Will AI Take HPC Jobs?

Is Devin coming for your HPC job? Or Claude?

Will AI Take HPC Jobs?

HPC engineers usually like to boast that AI has been made possible through the use of HPC techniques if not products to train the AI models that power todays LLMs. That’s not the topic for today though. 

What I’m wondering is, when will we see the output of that same AI help HPC? 

The internet has been awash lately with talk of Devin, Claude and ChatGPT coming for software engineering jobs. I’ve not heard many HPC engineers talking about how its helping in that domain let alone worried about losing their jobs to a silicon overlord. 

Given the frankly insane projected power usage (and therefore cost) for training the next generation of AI models you’d thnk that someone would be pointing that AI canon straight back at the HPC needed to train it right?

I’ve previously looked at using AI/ML in some of our clients’ HPC estates to optimise compute utilisation. There is definitely scope to do so. The biggest problem right now in my opinion is that many of those HPC systems run fairly opaque codebases. There is little to no information exposed about each task. Couple that with a legacy codebase that there is little appetite to change, and progress is slow.

Anyone fancy sharing their experiences of using AI in HPC (rather than using HPC to train AI)?