Does anyone know if there any current or near-term applications for these? Seems super cost-prohibitive, especially if you can just throw more compute at the problem instead. Super cool (hah) either way.
Just guessing since the article was information deficient: mainframes. They’re so expensive anyway that adding on a closed cycle cryogenic cooler isn’t even in the top ten line items on the bill.
At cryogenic temperatures, even older architectures not optimised for it can hit 9 GHz. An optimised modern CPU could probably run at well over 10 GHz, which is a big performance boost for legacy single-threaded code. Cooling and “overclocking” the CPUs would be cheaper than rewriting the code.
More extreme frequencies (over 100 GHz) can be achieved with existing but more exotic digital electronics such as Rapid Single Flux Quantum (RSFQ), which uses liquid helium cooling. This is used in some military radars, radio telescopes, and a few other similar high-speed signal processing scenarios. If ordinary silicon could run at, say, 20 GHz with cheaper LN cooling, that would lower the cost of some fancy radars.
Liquid nitrogen costs less than a $1/liter. Not sure what kind of reservoir volume they are talking, but indeed talking peanuts in comparison to everything else. The salary of the guy who has to maintain the cooling is going to be more.
> At cryogenic temperatures, even older architectures not optimised for it can hit 9 GHz. An optimised modern CPU could probably run at well over 10 GHz, which is a big performance boost for legacy single-threaded code. Cooling and “overclocking” the CPUs would be cheaper than rewriting the code.
When it will work, yes. You just need to take into account that some material characteristic is temperature dependent. /s
I would imagine that the first roll-out of this would be with the quantum computer division, as they already have experience with cryogenic cooling. This requires some major redesigns in chassis, encapsulation, boards, and manufacturing processes, so I would assume this would happen first in some esoteric HPC application before hitting POWER. Mainframes would most likely be the last ones - an unreliable fast mainframe is a worse value proposition than a reliable slow one.
My first thought was that this could have some "co-processor" style applications in quantum computers, but I think the "real" answer is that if datacenter workloads could 2x to 10x in efficiency using chips made with liquid nitrogen cooled transistors vs. traditional silicon, it becomes economical to move in this direction.
Only to a limited extent. The improvements don't scale at very low temperatures. Some weird stuff happens too: metals or resistors can get superconducting. Not always an advantage.
