While we here on the forums try to reach the highest 3DMark scores we can, the researchers at Oak Ridge National Laboratory (ORNL) have been working to build the world's fastest supercomputer for open scientific research. To achieve 20+ petaflops the researchers are using 18,688 nodes that each hold an NVIDIA GPU based on the Tesla K20 architecture and an AMD Opteron 6274 CPU with 16 cores, along with 710 terabytes of memory. Despite its name though, Titan is the same size as the supercomputer it is replacing, Jaguar, while being 10 times more powerful.
Titan represents more than just an achievement for computing power because of this hybrid CPU+GPU design, which many researchers are looking forward to take advantage of. In fact, it is only because of the addition of the GPUs that Titan can be as small as it is and as efficient. Titan is five times more energy efficient than Jaguar and to have simply upgraded Jaguar to the 20+ PFLOP mark would have required growing it to four times its current size. Needless to say, both ORNL and NVIDIA are proud of this technological accomplishment.
The ability to perform over 20,000,000,000,000,000 floating point operations per second (flops) may seem unnecessary to many of us who are happy with 5 GHz overclocks, but once Titan starts crunching numbers, it will likely never stop as more and more researchers request some of its time. Models for materials science, combustion, climate science, and nuclear energy are just some of what it will run both faster and more accurately than its predecessor.
Images courtesy of Oak Ridge National Laboratory.