Racing Lines & HPC
HPC is used for F1, but what can HPC learn from the world of motor racing?
When it comes to cloud adoption for #highperformancecomputing many of my client conversations start out around cost. Will it be cheaper than running on-prem? Often this is a really difficult question to answer too. Getting the on-prem cost (in a way that’s comparable to cloud) is often really difficult if not impossible. I get it though. Who on earth wants to increase their op-ex especially in the current economic climate.
Sometimes though that’s just the wrong question to ask.
When it comes to cloud adoption for #highperformancecomputing many of my client conversations start out around cost. Will it be cheaper than running on-prem? Often this is a really difficult question to answer too. Getting the on-prem cost (in a way that’s comparable to cloud) is often really difficult if not impossible. I get it though. Who on earth wants to increase their op-ex especially in the current economic climate.
Using public cloud we can spin up a literal super computer in minutes. Then when you’re done (maybe only a few minutes later) simply turn it all off again, paying only for the time it was running.
It goes without saying doing something like this would be impossible on prem. I can already hear the groans asking, sure but what’s the point of that?
Ever wanted to validate a new idea, perhaps a new financial risk model, maybe even a whole new trading desk? But can’t because running it on the on-prem HPC cluster which is fully occupied running production workloads would take months?
Ever wanted to know your trading risk for the derivs desk at the end of the day instead of the next morning due to large market events?
The best part: It wouldn’t even cost any more to run a large cluster for a short amount of time vs a small cluster for hours on end. CPU hours cost the same any which way you slice them up.
So next time you’re weighing up that cloud adoption, don’t forget the lost opportunity costs in the equation. It may not always make sense to run all your workload on cloud, but it doesn’t make sense to have zero cloud capability either.
Look past that first corner into the next bend.
