Innovation@CIT: HPC, AI and 24/7 mission critical workloads all together on one platform with vGPU

At the CIT we work everyday to create innovative and reliable IT to support education and research. This is one of our latest achievements. We have this very innovative idea to change from "desktop at day, compute at night" to "desktop at day, compute in every way. Use the resources wise, so you can make use of them twice”. What this means is that they will make it possible to share resources of our 24/7 mission critical VMware virtualisation platform with HPC-GPU. The platform is spread across 3 of our data centers, in order to easily handle failures. In fact, by design, 30% of the resources are reserved for failure and doing nothing. We call this reserved overcapacity, which we are using to run HPC workloads. The platform now has 36 nodes with each a NVIDIA V100 and 2 Intel 6150 CPUs and 768 Gb memory. Cool idea, don't you think?

There has been a lot of interest in our innovative solution not only from within the university but also from outside of the university. This attention has resulted in several presentations.

Next week Gemma van der Voorst and Wietze Albers will present at NVIDIA’s GPU Technology Conference (GTC) Europe in Munich and at the same time Henk-Jan Zilverberg will present the poster at the Digital Infrastructures for Research 2018 conference in Lisbon.

In November Gemma van der Voorst and Wietze Albers will present at VMworld Europe in Barcelona.