APEX Systems
Historical
October 13, 2020 — NVIDIA announced another large deployment in Southeast Asia of the NVIDIA DGX POD AI accelerated computing infrastructure to drive artificial intelligence (AI) research at CMKL University in Thailand. The cluster will be the central computing node that connects to research and university nodes across the country.
As part of Thailand’s AI research infrastructure, CMKL is setting up the AI computing cluster to provide capabilities for data exchange, and an AI analytics platform to support competency building in the new normal economy. This will help drive research in various domains such as food and agriculture, healthcare and smart city.
System Resources
System Diagram
System Overview
- Nvidia DGX A100 (40GB) - 6 nodes
| Components / Node | |
|---|---|
| CPU Model: | AMD EPYC™ Rome 7742 @2.25 GHz |
| Number of socket(s): | 2 |
| Cores per socket(s): | 64 |
| Cores per socket(s): | 64 |
| CPU total cores: | 256 |
| CPU L3 Cache: | 256 MB |
| Memory Type: | DDR4-3200MT/s ECC RDIMM |
| Memory total capacity: | 1024 GB |
| Accelerator (GPGPU) Model : | Nvidia A100 40 GB |
| Accelerator Unit : | 8 |
| Accelerator total Memory : | 320 GB |
| Accelerator Interconnection : | NVLink 3.0 & 6 NVSwitch 2.0 |
Networking
- Nvidia Mellanox InfiniBand HDR 200 Gb/s
- Maximum bandwidth: 1.6 TB/s IB GPU-to-GPU between nodes
- Nvidia Mellanox Spectrum Ethernet 100 Gb/s
Storage
- Total 3 PB of high-performance parallel storage
- 2.2 PB of SAS 12G HDD storage
- 550 TB of NVMe SSD storage