Overview of AI-Compliant Advanced Computer System
This system will be discontinued at the end of FY2023.
In December 2019 the Information Initiative Center at Hokkaido University introduced the “AI-Compliant Advanced Computer System” for Hokkaido University researchers with the aim of facilitating and accelerating research projects by transferring artificial intelligence technology to a variety of research fields, including artificial intelligence research using accelerators and cyber-physical systems.
System features
- Equipped with 4 high-performance GPUs per node (Tesla V100 SXM2, 32 GB memory)
- High-performance common storage equipped with the latest NVMe SSDs
- Supports GPU-optimized machine learning container, and operates job scheduler
- Linkable to Interdisciplinary Large-Scale Computer System (High Performance Inter-cloud)
System specifications
Computation node specifications | ||||
Product name | FUJITSU PRIMERGY CX400 M4 (chassis), CX2570 M5 (computation nodes,water cooling) | |||
No. of nodes | 8 nodes (4 chasses) | |||
271 TFLOPS (double precision) | ||||
Node specs. | CPU | Intel Xeon Gold 6230 (2.1 GHz/20-core) x 2 | ||
2.688 TFLOPS | ||||
GPU | NVDIA Tesla V100 SXM2 (32GB) x 4 | |||
31.2 TFLOPS (double precision),62.8 TFLOPS (single precision) | ||||
Mem. | 384GB (32GB DDR4-2933 RDIMM x 12) | |||
SSD | 1.6TB NVMe SSD x 1 | |||
Inter-Connection | 100Gbps Intel Omni-Path x 1 | |||
Common storage specifications | ||||
Product name | DDN ES200NV | |||
Controller | Active/active dual controller | |||
Capacity | 76.8TB (3.84TB dual-port NVMe SSD x 21) | |||
Max. performance | 20GB/s, 1M IOPS | |||
Inter-Connection | 100Gbps Intel Omni-Path x 2 | |||
Software specifications | ||||
Machine learning | TensorFlow, PyTorch, Chainer, Caffe, etc. | |||
Parallelizing compiler | PGI Professional Edition |