Intel lays foundation for 1000W CPUs with experimental liquid cooling

Table of Contents
As chips become more powerful, they not only become increasingly power hungry, but they also output a lot more heat. Unfortunately, silicon works best when it is kept cool, and rising temperatures can lead to thermal throttling, which can degrade performance or even damage the chip. To tackle this problem, Intel has revealed it is experimenting with a liquid cooling solution.
At the recent Foundry Direct Connect event, Intel showcased a new experimental package-level water cooling solution that has been designed to tackle rising temperatures and do a more efficient job at keeping CPUs cool. Not just your average desktop CPU, though, we’re talking about powerful chips operating at up to 1000W.
1000W liquid prototype unveiled
The prototype has been designed to work on both LGA (Land Grid Array) and BGA (Ball Grid Array) CPUs. The demonstrations used Intel's Core Ultra line, along with Xeon server processors. Similar to commercially available AIO coolers, the cooling solution runs through a specially designed compact cooling block that sits on top of the CPU die. This has microchannels made of copper to guide coolant flow, which can be optimised to target specific hotspots on the CPU.
In practice, this works similarly to attaching a commercial AIO cooling system to a delidded CPU. Intel says this system can dissipate up to 1000 watts of heat using only standard liquid cooling fluid. This kind of thermal load is far more than would be generated on a standard consumer gaming rig, but would be valuable for high-end workstations or machine learning workloads.
The system is also said to use solder or liquid metal for the thermal interface, instead of the standard thermal paste found in most consumer setups. Liquid metal can offer better contact than standard thermal paste or thermal gel.
According to reports from Tom's Hardware, this system isn't just an internal experiment. Intel is reported to be working on how to produce this system at scale for real-world applications and deployment. There is, however, no timeline for when this system will be made available for sale.