Skip to content

Commit 2cfda7e

Browse files
authored
Merge pull request GoogleCloudPlatform#4512 from arpit974/update-A3M-slurm-gcp
update A3M blueprint with new slurm gcp verion.
2 parents 5947f83 + 4f70d56 commit 2cfda7e

1 file changed

Lines changed: 1 addition & 36 deletions

File tree

examples/machine-learning/a3-megagpu-8g/a3mega-slurm-blueprint.yaml

Lines changed: 1 addition & 36 deletions
Original file line numberDiff line numberDiff line change
@@ -77,12 +77,6 @@ deployment_groups:
7777
enabled: true
7878
world_writable: true
7979
runners:
80-
- type: data
81-
destination: /etc/apt/preferences.d/block-broken-nvidia-container
82-
content: |
83-
Package: nvidia-container-toolkit nvidia-container-toolkit-base libnvidia-container-tools libnvidia-container1
84-
Pin: version 1.17.7-1
85-
Pin-Priority: 100
8680

8781
# The following holds NVIDIA software that was already installed on the
8882
# accelerator base image to be the same driver version. This reduces the
@@ -148,7 +142,7 @@ deployment_groups:
148142
apt-get install -y git
149143
ansible-galaxy role install googlecloudplatform.google_cloud_ops_agents
150144
ansible-pull \
151-
-U https://github.com/GoogleCloudPlatform/slurm-gcp -C 6.9.1 \
145+
-U https://github.com/GoogleCloudPlatform/slurm-gcp -C 6.10.2 \
152146
-i localhost, --limit localhost --connection=local \
153147
-e @/var/tmp/slurm_vars.json \
154148
ansible/playbook.yml
@@ -214,41 +208,12 @@ deployment_groups:
214208
hosts: all
215209
become: true
216210
vars:
217-
distribution: "{{ ansible_distribution | lower }}{{ ansible_distribution_version | replace('.','') }}"
218-
package_url: https://developer.download.nvidia.com/compute/cuda/repos/{{ distribution }}/x86_64/cuda-keyring_1.1-1_all.deb
219-
package_filename: /tmp/{{ package_url | basename }}
220211
enable_ops_agent: $(vars.enable_ops_agent)
221212
enable_nvidia_dcgm: $(vars.enable_nvidia_dcgm)
222213
nvidia_packages:
223214
- cuda-toolkit-12-8
224215
- datacenter-gpu-manager-4-cuda12
225216
tasks:
226-
- name: Download NVIDIA repository package
227-
ansible.builtin.get_url:
228-
url: "{{ package_url }}"
229-
dest: "{{ package_filename }}"
230-
- name: Install NVIDIA repository package
231-
ansible.builtin.apt:
232-
deb: "{{ package_filename }}"
233-
state: present
234-
- name: Reduce NVIDIA repository priority
235-
ansible.builtin.copy:
236-
dest: /etc/apt/preferences.d/cuda-repository-pin-600
237-
mode: 0o0644
238-
owner: root
239-
group: root
240-
content: |
241-
Package: nsight-compute
242-
Pin: origin *ubuntu.com*
243-
Pin-Priority: -1
244-
245-
Package: nsight-systems
246-
Pin: origin *ubuntu.com*
247-
Pin-Priority: -1
248-
249-
Package: *
250-
Pin: release l=NVIDIA CUDA
251-
Pin-Priority: 400
252217
- name: Install CUDA & DCGM
253218
ansible.builtin.apt:
254219
name: "{{ item }}"

0 commit comments

Comments
 (0)