Nodes in Colibri

Colibri consists of a login node, head node, and compute nodes. Some useful information about these nodes are as follows:

Login Node

Whenever you log in to the Colibri, you are directed to the login node. You can use this login node to install software but don’t run any computations in it. In Colibri, hpc-s13-n13 is the login node.

Remember not to run any programs in the login node because it could damage the node if improperly executed.

Head Node

A head node is a simply configured system that acts as an intermediate between the compute nodes and the login node. You can run the jobs on the compute nodes by using the Slurm scheduling system tools on the head node. hpc-colibri-h1 is the head node in the Colibri.

Compute Nodes

Colibri has 13 compute nodes which are categorized into 8 CPU nodes and 5 GPU nodes. The compute nodes are the nodes where your programs and jobs are/should be executed.

CPU Nodes

There are 8 CPU nodes available in Colibri and the table below shows the hardware configuration of the CPU nodes.

  • SLURM Resources

  • Physical Resources

Compute Nodes CPU Per Node(Threads) RAM(GB) Available Features

hpc-s13-n[5–12]

190

742

amd, turin, epyc-9655

Compute Nodes CPU CPU Per Node Core Per CPU Cores/Threads Per Node RAM(GB/TB)

hpc-s13-n[5–12]

AMD EPYC 9655 2.60GHz

2

96

192/192

768 GB

Compute nodes are generally not accessed by the HPC user. Sometimes, you may need to check the data generated by a program on the compute nodes, which may not be clear from the head node. Under those circumstances, what you can do is that you can submit interactive jobs using Slurm to check the results of the compute nodes.

Please refer to the page → Interactive jobs in Slurm for more insights.

GPU Nodes

If your job needs GPU, Colibri has 5 GPU nodes and you are always welcome to use it as well. Configuration about the GPU nodes are listed in the following table:

  • SLURM Resources

  • Physical Resources

GPU Nodes CPU Per Node(Threads) RAM(GB) Available Features GPU Per Node GPU Memory

hpc-s13-n[1–2], hpc-r13-n1

126

1498

amd, turin, epyc-9575f, gpu, gpu-nvidia, nvidia_rtx_pro_6000, nvidia_rtx_pro_6000-96g

8

96 GB

hpc-s13-n[3–4]

126

1498

amd, turin, epyc-9575f, gpu, gpu-nvidia, nvidia_h200_nvl, nvidia_h200_nvl-141g

4

141 GB

GPU Nodes CPU CPU Per Node Core Per CPU Cores/Threads Per Node RAM(GB/TB) GPU GPU Per Node

hpc-s13-n[1–2], hpc-r13-n1

AMD EPYC 9575F 3.30GHz

2

64

128/128

1.5 TB

NVIDIA RTX PRO 6000

8

hpc-s13-n[3–4]

AMD EPYC 9575F 3.30GHz

2

64

128/128

1.5 TB

NVIDIA H200 NVL

4

Partitions

Public Partitions

Partition Name Default Status Assigned Nodes

general-cpu

Default

hpc-s13-n[5–11]

general-gpu

No

hpc-s13-n[1–3]

interactive-cpu

No

hpc-s13-n12

interactive-gpu

No

hpc-r13-n1

backfill

No

hpc-s13-n[1–11]

Condo Partitions

Partition Name Assigned Nodes

inbre

hpc-s13-n[1–3,5–11]

prism

hpc-s13-n4

Each Node in Colibri has feature tags assigned to it. Users can select nodes to run their jobs based on the feature tags using sbatch or srun --constraint flag. For more information, visit the page point_right Slurm Node Features