GPUs: Can you share yours?
A little more flavour on the headline topics for the next HPC Club event on the 12th of March. The intersection of HPC and AI and sharing infrastructure.
Tell me if this sounds familiar. A part of the business owns a not insignificant number of GPUs. Ok its quite a lot of GPUs. They’re heavily utilised, but only within certain (normally well defined) time windows. Outside of those time periods they sit idle. There is no way they are about to be shared if there’s the slightest risk of them not being available when needed. So, they’re not shared at all. When it doubt, take the nuke it from space approach.
Two or three years ago that was fine. No one else wanted them anyway. Keep yer stinking GPUs! Now? More like trying to guard gold in a secret cave that everyone has a treasure map for. There’s demand from lots of new consumers. Many of which may not be profit centres or have much political standing within the business.
And you have the fun job of squaring that particular GPU shaped circle.
Want to chat to other people who have the same problem? Hear from Martin Callaghan (from Red Oak Consulting) who knows a thing or two about this space?
Come along to HPC Club on 12th March.
