July was the hottest month in recorded history, scorching past a 40-year record. Similarly, data centers have also begun to reach record temperatures, and standard air-cooling solutions can no longer suffice.
Liquid cooling solutions are growing in popularity as a potential solution to mitigate increasing data center temperatures. Cloud service providers have begun to encounter power density issues because of more powerful data processing units now available on the market.
As power density increases with little sign of stopping, reaching beyond the 100kW per rack mark, data center operators have begun the process of fitting and retrofitting cooling solutions that can accommodate these new demands.
The current industry standard cooling solution for data centers involves air-cooling methods, which typically utilize a heatsink installed adjacent to the processing unit. Heatsinks utilize electric fans to disperse heat into the air and away from critical components. Air-cooling methods are usually sufficient for server racks that consume up to 50kW of power. However, beyond this figure, fan sets become cost and power inefficient.
Liquid cooling solutions come in a few varieties depending on factors such as available budget, space, and expertise so Silverlinings is bringing you the top three solutions:
1. Immersion cooling
Immersion cooling or ‘the carbonite method’ is a liquid cooling solution that involves completely submerging the server in coolant. The coolant is more thermally conductive than air but does not conduct electricity (zzzap!) or make metals rusty (tin roof!). Immersion cooling solutions come in two varieties, single and dual phase.
Single Phase: In single-phase immersion cooling the coolant will always stay in a liquid state and travels through a heat exchanger creating a cooling circuit. The servers are submerged vertically in a hydrogen-based coolant bath that can be left open, as there is no risk of evaporation.
Single-phase immersion could be considered the simplest application of liquid cooling technology. Outside of ensuring that critical hardware within the server is not permeable to the coolant used, single phase immersion is simpler to install, easier to manage, and more affordable for enterprises running server racks consuming up to 70kW of power.
Dual Phase: Dual-phase immersion cooling disperses heat through a process of low-temperature evaporation. This is achieved by putting fluorinated coolant in direct contact with the hottest parts of the server. The gas is then cooled via a heat exchanger until it returns to its liquid state.
Dual phase cooling is the more resource-efficient solution, however it also introduces a host of extra risks, such as micro-cavitation (or the high-velocity collapse of tiny bubbles in the liquid) due to the boiling process and respiratory damage due to the potential inhalation of fluorinated fluid fumes, which are toxic.
Despite this, interest in dual phase immersion solutions has remained, likely due to its efficiency over alternatives.
[Nerd alert: In the 2007 sci-fi movie Sunshine from director Danny Boyle, the lead engineer, played by Chris Evans (a.k.a. Captain America) attempts to cool down his spaceship's immersion cooled supercomputer by diving into the coolant reservoir and manually lowering the processing units into the dielectric fluid. This is not recommended best practice for increasing the efficiency of immersion cooling systems].
2. Direct to chip
Direct to chip cooling is a non-immersion method of liquid cooling that uses tubing and pumps to deliver coolant — you guessed it — directly to the chip. This method of cooling is very effective as the tubing can be routed to target critical processing units. Additionally, as this method can incorporate dual-phase cooling, direct to chip systems are able to mitigate temperature on server racks consuming up to 100kW of power. The major selling point of direct to chip cooling in comparison to immersion cooling is customization and ease of access, taking the best parts of immersion cooling while leaving behind the draining, refilling and clean up that potential users may be wary of.
3. Rear door heat exchangers
Since its introduction, rear door heat exchangers have been widely used in data centers. Standard designs, which are typically rated to cool servers that consume up to 20kW of power, are easily implemented and compatible with most traditional server configurations.
Using fans, units draw hot air through the back of the server, warming coolant that runs through the server’s radiator coils, a coolant distribution unit receives the heated fluids through these radiator coils and sends them back to the server once they have cooled.
Rear door heat exchangers, while not the coldest liquid cooling solution, do offer some significant benefits. They take up minimal floor space, which is useful in dense computing environments. Additionally, they are low maintenance in comparison to other solutions, only occasionally needing to be cleaned. Despite these benefits rear door heat exchangers make up the lower ranks among liquid cooling solutions as data centers evolve past outdated standards of power consumption.