CPU get hot, really hot and current cooling technologies are not great at getting to that heat. Since the CPU has some components that get hotter than others and the heat spreader is pretty inefficient at transferring that heat, your CPUs thermal limit is basically the limit of the hottest component, not the average temperature across the die.
A team of Georgia Tech researchers have now shown it’s possible to get that cooling fluid right where it’s needed by cutting microfluidic passages into FPGA devices such as CPUs in existing computers. This technique gets the fluid within a few hundred microns of the actual transistors. This is orders of magnitude better than the many millimeters of material between these components and the water in current water cooling setups. How well did it work? Using a 28nm FPGA device as a testbed the team was able to beat air cooling solutions by 60%. There is no heatsink, no cooling block, just an inlet and outlet pipe connecting directly to the die itself.
The test system used water at 20C and a flow rate of 147 ml/m. A flow rate many time that of existing water cooling solutions, where flow rates are measured in litres per minute, typically in the 3 to 4 L/m range. The chip was cooled to less than 24C, compared to the 60C achieved via standard air cooling.
This research is unlikely to find its way into desktop or even high-end professional setups any time soon, but it has potential for new high performance computer design. SO don’t be surprised if we see a prototype supercomputing cluster or datacenter using this technology in the next few years. The researchers are also speculating that this approach could allow for chip stacking with short, high bandwidth interconnects. Another indication of supercomputing potential at room temperature.
This could also mean a long awaited jump in clock speed. We’ve seen overclockers achieve insane clock speed figures for brief spans of time using liquid nitrogen, but perhaps microfluidics holds the key to breaking the 4Ghz barrier that desktop CPUs seem to have hit.
We’ll be keeping a close eye on this line of research without a doubt.
A team of Georgia Tech researchers have now shown it’s possible to get that cooling fluid right where it’s needed by cutting microfluidic passages into FPGA devices such as CPUs in existing computers. This technique gets the fluid within a few hundred microns of the actual transistors. This is orders of magnitude better than the many millimeters of material between these components and the water in current water cooling setups. How well did it work? Using a 28nm FPGA device as a testbed the team was able to beat air cooling solutions by 60%. There is no heatsink, no cooling block, just an inlet and outlet pipe connecting directly to the die itself.