From the paper "Computational Sprinting." Showing temperature vs processing power.
Computers and mobile devices run a constant CPU speed to process all of our software. That generates heat that has to be cooled either by a heatsink or some other mechanism. To make the CPU run at faster speeds usually requires overclocking which generates even more heat. What if you could get incredible speeds without the need for bulky cooling systems even for mobile devices?
That is the idea a combined team of researchers from the University of Pennsylvania and the University of California want to try with what they call ‘Computational Sprinting’. The idea is to have a multi-core (over 12 cores) CPU use incredibly fast ‘burst’ speed instead of a sustained speed which would give the user super-speed while running tasks or apps needed at that particular instance. For example, a smartphone would use one core for typical tasks while leaving the additional cores dormant until needed for more complex applications. The team demonstrated the concept in a virtual environment that ran a chip with 16 cores and found that Computational Sprinting increased performance by a factor of 10!
Again, the by-product of increased speed is heat. It is a CPU killer at worst and decreases the life-span of the chip at best. Exploiting thermal capacitance is the team's direction. The amount of capacitance dictates the amount of heat that a "sprint" can produce. Storing heat in the devices case or other passive components is always an option. The team looked into placing small bit of metal near the chip, but the heat storage capacity is low. They are also looking into phase-change materials (PCM), where heat is stored in a material as it transitions between a solid to a liquid. Between sprints, the PCM returns to its original state. Currently PCMs are the only way large amounts of heat from sprinting can be handled by their prototype.
Their experiments showed that the responsiveness of a 16W chip can be handled by a 1W mobile device using parallel computation. If the correct PCM can be chosen, we all could have our mobile devices melting in our hands in the near future.
Cabe
