Facebook has started the Open Compute Project (OCP), which will share new server and datacentre designs that are radically more efficient than current systems.
Chief executive Mark Zuckerberg said that as Facebook had grown, it had begun looking for better server and datacentre performance than it had been getting from current off-the-shelf products. For the last 18 months, engineers had been working on new designs that would fit large-scale computing needs.
"We want to share that knowledge with the industry and make server and datacentre design open," he said.
"We're trying to foster ecosystems for the development of business startups. It's really cool. We're not the only ones who need this hardware and by sharing there will be more demand for the stuff we need, which makes it cost effective."
Jonathan Heiliger, vice president of technical operations at Facebook, said that central to its strategy was power usage effectiveness (PUE), which is the ratio of power spent on computing versus that used to run and cool the facility.
The ideal PUE was a rating of 1.0 – meaning 100 per cent of power went to computing – but typically datacentres operated at a PUE of 1.5. Facebook's new datacentre in Prineville, Oregon operated at a PUE of 1.07, which is a major improvement.
"A typical datacentre consumes about $1m per MW each year, so this design would cut the annual power budget for an average site from $10m to $6m," said Rackspace's chairman Graham Weston.
"We had developing our own intellectual property around this issue, but will be flushing that to go with this open source design, because we believe in open source."
Dubbed Antlia 2, the ghost galaxy sits just 130,000 light-years away from the Milky Way
Delays to the roll-out of age verification for adult websites hasn't stopped government from considering extending them to more websites
Bluehole confirms rumours that Playstation 4 port is coming on 7 December
Atmospheric iodine works as a significant sink of tropospheric ozone, nullifying the harmful pollutant