Skip to main content

Titan Supercomputer Packs 46,645,248 Nvidia CUDA Cores

Titan is based on Cray's XK7 system architecture boasting a heterogeneous environment consisting of CPUs and GPGPUs.

The computer is expected to deliver about 10 times the computing performance of ORNL's Jaguar supercomputer, which is currently listed as the world's sixth fastest supercomputer with a maximum sustained performance of 1.94 PFlops. Titan would top the Top500 list at this time as it is significantly more capable as the currently leading system - Lawrence Livermore National Laboratory's BlueGene/Q-based Sequoia supercomputer (16.3 PFlops).

Titan has 18,688 nodes, with each holding a 16-core AMD Opteron 6274 processor and an Nvidia Tesla K20 graphics processing unit. That translates to 299,008 CPU cores, as well as 46,645,248 CUDA cores, 91.25 TB ECC GDDR5 memory, and 584 TB registered ECC DDR3 memory. Each node integrates a total of 38 GB of memory, translating to 710,144 GB or 693.5 TB of total system memory. The entire installation requires 4,352 square feet of floor space.

What makes Titan especially impressive is when we remember that the first exaflop supercomputer is promised for 2020. For the next seven years, supercomputing performance will have to climb by a factor of 50 over Titan and there is reason to believe that the industry can achieve this goal. Over the past seven years, supercomputer performance jumped by almost 120x. In June 2005, the world's fastest supercomputer was Lawrence Livermore National Laboratory's BlueGene/L with 65,536 cores, which were good for 136.8 TFlops.

Contact Us for News Tips, Corrections and Feedback