Cloud giants 'ran out' of fast GPUs for AI boffins Top cloud providers struggled to provide enough GPUs on-demand last week, AI experts complained to The Register. As a deadline for research papers loomed for a major conference in the machine-learning world, teams around the globe scrambled to rent cloud-hosted accelerators to run tests and complete their work in time to …
Glad it's not just undergrads who wait until the last minute to write an essay!
This sounds like more than writing the paper, it is doing the actual research to go into it. So what are these researchers doing the rest of the time if it's not their AI models? Is it "There's a deadline coming up, so I'd better do some work"?
The old rule with research supercomputers was that you work out how long your problem is going to take to run on current hardware, then wait half that time and buy the new version of the hardware which will do the job in less than half the time, thus saving you time overall.
"...as AI development ramps up, hosting providers must scale to meet demand"
But the demand isn't there 95% of the time. You can't possibly accommodate every last spike in activity and still run a business. Not without it costing more than keeping the system internal to the researchers.
Bottom line, if it's a critical computing resource, make sure *you* can turn the taps on and off.
I know that Google has, on several occasions, affected the global supply for components... remember when memory was extremely difficult to source a few years ago. Pretty crazy, "I would like to place an order for memory", "Alright, how much do you need?", "How much do you have?", "On hand in the warehouse?", "No... on this planet."
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
