Thanks in part to the popularity of closed-loop water coolers, liquid cooling is becoming more and more prevalent in high-end desktop PCs. Servers are getting in on the action, too, although you won't find them using the same cooling gear. IBM developed a custom liquid cooling system for SuperMUC, Europe's most powerful supercomputer.
Cooling one or two components inside a PC is nothing compared to chilling a supercomputer with 18,000 Xeon processors. The accompanying 324TB of memory is also liquid-cooled. IBM refers to the cooling system as a "hot-water" design that can operate without the aid of a chiller or compressor. The water circulating through the plumbing gets up to 45°C, and it's pumped through an exchanger that provides heat to the buildings surrounding the Leibniz Supercomputing Center where SuperMUC is housed.
According to IBM, using water cooling in this environment results in a 40% reduction in energy consumption versus a traditional air cooling. That's good for savings of up to a million euros a year. Tree-huggers should be happy, too. SuperMUC is located in Germany, where all government-funded institutions are required to get their energy from sustainable sources.
IBM has been working with water cooling for decades, and it has interesting plans for the future. This presentation (PDF) on SuperMUC teases the potential of liquid cooling tucked between the layers of a 3D semiconductor. Keeping the liquid away from electrical interconnects is going to be challenging, but IBM already has a "test vehicle" for the technology.
|Samsung's DDR4 modules for servers have quadruple-stacked memory dies||30|
|Apple sets date for expected iPhone 6 reveal||1|
|This 8'' Windows 8.1 tablet will cost only $149||29|
|Amazon sale discounts hundreds of downloadable PC games||45|
|Wednesday Evening Shortbread||43|
|Asus shows glimpse of ZenWatch; Apple 'wearable' coming Sept 9||24|
|Zotac's ''Pico'' PC runs Windows, slips into a pocket||78|
|Dropbox Pro now offers 1TB of storage for $9.99 a month||40|
|Now we can lose our data 8TB at a time.||+44|