GPU supercomputer

A GPU supercomputer is a networked group of computers with multiple graphics processing units working as general-purpose GPUs (GPGPUs) in tandem on a single task.

GPU supercomputers enable faster processing of tasks because of the inherently parallel nature of GPUs. Because graphics cards are constructed for massive parallelism, they can dwarf the calculation rate of even the most powerful CPUs for many parallel processing tasks. While a CPU may have upwards of 12 cores, a GPU can have thousands of simplified shader cores. These same shader cores that allow multiple pixels to be rendered simultaneously can similarly process multiple streams of data at the same time. When tasks are recoded for GPU-based computation, it can enable the handling of huge workloads.

Heterogeneous CPU/GPGPU has been a significant development in high-performance computing (HPC). The combined use of CPU and GPU makes GPU supercomputers more effective by using all resources fully, improving not only performance but energy efficiency as well.

A previously CPU-based supercomputer, Jaguar, was upgraded with Nvidia GPUs to become Titan, a GPU supercomputer. Its performance went from 17.59 petaFLOPS to 27 petaFLOPS. The commercial Nvidia GTX Titan GPU is named for the supercomputer.

Have a look inside the Titan supercomputer: