Google joins Facebook’s Open Compute Project as web giant starts work on standardised racks for artificial intelligence, machine learning
Announced at the annual Open Compute Summit in San Jose, not only has Google joined the OCP but it is already working with Facebook on new hardware.
It was 2011 when Facebook took the lid off its server designs, and one-upped competitors who thought keeping their own data centre hardware secret would give them a competitive edge.
“It’s time to stop treating data centre design like Fight Club and demystify the way these things are built,” said Jonathan Heiliger at the time, who was then a vice president of operations at Facebook.
Slowly but surely, others joined the OCP, agreeing the mutual benefits to everyone in sharing data centre designs to advance computing in an age of massive scalability and artificial intelligence.
But up until this week, just Amazon and Google remained out of the OCP. Today, it’s just Amazon keeping tight lipped about its data centres as Google admits it now wants to help standardise designs for hardware such as GPUs for machine learning applications.
“The Open Compute community is an established collection of consumers and producers, and we see an opportunity to contribute our experience and expand the Open Rack specification,” wrote John Zipfel, Technical Program Manager at Google, on Google’s blog yesterday.
“Today’s launch is a first step in a larger effort. We think there are other areas of possible collaboration with OCP. We’ve recently begun engaging the industry to identify better disk solutions for cloud based applications. And we think that we can work with OCP to go even further, looking up the software stack to standardize server and networking management systems. We look forward to new and exciting advancements to come with the OCP community.”
Google and Facebook will work together on a new server rack, which they claim will be able to deliver up to four times more power to the rack’s servers, going from 12 volts to 48 volts.
“We kicked-off the development of 48V rack power distribution in 2010, as we found it was at least 30 percent more energy efficient and more cost effective in supporting these higher-performance systems,” wrote Zipfel.
“Our 48V architecture has since evolved and includes servers with 48V to point-of-load designs, and rack-level 48V Li-Ion UPS systems. Google has been designing and using 48V infrastructure at scale for several years, and we feel comfortable with the robustness of the design and its reliability.
“As the industry’s working to solve these same problems and dealing with higher-power workloads, such as GPUs for machine learning, it makes sense to standardize this new design by working with OCP.”