cd13f7 Mon Sep 17 00:00:00 2001 From: Tyler Michael Smith Date: Sat, 8 Aug 2026 21:17:50 -0400 Subject: [PATCH] [Build] Upgrade runtime image to Ubuntu 24.04, pick up rdma-core > 44 (#51058) Signed-off-by: Tyler Michael Smith Co-authored-by: OpenAI Codex Co-authored-by: Michael Goin --- docker/Dockerfile | 4 ++-- docker/versions.json | 6 +++--- docs/getting_started/installation/gpu.cuda.inc.md | 2 ++ 3 files changed, 7 insertions(+), 5 deletions(-) diff --git a/docker/Dockerfile b/docker/Dockerfile index 22a55fe1ec..c020488828 100644 --- a/docker/Dockerfile +++ b/docker/Dockerfile @@ -24,7 +24,7 @@ ARG CUDA_VERSION=13.0.3 ARG PYTHON_VERSION=3.12 -ARG UBUNTU_VERSION=22.04 +ARG UBUNTU_VERSION=24.04 # DeepEPv2 requires NCCL >= 2.30.4 (GIN backend). # This version is used for CUDA 12+ builds to keep the DeepEP build and # runtime NCCL versions compatible. @@ -765,7 +765,7 @@ RUN --mount=type=cache,target=/opt/uv/cache \ # Install gdrcopy (saves ~6s per build) # TODO (huydhn): There is no prebuilt gdrcopy package on 12.9 at the moment ARG GDRCOPY_CUDA_VERSION=12.8 -ARG GDRCOPY_OS_VERSION=Ubuntu22_04 +ARG GDRCOPY_OS_VERSION=Ubuntu24_04 ARG TARGETPLATFORM COPY tools/install_gdrcopy.sh /tmp/install_gdrcopy.sh RUN set -eux; \ diff --git a/docker/versions.json b/docker/versions.json index d681acfed9..185af3102b 100644 --- a/docker/versions.json +++ b/docker/versions.json @@ -8,7 +8,7 @@ "default": "3.12" }, "UBUNTU_VERSION": { - "default": "22.04" + "default": "24.04" }, "NCCL_VERSION": { "default": "2.30.7" @@ -17,7 +17,7 @@ "default": "pytorch/manylinux2_28-builder:cuda13.0" }, "FINAL_BASE_IMAGE": { - "default": "nvidia/cuda:13.0.3-base-ubuntu22.04" + "default": "nvidia/cuda:13.0.3-base-ubuntu24.04" }, "GET_PIP_URL": { "default": "https://bootstrap.pypa.io/get-pip.py" @@ -74,7 +74,7 @@ "default": "12.8" }, "GDRCOPY_OS_VERSION": { - "default": "Ubuntu22_04" + "default": "Ubuntu24_04" }, "BITSANDBYTES_VERSION_X86": { "default": "0.46.1" diff --git a/docs/getting_started/installation/gpu.cuda.inc.md b/docs/getting_started/installation/gpu.cuda.inc.md index 925d9605ab..4b82178a4d 100644 --- a/docs/getting_started/installation/gpu.cuda.inc.md +++ b/docs/getting_started/installation/gpu.cuda.inc.md @@ -321,6 +321,8 @@ You can add any other [engine-args](https://docs.vllm.ai/en/latest/configuration vLLM's Docker image comes with [CUDA compatibility libraries](https://docs.nvidia.com/deploy/cuda-compatibility/index.html) pre-installed. This allows you to run vLLM on systems with NVIDIA drivers that are older than the CUDA Toolkit version used in the image, but only supports select professional and datacenter NVIDIA GPUs. +For CUDA 13 images, the minimum host kernel is Linux 4.15 when running normally because CUDA 13 requires an R580 or newer driver. Compatibility mode supports R535 and R570 host drivers; R535 lowers the minimum host kernel to Linux 3.10, while R570 still requires Linux 4.15. Upgrading the container userland to Ubuntu 24.04 does not raise these driver-defined kernel requirements. See NVIDIA's [forward compatibility matrix](https://docs.nvidia.com/deploy/cuda-compatibility/forward-compatibility.html#use-the-right-cuda-forward-compatibility-package), [R535 requirements](https://download.nvidia.com/XFree86/Linux-x86_64/535.104.05/README/minimumrequirements.html), and [R580 requirements](https://download.nvidia.com/XFree86/Linux-x86_64/580.76.05/README/minimumrequirements.html). + To enable this feature, set the `VLLM_ENABLE_CUDA_COMPATIBILITY` environment variable to `1` or `true` when running the container: ```bash