Titan Debuts with NVIDIA Tesla GPUs, Boasting Peak Performance of 20 Petaflops
Titan, touted as the world's first open-source supercomputer, made its debut at Oak Ridge National Laboratory (ORNL) in the United States. Titan is 20 times faster than Jaguar, its predecessor, as the former is capable of a peak performance of 20 petaflops - 20 million billion floating-point operations per second. This supercomputing system also marks the first deployment of the NVIDIA Tesla K20 GPU as Titan's peak performance is mainly due to the presence of 18,688 such GPUs.
Essentially, the Titan is a Cray XK7 supercomputer that is highly scalable in terms of petascale architecture. It comprises an array of XK7 blade servers and each server is powered by a pair of XK7 node that is essentially an AMD 16-core Opteron CPU and an NVIDIA Tesla K20 GPU. Each node is in communication with the other via Cray's Gemini interconnect.
Although Titan' peak performance is 20 times faster than the Jaguar supercomputer, it actually occupies roughly the same physical footprint as the Jaguar while consuming more power at nine megawatts, versus Jaguar's power requirements of seven megawatts. However, Titan is actually about five times more power efficient than its predecessor. In order to share the computing prowess of the Titan, the ORLC has made it an open-source supercomputing resource to the academic and research community with subscription invites for their members to leverage on its GP-GPU compute prowess. According to NVIDIA, response has been overwhelming and research work in fields like material science, nuclear energy and climate change will stand to benefit greatly from this initiative.