This Small Business Innovation Research (SBIR)Phase I project proposes to develop a high performance mineral oil submersion cooling system for computer servers. While CPU (Central Processing Unit) over-clocking offers the potential to increase output from existing servers by up to 30%, it will also increase the heat generated in a small space in a non-linear fashion likely affecting server reliability. Inefficient cooling is a key driver in data center energy consumption, which has recently risen from 1% of US electricity to a predicted 3%. Mineral oil is better and heat dissipation than air, providing better cooling performance to allow over-clocking of computer servers while reducing the amount of cooling energy needed. The proposed Phase I research objectives are to: (i) design and install a cooling system specifically for high performance cooling; (ii) investigate and implement the best method to over-clock servers; (iii) document server performance and cooling requirements of an over-clocked system over a range of CPU clock speeds; (iv) document and improve system ease-of-use and ergonomics. The anticipated technical result is to quantify the performance and cost and energy benefits of a mineral oil immersion cooling system for over-clocked servers.
The broader impact/commercial potential of this project comes from (a) increasing server processing power while (b) lowering energy use and (c) lowering build-out costs of a computer data center. For computational-heavy research institutions, over-clocking offers the potential to solve more of society's research needs with fewer resources as computer servers would perform significantly more computations than before. Also, lowering the energy of a large contributor to incremental US electricity demand will greatly benefit the environment. Finally, the build-out costs of a data center, which scales roughly in-line with power consumption, will also be greatly reduced as cooling energy is reduced.
Computer data centers are projected to use nearly 3% of US electricity in 2011, up from 1% in the year 2000. However, less than 50% of that electricity actual goes to components like the CPU that do actual computation or work, and the other 50%+ is wasted primarily on cooling inefficiencies. Our project investigated a new cooling method for computers servers in data centers by developing and installing a fluid submersion cooling method that uses 90% less cooling energy than the current method of air cooling. Our fluid, which not electrically conductive and thus is safe for electrical components, requires much less cooling energy and less cooling equipment because it has much better thermal properties than air, which means it is much more effective at capturing and transferring heat. Our system’s effectiveness drives the ability to use hardware in new methods such as over-clocking, which is getting a system to perform more computations than it was previously capable of. Overall, the cost of running a data center, including server purchases, center build-out cost, and electricity spend, will decrease by 20-30% using our system. The move away from air to a fluid for cooling is common: many devices such as automobile engines and electric transformers started being cooled by air but moved to fluids to take advantage of superior cooling. The move from air to fluid by data centers will depend on low system cost and usability. Since our low-cost method uses a non-toxic fluid similar to baby oil and has a simple and easy-to-use architecture, it provides data centers the cost justification they need to switch cooling methods.
