NVIDIA unveils Tesla K40 accelerator, groups with IBM on GPU-based mostly supercomputing
NVIDIA’s Tesla GPUs are already mainstays in supercomputers that want specialised processing energy, they usually’re turning into much more necessary now that the corporate is launching its first Tesla constructed for giant-scale tasks. The brand new K40 accelerator solely has 192 extra processing cores than its K20x ancestor (2,880, just like the GeForce GTX 780 Ti), nevertheless it crunches analytics and science numbers as much as forty % quicker. A leap to 12GB of RAM, in the meantime, helps it deal with knowledge units which might be twice as huge as earlier than. The K40 is already out there in servers from NVIDIA’s companions, and the College of Texas at Austin plans to make use of it in Maverick, a distant visualization supercomputer that ought to be up and operating by January.
As a part of the K40 rollout, NVIDIA has additionally revealed a partnership with IBM that ought to convey GPU-boosted supercomputing to enterprise-grade knowledge facilities. The 2 plan on bringing Tesla GPU help to IBM’s Power8-based mostly servers, together with each apps and improvement instruments. It isn’t clear when the deal will bear fruit, however do not be stunned if it turbocharges a company mainframe close to you.