At Virtual SC20, we talked with Solutions Architect Andrew O’Neill of Bethel, WA-based Silicon Mechanics, a systems integrator founded in 2001 with a focus on HPC, artificial intelligence and enterprise storage solutions. Clients turn to Silicon Mechanics bespoke open source solutions, along with service expertise to address complex computing challenges – which these days so often means AI.
We have seen a huge increase in AI and machine learning, and we’ve also seen a very large increase in general GPU compute as well. So a lot of people that were not doing computing on GPUs are now doing compute on a lot of GPUs. One of the features that we see really moving this forward is the new Nvidia A100 GPU , (the) Multi-Instance GPU, allowing departments that may just be looking at getting their feet wet in using a portion of the GPU and sharing it between multiple users.
Another major impact on demand for Silicon Mechanics’ capabilities: COVID19, including a federal government engagement in New York calling for a soup-to-solution, from system design and installation to training syllabus.
Not only was the user going to be using a brand new HPC cluster, they were also going to be their own system administrator. We’d originally planned for multiple days of on-site training, and we had to very quickly pivot and shift in order to successfully do it remotely. (We)…allowed the customer not only to get up to speed quickly, but to have a manual that they can reference in the future if they had any questions.
