14,600 NVIDIA Tesla GPUs Power Oak Ridge Titan Supercomputer

While Intel is celebrating the Stampede HPC system and the role played by Xeon Phi in its creation, NVIDIA has its own reason to be giddy: the soon-to-be fastest supercomputer of them all.

Some may have already heard about Titan, but for those who haven't, Oak Ridge National Laboratory (ORNL) is currently building it.

We've now learned, from a report, that there will be 14,592 Tesla K20 graphics processing units involved in the building of the HPC conglomerate.

While 16-core AMD Opteron CPUs do the regular computing, NVIDI's GPU accelerators do the parallel processing tasks.

The Titan will have 18,688 nodes, each with a 32 GB of memory. The cumulative floating-point performance should end up at above 20 Petaflops (Sequoia, the current leader, offers 2.01327).

So far, ORNL has received only 32 Kepler-based Tesla K2 units. 1000 more should be delivered this week, while Titan itself will be completed by March 2013.