Hardware Resources

Available Hardware

Memory is the RAM Slurm can allocate on the node (its RealMemory, in GB). CPUs is the count Slurm schedules with --cpus-per-task - hardware threads where SMT is enabled.

CPU Nodes

test-c0 belongs only to the dev-cpu partition (the default when no -p is given; 1-hour limit). c0-c6 make up the cpu partition.

GPU Nodes

test-g0 belongs only to the dev-gpu partition (1-hour limit). Nodes temporarily reserved for a group (listed on the Overview page) are out of the shared gpu / low-gpu pool while reserved.

Private GPU Nodes

Private nodes are not part of the shared gpu / low-gpu pool; each is reachable only through its owner's group-restricted partition (see the Jobs page).

GPU Architectures Overview

The cluster has a variety of GPUs across different NVIDIA architectures:

The names below are marketing names; the exact Slurm type strings for --gpus=<type>:<n> (e.g. nvidia_rtx_6000_ada_generation) are listed on the Jobs page.

Blackwell Architecture (Latest)

• RTX Pro 6000 Max-Q: 96GB VRAM (g0, g7)
• RTX Pro 6000 SE: 96GB VRAM (g2, g11)
• RTX Pro 5000: 48GB VRAM (priv-g1, priv-g2, priv-g3, priv-g6)

Ada Architecture

• RTX 6000 Ada Generation: 48GB VRAM (test-g0, g5, g16, g20, priv-g13)
• RTX 5000 Ada Generation: 32GB VRAM (g3)
• RTX 4000 Ada Generation: 20GB VRAM (priv-g10)
• L40S: 48GB VRAM (g21, priv-g14)

Ampere Architecture

• RTX A6000: 48GB VRAM (g1, g6, g14, g15, g19, g22)
• RTX A5000: 24GB VRAM (priv-g4, priv-g5)
• RTX A4000: 16GB VRAM (g4, g8, g9, g17, g18)

Turing Architecture

• RTX 2080 Ti: 11GB VRAM (g10, g12, g13, priv-g0, priv-g8, priv-g9, priv-g12, priv-g15)
• Quadro RTX 6000: 24GB VRAM (priv-g7, priv-g11)