Tesla V100 Nvlink GPU node

High-performance GPU-based intensive compute node

In less than a month, a new compute node will augment the current GPU-based systems available at the Swiss Geocomputing Centre. Acquired thanks to the support of the FINV – Unil fond, this intensive compute node is the building block of tomorrow’s most powerful supercomputers. The 4U rack is composed of a 2U unit hosting two Intel Scalable Silver 4112 (4 cores) processors and 768 GB of RAM. The second 2U unit contains 8 state-of-the-art Tesla Volta 100 Nvlink GPUs. Each of the graphical processor is equipped with 32 GB of on-chip memory, and achieves a theoretical memory bandwidth close to 1 TB/s. The system features 8 Tesla V100 cards interconnected with the Nvidia Nvlink technology, and is capable to deliver an overall processing power up to 40’000 cores, which is the equivalent of 5’000 usual 8 cores computer processors (CPU).

Here is an insight of this computing monster, imaged while visiting today Colfax International, our provider and main tech partner, in Sunnyvale CA. Enjoy 🙂

View inside the 2u unit hosting the 8 Tesla V100 Nvlink GPUs. Visible is the passive cooling installation on their top.
View inside the 2u unit hosting the intel Scalable processors and the 768 Gb of RAM.