Capacity
Suggested: Medium (M)
Hardware Configuration
32 GPUs • steps of 8 GPUs (0.5 unit)
60 GPUs • steps of 30 GPUs (whole units)
Network & Storage
Options
Generate Slurm partitions for MI350P Advanced Services nodes
Show TDP-based power draw
Departments
Ratio: 32:60 — Valid (16:30)
Expected: 16:30 per unit pair
Bill of Materials
| System | Units | GPUs | GPU Model | GPU Mem | Total Mem | VMs | Storage | Network |
|---|---|---|---|---|---|---|---|---|
| Teaching Cluster | 2 | 32 | AMD Instinct MI300X | 192 GB | 6144 GB | 16–32 | 64 TB | InfiniBand HDR |
| MI350P Advanced Services | 2 | 60 | AMD Instinct MI300X | 192 GB | 11520 GB | 15–30 | 120 TB | InfiniBand HDR |
| Total | 4 | 92 | — | — | 17664 GB | 31–62 | 184 TB | — |
Per-Student GPU Memory
35.33 GB
VM Config
32 vCPU / 256 GB
Storage per GPU
2 TB
Slurm Partition Configuration (MI350P Advanced Services)
| Partition | Nodes | GPUs/Node | Total GPUs |
|---|---|---|---|
| gpu-inference | 30 | 2 | 60 |
| gpu-batch | 15 | 4 | 60 |
Slurm can schedule across all 60 MI350P Advanced Services GPUs for large batch jobs.
Power & Cooling Estimate
GPU TDP
750W
Total Draw
82.8 kW
Includes
+20%
overhead
Estimate: 92 GPUs × 750W + 20% cooling/networking overhead
Storage Breakdown
Teaching Cluster (2 units × 16 GPUs × 2TB)64 TB
MI350P Advanced Services (2 units × 30 GPUs × 2TB)120 TB
Total184 TB (NVMe Only)