TamIA

Field	Value
Availability	March 31, 2025
Login Node	tamia.alliancecan.ca
Automation Node	Automation Node: robot.tamia.ecpia.ca
Globus Collection	TamIA's Globus v5 Server
Copy Node (rsync, scp, sftp,...)	tamia.alliancecan.ca
Portal	https://portail.tamia.ecpia.ca/

TamIA is a cluster dedicated to the needs of the Canadian scientific community in artificial intelligence. TamIA is located at Université Laval and is co-managed with Mila and Calcul Québec. Its name recalls the tamia, a rodent mammal present in North America.

This cluster is part of the Pan-Canadian Artificial Intelligence Computing Environment (PACEI).

Specific Features

• It is our policy that TamIA compute nodes do not have Internet access. To request an exception, please contact technical support, explaining what you need and why.
• Note that the crontab tool is not available.
• Note that the VSCode integrated development environment is forbidden on the login nodes due to its heavy load. It is still permitted on compute nodes.
• Each job should have a minimum duration of one hour (at least five minutes for test jobs) and you cannot have more than 1000 jobs (running and pending) at a time.
• The maximum duration for a job is one day (24 hours).
• Each job must use all GPUs on the allocated servers: 4 for H100s and 8 for H200s.

Access

1. To access the computing cluster, each researcher must complete an access request in the CCDB (select the 'Artificial Intelligence' tab, then 'tamIA'). Effective access to the cluster may take up to an hour after completing the access request.
2. To submit compute jobs, you must be a member of a Resource Allocation Project (RAP) prefixed with aip-. If you are a Principal Investigator and do not yet have such a RAP, you must submit a declaration of intended use of artificial intelligence.

The procedure for sponsoring other researchers is as follows:

• On the CCDB homepage, consult the 'Project with Resource Allocation' table;
• Look for the aip- project RAP ID and click on it to be redirected to the RAP management page;
• At the bottom of the RAP management page, click on Manage Project Membership;
• In the 'Add Members' section, enter the member's CCRI.

The computing cluster is accessible only from Canada.

Storage

Space	File System	Details
HOME	Lustre File System	* This space is small and cannot be expanded; you will need to use your project space for large storage needs. Small fixed per-user quotas There is currently no automatic backup. (Planned for Spring 2026)
SCRATCH	Lustre File System	* Large space for storing temporary files during computations. No automatic backup system. Large fixed per-user quotas. * There is an automatic purging of old files in this space.
PROJECT	Lustre File System	* This space is designed for sharing data among group members and for storing large amounts of data. Large adjustable per-project quotas. There is an automatic daily backup.

At the very beginning of this page, a table lists several connection addresses. For data transfers via Globus, you must use the Globus Drop Point. However, for tools like rsync and scp, you must use the address of the Copy Node.

High-Performance Networking

The InfiniBand Nvidia NDR network connects all nodes in the cluster. Each GPU is connected to an NDR200 port via an Nvidia ConnectX-7 card. Each server therefore has 4 or 8 NDR200 ports connected to the InfiniBand fabric.

The InfiniBand network is non-blocking for compute servers and consists of 2 layers of switches arranged in a "fat-tree" topology. Storage and compute nodes are connected via 4 or 8 400Gb/s connections to the network core.

Node Specifications

Nodes	Cores	Available Memory	CPU	Storage	GPU
12	64	1024GB	2 x Intel Xeon Gold 6448Y 2.1 GHz, 32C	1 x 7.68TB SSD	8 x NVIDIA HGX H200 SXM 141GB HBM3 700W, connected via NVLink
53	48	512GB	2 x Intel Xeon Gold 6442Y 2.6 GHz, 24C	1 x 7.68TB SSD	4 x NVIDIA HGX H100 SXM 80GB HBM3 700W, connected via NVLink
8	64	512GB	2 x Intel Xeon Gold 6438M 2.2G, 32C/64T	1 x 7.68TB SSD	None

Available Software Environments

The standard software environment StdEnv/2023 is the default environment on TamIA.

GPU Jobs

Jobs are assigned to full nodes. Use one of the following Slurm options:

• For a job on a node with H100 GPUs: --gpus=h100:4
• For a job on a node with H200 GPUs: --gpus=h200:8
• For multi-node jobs, use --gpus-per-nodes=h100:4 or --gpus-per-nodes=h200:8.

Monitoring Your Jobs

From the portal, you can monitor your GPU and CPU compute jobs in real-time or view past jobs to maximize resource utilization and reduce queue waiting times.

Specifically, for a job, you can visualize:

• compute core usage;
• memory usage;
• GPU usage.

It is important to use the allocated resources effectively and to adjust your requests if compute resources are underutilized or unused. For example, if you request four CPU cores but only use one, you should adjust your submission file accordingly.