Chasing seconds and cents

Chasings tenths of a second on track or in your HPC workload? Got money to burn or a budget to stick to?

Chasing seconds and cents

Over the weekend while clearing out the loft (attic for my American friends) I found some old photographs. Those of you who have grown up in an era before the smartphone may remember these old physical images produced on glossy paper, the rest of you can ask ChatGPT about them I guess

The picture set off a chain of ideas in my head which I thought I’d share, and it would be wrong to do so without sharing the picture that triggered them. This meant I had to find a scanner. Oh well, you like random pictures right? 

Ah yes, that chain of thoughts. Running #HPC is rather like motorsport. Naturally, there’s the obvious comparison of chasing tenths of a second. Whether that’s lap times or the time to run a PV calculation or CFD analysis. The similarity goes a bit deeper though.

Much as in the world of motorsport, improving performance in #highperformancecomputing is easy if you have unlimited or large budgets. Just spin up as much HB176rs_v4 or hpc7a.96xlarge (they’re the fastest cloud compute you’re going to get your hands on) as you can get your hands on. 

What if you don’t have money to burn though? What if you’re racing in a single marque series with barely enough money to buy tyres?

I guess clearing out the loft also reminded me of a quote I used in my Meng (Aero) write up:

“We haven’t the money, so we’ve got to think” – Ernest Rutherford

If you want to optimise for cost as well as performance on cloud, you better know which instance types to pick and which regions to run them in.

Luckily we can help with that :)