diff --git a/buildlib/az-distro-release.yml b/buildlib/az-distro-release.yml index 5d47119e15f..f539f6f67bb 100644 --- a/buildlib/az-distro-release.yml +++ b/buildlib/az-distro-release.yml @@ -31,6 +31,19 @@ jobs: ubuntu22_cuda11: build_container: ubuntu22_cuda11 artifact_name: $(POSTFIX)-ubuntu22.04-mofed5-cuda11.tar.bz2 + centos7_cuda12: + build_container: centos7_cuda12 + artifact_name: $(POSTFIX)-centos7-mofed5-cuda12.tar.bz2 + ubuntu18_cuda12: + build_container: ubuntu18_cuda12 + artifact_name: $(POSTFIX)-ubuntu18.04-mofed5-cuda12.tar.bz2 + ubuntu20_cuda12: + build_container: ubuntu20_cuda12 + artifact_name: $(POSTFIX)-ubuntu20.04-mofed5-cuda12.tar.bz2 + ubuntu22_cuda12: + build_container: ubuntu22_cuda12 + artifact_name: $(POSTFIX)-ubuntu22.04-mofed5-cuda12.tar.bz2 + container: $[ variables['build_container'] ] diff --git a/buildlib/azure-pipelines-release.yml b/buildlib/azure-pipelines-release.yml index c13bf6e8a2c..41863120fde 100644 --- a/buildlib/azure-pipelines-release.yml +++ b/buildlib/azure-pipelines-release.yml @@ -15,16 +15,25 @@ variables: resources: containers: - container: centos7_cuda11 - image: rdmz-harbor.rdmz.labs.mlnx/ucx/centos7-mofed5-cuda11:2 + image: rdmz-harbor.rdmz.labs.mlnx/ucx/x86_64/centos7-mofed5-cuda11:3 options: $(DOCKER_OPT_VOLUMES) - container: centos8_cuda11 - image: rdmz-harbor.rdmz.labs.mlnx/ucx/centos8-mofed5-cuda11:2 + image: rdmz-harbor.rdmz.labs.mlnx/ucx/x86_64/centos8-mofed5-cuda11:3 - container: ubuntu18_cuda11 - image: rdmz-harbor.rdmz.labs.mlnx/ucx/ubuntu18.04-mofed5-cuda11:3 + image: rdmz-harbor.rdmz.labs.mlnx/ucx/x86_64/ubuntu18.04-mofed5-cuda11:3 - container: ubuntu20_cuda11 - image: rdmz-harbor.rdmz.labs.mlnx/ucx/ubuntu20.04-mofed5-cuda11:3 + image: rdmz-harbor.rdmz.labs.mlnx/ucx/x86_64/ubuntu20.04-mofed5-cuda11:3 - container: ubuntu22_cuda11 - image: rdmz-harbor.rdmz.labs.mlnx/ucx/ubuntu22.04-mofed5-cuda11:3 + image: rdmz-harbor.rdmz.labs.mlnx/ucx/x86_64/ubuntu22.04-mofed5-cuda11:3 + - container: centos7_cuda12 + image: rdmz-harbor.rdmz.labs.mlnx/ucx/x86_64/centos7-mofed5-cuda12:3 + options: $(DOCKER_OPT_VOLUMES) + - container: ubuntu18_cuda12 + image: rdmz-harbor.rdmz.labs.mlnx/ucx/x86_64/ubuntu18.04-mofed5-cuda12:3 + - container: ubuntu20_cuda12 + image: rdmz-harbor.rdmz.labs.mlnx/ucx/x86_64/ubuntu20.04-mofed5-cuda12:3 + - container: ubuntu22_cuda12 + image: rdmz-harbor.rdmz.labs.mlnx/ucx/x86_64/ubuntu22.04-mofed5-cuda12:3 stages: - stage: Prepare diff --git a/buildlib/dockers/docker-compose.yml b/buildlib/dockers/docker-compose.yml index 57ba453f711..767ac468bc8 100644 --- a/buildlib/dockers/docker-compose.yml +++ b/buildlib/dockers/docker-compose.yml @@ -1,11 +1,16 @@ version: "3.4" +# To build images, run: +# docker compose -f docker-compose.yml build -# Find driver version based on CUDA version, OS and CPU arch (515 in this case): +# Find MOFED_VERSION per OS and arch: +# https://network.nvidia.com/products/infiniband-drivers/linux/mlnx_ofed/ + +# Find NV_DRIVER_VERSION per CUDA version, OS and arch: # https://developer.nvidia.com/cuda-11-7-0-download-archive?target_os=Linux&target_arch=x86_64&Distribution=Ubuntu&target_version=22.04&target_type=runfile_local services: centos7-mofed5-cuda11: - image: centos7-mofed5-cuda11:2 + image: centos7-mofed5-cuda11:3 build: context: . network: host @@ -15,19 +20,9 @@ services: MOFED_OS: rhel7.6 CUDA_VERSION: 11.4.0 OS_VERSION: 7 - centos7-mofed5.4-cuda11: - image: centos7-mofed5.4-cuda11 - build: - context: . - network: host - dockerfile: centos-release.Dockerfile - args: - MOFED_VERSION: 5.4-1.0.3.0 - MOFED_OS: rhel7.6 - CUDA_VERSION: 11.2.0 - OS_VERSION: 7 + ARCH: x86_64 centos8-mofed5-cuda11: - image: centos8-mofed5-cuda11:2 + image: centos8-mofed5-cuda11:3 build: context: . network: host @@ -37,6 +32,7 @@ services: MOFED_OS: rhel8.2 CUDA_VERSION: 11.4.0 OS_VERSION: 8 + ARCH: x86_64 ubuntu18.04-mofed5-cuda11: image: ubuntu18.04-mofed5-cuda11:3 build: @@ -48,6 +44,7 @@ services: UBUNTU_VERSION: 18.04 CUDA_VERSION: 11.4.0 NV_DRIVER_VERSION: 470 + ARCH: x86_64 ubuntu20.04-mofed5-cuda11: image: ubuntu20.04-mofed5-cuda11:3 build: @@ -59,6 +56,7 @@ services: UBUNTU_VERSION: 20.04 CUDA_VERSION: 11.4.0 NV_DRIVER_VERSION: 470 + ARCH: x86_64 ubuntu22.04-mofed5-cuda11: image: ubuntu22.04-mofed5-cuda11:3 build: @@ -70,3 +68,52 @@ services: UBUNTU_VERSION: 22.04 CUDA_VERSION: 11.7.0 NV_DRIVER_VERSION: 515 + ARCH: x86_64 + centos7-mofed5-cuda12: + image: centos7-mofed5-cuda12:3 + build: + context: . + network: host + dockerfile: centos-release.Dockerfile + args: + MOFED_VERSION: 5.0-1.0.0.0 + MOFED_OS: rhel7.6 + CUDA_VERSION: 12.0.0 + OS_VERSION: 7 + ARCH: x86_64 + ubuntu18.04-mofed5-cuda12: + image: ubuntu18.04-mofed5-cuda12:3 + build: + context: . + network: host + dockerfile: ubuntu-release.Dockerfile + args: + MOFED_VERSION: 5.0-1.0.0.0 + UBUNTU_VERSION: 18.04 + CUDA_VERSION: 12.0.0 + NV_DRIVER_VERSION: 525 + ARCH: x86_64 + ubuntu20.04-mofed5-cuda12: + image: ubuntu20.04-mofed5-cuda12:3 + build: + context: . + network: host + dockerfile: ubuntu-release.Dockerfile + args: + MOFED_VERSION: 5.0-1.0.0.0 + UBUNTU_VERSION: 20.04 + CUDA_VERSION: 12.0.0 + NV_DRIVER_VERSION: 525 + ARCH: x86_64 + ubuntu22.04-mofed5-cuda12: + image: ubuntu22.04-mofed5-cuda12:3 + build: + context: . + network: host + dockerfile: ubuntu-release.Dockerfile + args: + MOFED_VERSION: 5.4-3.6.8.1 + UBUNTU_VERSION: 22.04 + CUDA_VERSION: 12.0.0 + NV_DRIVER_VERSION: 525 + ARCH: x86_64 diff --git a/buildlib/dockers/ubuntu-release.Dockerfile b/buildlib/dockers/ubuntu-release.Dockerfile index 377dbcbdfe6..a8e73d038cc 100644 --- a/buildlib/dockers/ubuntu-release.Dockerfile +++ b/buildlib/dockers/ubuntu-release.Dockerfile @@ -20,6 +20,7 @@ RUN apt-get update && \ libnvidia-compute-${NV_DRIVER_VERSION} \ make \ maven \ + pkg-config \ udev \ wget \ environment-modules \