Ephemeral Super Computers
We’re already familiar with ephemeral infrastructure in the cloud for many things. Are ephemeral super computers really such a surprise?
Ephemeral. Fleeting. Transient.
We’re used to certain types of infrastructure being ephemeral. It’s even a highly desired target state. Think build and deployment machines. Think test environments.
And of course, the cloud makes all of this much more feasible.
What about super computers though? We don’t normally tend to think of these as ephemeral. Quite the opposite in fact. They’re usually large physical deployments taking years to build and running for decades.
They don’t have to be though. You can have ephemeral super computers.
It’s not the stuff of dreams either. Not only is it possible today but some of our clients already do exactly this. Spin up a supercomputer. Run your workload. Shut it down. Pay for what you used.
The hyperscalers even give you the basic building blocks you need to do this. (Speak to us for the rest, especially if you don’t want it to be financially crippling!)
So why isn’t this more common? Whenever I speak to the end users of HPC the biggest complaint is always lack of access to more compute. If you can spin up compute on demand now, why isn’t everyone doing it all the time?
I don’t have an answer to that question but it’s one I would really like to understand better.
Maybe it’s because we’re still missing the ability to do that from an end user perspective (risk managee, engineer, meteorologist?) Or maybe it’s just one of those things where users didn’t tell you what they really wanted?
