|
The Cluster Computing System comprises 104 CPU nodes of HPE Apollo 2000 and 9 GPU nodes of HPE DL380.
Hardware (CPU Nodes)
The CPU nodes consist of three types of HPE Apollo 2000 with different installed memory (total 104 nodes). The numbers of nodes with 256GB, 512GB and 1,024GB memory are 44, 40, and 20 respectively. Each compute node is equipped with two sockets of Intel 64bit Xeon processor G6348 2.6GHz, each having 28 cores, for a total of 56 cores.
The system architecture of the cluster minimizes the impact of node failures on the overall system and offers good maintainability.
| CPU node | Apollo2000 |
|---|
| CPU | Intel Xeon Gold 6348 (Ice Lake) |
| Clock frequency | 2.6GHz |
| L3 Cache | 42MB |
| Number of cores per a node | 56 |
| CPU peak performance per a node | 4.65TFLOPS |
| Memory per a node | DDR4-3200 256GB | DDR4-3200 512GB | DDR4-3200 1,024GB |
| Storage per a node | 480GB SATA RI SSD |
| OS | Red Hat Enterprise Linux 8.7 |
| Interconnect | Between nodes:
InfiniBand HDR200 (200Gbps) |
| Number of nodes | 44 | 40 | 20 |
| CPU peak performance | 205TFLOPS | 186TFLOPS | 93.2TFLOPS |
| Batch queue | SMALL, APC |
Hardware (GPU Nodes)
The GPU nodes consist of DL380G11 (9 nodes).
The DL380G11 is a compute node equipped with two sockets of Intel 64bit Xeon processor P8462Y+ 2.8GHz, each having 32 cores, for a total of 64 cores.
It has 1,024GB of memory and is equipped with two NVIDIA H100 PCIe (80GB memory) GPU boards.
When using GPU nodes, a dedicated batch queue (APG queue) must be used.
Note that the login nodes also have GPUs, but those boards are NVIDIA A100 PCIe (80GB memory), so please be aware.
| GPU node | DL380G11 |
|---|
| CPU Model | Intel Xeon Platinum 8462Y+ (Sapphire Rapids) |
| Base Clock Frequency | 2.8GHz |
| L3 Cache | 60MB |
| Number of cores per a node | 64 |
| Theoretical CPU Performance per Node | 5.73TFLOPS |
| Main Memory per Node | DDR5-4800 1,024GB |
| Internal Disk per Node | 480GB SATA RI SSD x2 (RAID1) |
| GPU per Node | NVIDIA H100 PCIe 80GB x2 |
| OS | Red Hat Enterprise Linux 8.7 |
| Interconnect | Node-to-Node:
InfiniBand HDR200 (200Gbps) |
| Number of Nodes | 9 |
| Total Theoretical CPU Performance | 51.6TFLOPS |
| Batch Queue | APG |
| GPU | NVIDIA H100 (PCIe) |
|---|
| Compute Capability | 9.0 |
| Micro architecture | NIVIDA Hopper |
| TF32 Tensor Cores (TFlops) | 378
756 (with sparsity） |
| FP64 Tensor Cores (TFlops) | 51 |
| FP32 (TFlops) | 51 |
| FP64 (TFlops) | 26 |
| Number of Tensor Cores | 456 |
| Number of FP32 Cores | 14,592 |
| Interconnect Interface | PCI Express Gen5 |
| Interconnect Bandwidth | 128GB/s |
| GPU Memory | 80GB HBM2e |
| GPU Memory Bandwidth | 2,000 GB/s |
Software
The operating system running on Apollo 2000 and DL380 is Red Hat Enterprise Linux (RHEL), the same as HPE Superdome Flex (hereafter SDF).
Additionally, GPU nodes equipped with GPU cards can also run GPU‑enabled applications.
| 
