Deep Learning

The Mesocenter has a dedicated infrastructure for Deep-Learning. This consists of a server containing GPU Volta 7 cards and a DGX with 4 GPU Volta cards.

The set produces a total power of 85 Tflop/s

CPU

• Processor Model: Intel(R) Xeon(R) Silver 4110 CPU @ 2.10GHz
• Number of CPU cores: 16
• Memory: 128G

3xGPU

• Tesla V100
• Connection PCI express 32 Go/s
• Performance 7.8 TeraFLOPS DP
• Performance Deep Learning 112 TeraFLOPS
• Power Consumption 250 Watts
• GPU Memory 16 Go

Project Owner

• Mésocentre & I-SITE (Advances)

CPU

• Processor Model: Intel(R) Xeon(R) Silver 4110 CPU @ 2.10GHz
• Number of CPU cores: 32
• Memory: 256G

4xGPU

• Tesla V100
• Connection PCI express 32 Go/s
• Performance 7 TeraFLOPS DP
• Performance Deep Learning 112 TeraFLOPS
• Power Consumption 250 Watts
• GPU Memory 32G

Project Owner

• Mésocentre

CPU

• Processor Model: Intel Xeon E5-2698 v4. 2,2 GHz
• Number of cores: 20cores
• Memory: 256Go DDR4

4xGPU

• Tesla V100
• Total GPU Memory 128 GB total system
• Total NVIDIA Tensor Cores 2,560
• Total NVIDIA CUDA® Cores 20,480
• Storage: Data: 3X 1.92 TB SSD RAID 0, system 1X 1.92 TB SSD
• Connection NVLINK 300 Go/s
• Performance Deep Learning 112 TeraFLOPS x 4
• Performance 7.8 TeraFLOPS x 4
• Power consumption 250 Watts

Project Owner

• I-SITE (Advances)

Unlike the cluster (Centos 6.5), the servers are running Ubuntu Linux. Thus, the servers do not share any software with the rest of the cluster. In the other word, there is no module command in the servers. Software are installed directly on the servers.

We can find all Deep learning software stack: Tensorflow, keras, …

• How to use the Dell server?
• How to use the DGX?

• https://github.com/ChristosChristofidis/awesome-deep-learning
• https://github.com/jtoy/awesome-tensorflow