HANDY AND FREE TO USE GPU SYSTEMS FOR MACHINE/DEEP LEARNING
What Needs to be Saved
From time to time, you may hear or read words like “Artificial Intelligence (AI)”, “Deep Learning (DL)” and “Machine Learning (ML)” which are related to the Smart Nation Initiatives since it was introduced in November 2014. If you happen be one of the researchers working on AI/DL/ML, you would be very keen to run your machine learning or deep learning on a powerful GPU system with multiple GPU cards and thousands of Tensor cores.
So, here are two ready GPU resources you can consider to run your large and computing intensive machine learning or deep learning workloads.
• On-Premise GPU system maintained by NUS Information Technology (Volta)
• Remote GPU system maintained by National Supercomputing Centre (NSCC) (AI System)
A comparison of the technical details of the Volta GPU system at NUS and the AI system at NSCC is tabulated below.
Table 1. Technical Specifications for NUS-Volta System and NSCC-AI System
| Technical Specifications | Volta GPU System at NUS | AI System at NSCC |
|---|---|---|
| GPU Server Model | Dell PowerEdge C4140 | Nvidia DGX-1 |
| Number of GPUs Per Server | 4 | 8 |
| Max Workload Per Server | 4 | 8 |
| GPU Model | Nvidia Tesla® V100-32GB | Nvidia Tesla® V100-16GB |
| GPU Interconnect | NVlink | NVlink |
| GPU Capacity | 20 GPU Cards | 48 GPU Cards |
| Supported Containers | Singularity | Singularity, Docker |
| Supported ML/DL Platform | Tensorflow, Caffe, Caffe2, PyTorch, Torch7, etc. | Tensorflow, Caffe, Caffe2, PyTorch, cntk, Theano, etc. |
| Job Management & Scheduler | PBS Pro | PBS Pro |
| Run Time Limit | 48 hours (extendible) | 24 hours |
| Eligible Users | NUS staff, NUS Students, and Research Collaborators | NUS staff, NUS Students, and Research Collaborators |
| Cost to Use | Free and Fair share | Free and Fair free |
| Application to Start | Link | Link |
| User Guide | Online User Guide | Online User Guide |
You can choose to use either one or both systems to run your deep learning. Since the same job scheduler, PBS Pro, is used on both system, job submission procedures on both the NUS’s Volta system and NSCC’s AI system are very much similar.
Please feel free to contact us via A.S.K. or email at gs.ude.sun@gnireenignEataD if you need any assistance for running machine/deep learning on the two GPU systems.
