Next Generation CPU Refrigerators

Iddo Genuth writes "Researchers at Purdue University are developing a miniature refrigeration system, small enough to fit inside laptop computers. According to the researchers, the implementation of miniature refrigeration systems in computers can dramatically increase the amount of heat removed from the microchips, therefore boosting performance while simultaneously shrinking the size of computers."

Comments

Re:How much juice?

[megaditto]

Could be pretty damn efficient if it's a heat pump.

A good AC unit usually consumes less than 10 times the energy it moves (a 1 kW window unit rated for 40,000 BTUs for example), but that depends how much colder the inside needs to be compared to the outside air.

In case of CPU coolers (cooling things hotter than ambient air), one could even GENERATE electricity if the size and cost of the "cooler" is not a concern (A thick diamond heatpipe to conduct heat away to distant thermocouples is how I would do it).

Re:Excellent

[somersault]

I don't think multi-core is going to cut it, it seems to me each processor needs it's own mememory and bandwidth to do massive calculations, and then sends the results of this information to where it is needed.

While multi-core isn't amazingly effective for doing 'massive calculations' of the variety that scientists usually do (compared to a supercomputer with thousands of nodes anyway), it is great for general purpose computing. It definitely helps for everyday use - whenever I use a single core computer (even with a high clockspeed), I notice the difference in responsiveness, especially when booting into Windows and all the system tray apps are loading, or running lots of applications at the same time. You have to remember that even if you're just running a single application on your dekstop, there are plenty of background processes too.

Not that I want to dissuade you from researching into more efficient processor methodologies, even if it's only for specific tasks - go ahead :) But when you get down to it, most tasks your average computer user does during the day are neither suitable for parallelisation, nor are they considered highly specialised. I'm just thinking of web browsing, chatting, checking email. Modern games do involve lots of operations that 20 years ago would be considered 'specialised', like 3D sound, graphics and physics processing, but we already have specialised processors for all of these things.

I'm really wondering if anyone has done any research into the geometry of information processing functions, of what can be specifically offloaded and what should not

I don't think you're giving the guys at places like Intel and AMD much credit.. if they hadn't thought about stuff like that then where did the idea for 'hyperthreading' and different CPU 'pipelines' come from? To me it seems that the only things that have changed in the last couple of decades is that we've gone from having computers that were mainly designed for integer arithmetic as far as hardware was concerned, to having computers with addons for floating point calculation, and now we have units capable of massively parallel floating point calculations and amazing amounts of memory bandwidth (graphics cards and supercomputers), and now we are getting APIs like CUDA to make use of graphics cards to do more supercomputer like things with our graphics cards. I'm not a CPU design engineer though, so the true progression is probably a bit more complex ;)