Wulver¶
Wulver is NJIT's newest cluster, which went into production in early 2024. To get started with Wulver, please check the details in get started on Wulver
Specifications:¶
General | Bigmem | GPU | |
---|---|---|---|
Partition | general |
bigmem |
gpu |
Nodes | 100 | 2 | 25 |
CPU Type | AMD EPYC 7753 | AMD EPYC 7753 | AMD EPYC 7713 |
CPU Speed (GHZ) | 2.45 | 2.45 | 2.0 |
CPUs/Socket | 64 | 64 | 64 |
Sockets/Node | 2 | 2 | 2 |
RAM/Node (GB) | 512 | 2048 | 512 |
GPU Type | - | - | nVIDIA A100 |
GPUs/Node | - | - | 4 |
GPU Memory (GB) | - | - | 80 |
Total CPUs | 12800 | 256 | 3200 |
Total RAM (TB) | 50 | 4 | 12.5 |
Total SUs/year | 112128000 | 2242560 | 28032000 |
Peak Performance (TFLOPs) | 501.76 | 10.035 | 1072.4 |
HPL (TFLOPs/Node) | - | - | - |
- 100 GB High throughput/low latency Infiniband network
- Arcastream Storage
- Parallel filesystem allowing multiple read/write operations
- Max performance 20 GB/s
- Hierarchical storage; NVME scratch, SAS, Cloud
- User transparent transfer from high-speed to medium to archive storage
- 1 PB total storage
- Virtualized Support and login nodes
- Management and deployment overseen by NJIT:
- Nvidia Bright Cluster Manager
- X-ISS system administration
- SLURM scheduler with full support
- InfiniBand support via Dell or directly from Mellanox
- Will be deployed in the Databank data center in Piscataway