From 9364eb7ca8d182a5a39083ff6df275646643179d Mon Sep 17 00:00:00 2001 From: Sam Skillman Date: Fri, 25 Jul 2025 15:56:55 +0000 Subject: [PATCH] Move base image forward, fixing lustre/nvidia/driver compat. --- .../a3ultra-slurm-blueprint.yaml | 21 ++----------------- .../a4-highgpu-8g/a4high-slurm-blueprint.yaml | 21 ++----------------- 2 files changed, 4 insertions(+), 38 deletions(-) diff --git a/examples/machine-learning/a3-ultragpu-8g/a3ultra-slurm-blueprint.yaml b/examples/machine-learning/a3-ultragpu-8g/a3ultra-slurm-blueprint.yaml index f5631e8fad..cb49b4a6d4 100644 --- a/examples/machine-learning/a3-ultragpu-8g/a3ultra-slurm-blueprint.yaml +++ b/examples/machine-learning/a3-ultragpu-8g/a3ultra-slurm-blueprint.yaml @@ -24,7 +24,7 @@ vars: # Image settings base_image: project: ubuntu-os-accelerator-images - image: ubuntu-accelerator-2204-amd64-with-nvidia-570-v20250712 + image: ubuntu-accelerator-2204-amd64-with-nvidia-570-v20250722 image_build_machine_type: n2-standard-16 build_slurm_from_git_ref: 6.10.0 # Cluster env settings @@ -147,24 +147,7 @@ deployment_groups: ansible.builtin.apt: deb: "{{ cuda_repo_filename }}" state: present - # The following 2 tasks work around a temporary issue with Ubuntu - # packaging of NVIDIA 570 driver series for kernel 6.8.0-1032 - # This command ensures that any holds are removed before attempting an upgrade. - # We ignore failures in case the packages were not held. - - name: Unhold NVIDIA driver packages - ansible.builtin.command: - cmd: apt-mark unhold linux-modules-nvidia-570-server-open-gcp linux-modules-nvidia-570-server-open-6.8.0-1032-gcp - become: true - changed_when: false - failed_when: false - - name: Install latest NVIDIA driver metapackage and kernel module - ansible.builtin.apt: - name: - - linux-modules-nvidia-570-server-open-gcp - - linux-modules-nvidia-570-server-open-6.8.0-1032-gcp - state: latest - update_cache: yes - become: true + - name: Reduce NVIDIA repository priority ansible.builtin.copy: dest: /etc/apt/preferences.d/cuda-repository-pin-600 diff --git a/examples/machine-learning/a4-highgpu-8g/a4high-slurm-blueprint.yaml b/examples/machine-learning/a4-highgpu-8g/a4high-slurm-blueprint.yaml index ebf1b432fc..911f73ee36 100644 --- a/examples/machine-learning/a4-highgpu-8g/a4high-slurm-blueprint.yaml +++ b/examples/machine-learning/a4-highgpu-8g/a4high-slurm-blueprint.yaml @@ -24,7 +24,7 @@ vars: # Image settings base_image: project: ubuntu-os-accelerator-images - image: ubuntu-accelerator-2204-amd64-with-nvidia-570-v20250712 + image: ubuntu-accelerator-2204-amd64-with-nvidia-570-v20250722 image_build_machine_type: n2-standard-16 build_slurm_from_git_ref: 6.10.0 # Cluster env settings @@ -148,24 +148,7 @@ deployment_groups: ansible.builtin.apt: deb: "{{ cuda_repo_filename }}" state: present - # The following 2 tasks work around a temporary issue with Ubuntu - # packaging of NVIDIA 570 driver series for kernel 6.8.0-1032 - # This command ensures that any holds are removed before attempting an upgrade. - # We ignore failures in case the packages were not held. - - name: Unhold NVIDIA driver packages - ansible.builtin.command: - cmd: apt-mark unhold linux-modules-nvidia-570-server-open-gcp linux-modules-nvidia-570-server-open-6.8.0-1032-gcp - become: true - changed_when: false - failed_when: false - - name: Install latest NVIDIA driver metapackage and kernel module - ansible.builtin.apt: - name: - - linux-modules-nvidia-570-server-open-gcp - - linux-modules-nvidia-570-server-open-6.8.0-1032-gcp - state: latest - update_cache: yes - become: true + - name: Reduce NVIDIA repository priority ansible.builtin.copy: dest: /etc/apt/preferences.d/cuda-repository-pin-600