Total Pageviews

Search This Blog

Wednesday, August 20, 2025

designing a supercomputer cooled with heat exchangers

Below is a prototype I built myself. It attracted the attention of the FBI who came to my apartment to ask me about it. It was cooled with mineral oil which does not conduct electricity. The same oil has been used to cool high voltage transformers for over a century. 


Computers are about as small and cheap as they're ever going to be. Advances in software are few and far between. Thus, the only way left to improve performance is through better heat transfer. Oil immersion cooling is not a new idea, but it has yet to be embraced fully. See these cluster computers for examples:





Cluster computing is a big part of the solution and was heavily advocated by computer pioneer Grace Hopper. She compared it to hitching multiple oxen to the same yoke if one was not strong enough to do the job alone. It's a similar idea to multiengine aircraft.

Microsoft did an interesting and successful experiment called Project Natick where they packed a data center into a waterproof shipping container and submerged it. 


I think it could be better to combine the cooling power of oil and water used a shell and tube heat exchanger. Cold lake water could be pumped into the shell, and the heated cooling oil would be pumped through the tubes to keep it extra cold.



There are many cold lakes in the continental US that could be used, and the water temperature would not change much even for the cooling needs of a very large supercomputer complex such as the one in Oak Ridge. It would also be cheaper in terms of water usage than the NSA data center complex in Bluffdale, Utah which is cooled by water evaporation. Mineral oil does not evaporate, and the volume and water temperature of most lakes stays constant.

The ideal location for a supercomputer complex cooled by heat exchangers would be on the shores of Lake Superior. Using freshwater eliminates many of problems brought on by sea salt and other complications such as tides and bad weather. The main risk would be water leaking into the oil, but it would settle to the bottom of computer tank because water is denser. 

If the complex was some distance below the water level of the lake, artesian pressure could be used instead of pumps to circulate the water. Instead returning the water to the lake, it would drain into a pond for evaporation. Al valve would be opened or closed to control the water level. The main drawback is maintenance, though I suppose that could done by divers depending on the size of the oil bath. 

In my experience, it will also be necessary to keep the oil bath covered. Insects are attracted to the oil and get stuck in it. These were the literal bugs I dealt with my oil-cooled computer. There were no adverse performance effects, but all the dead bugs were unsightly. 

Basically, the oil bath would be about the size of an Olympic swimming pool with a uniform depth of maybe five feet. A bundle of tubes carrying the cooling water would pass through the oil bath on one of its sides. On the bottom of the oil bath would be a layer of copper BBs or pennies to act as auxiliary heat sinks. In this way, the oil bath acts like the shell side a very large heat exchanger with the lake being the tube side. Another way to think of it is like the way a car radiator exchanges heat to cool an engine. 

CPUs and other such chips are designed to slow down once they reach a center temperature. But if the chips through better cooling never reach that temperature, there is no limit to the speed of the system clock. If something like Intel's terahertz processor was used with the cooling system described above, I can scarcely imagine how fast the computer would be. 


There are all kinds of computer science problems thought to be unsolvable only because of the current limits of processing speed. This is the reason for all the work with quantum computers. However, those require technology that could be decades away if it is possible at all. 


Blowing air on a machine to cool it down is 19th century thinking. It is time to use mineral oil and heat exchangers. 

No comments: