Getting Started with GPU Supercomputing
What is a GPU?
A Graphics Processing Unit (GPU) is a purpose-built hardware device originally designed for display rendering. The advancement of computer graphics, over the years, has driven the requirements for highly parallel processing structures which emphasize efficiency rather than general purpose applicability. The GPU was designed to do one thing – execute the same instructions over many pieces of data as fast as possible.
However, recent years have seen the emergence of exciting new technology allowing the utilization of the GPU for general purpose computations. The GPU has many cores (typically thousands today) designed to process large blocks of data in parallel. This contrasts with a typical Central Processing Unit (CPU), which has significantly fewer cores, each of which is more sophisticated and suited to working on independent instructions. The GPU is not practical for all types of computing, but for cases where the same instructions are to be carried out over large sets of data – like millions of particles – the GPU becomes a personal supercomputer.
More information about GPU supercomputing can be found here: http://www.nvidia.com/object/what-is-gpu-computing.html
What advantages does a GPU have over traditional parallel architectures?
While a GPU does not rival the world’s fastest parallel processing systems, it also doesn’t fill an entire room (or building in some cases), doesn’t require specialized cooling systems, doesn’t require an expensive high speed network and doesn’t require a team of personnel for management and utilization. A supercomputer in a box, GPUs are small and powerful, delivering acceleration to compute-intense applications, at a fraction of the cost of a parallel network. GPUs represent supercomputing for the masses.
Current High-Performance Computing (HPC) and gaming GPUs from NVIDIA®i contain well over 2,000 compute cores, fit and function in a standard workstation computer, and boast 1.3 TFlops of double-precision performanceii. Only ten years ago, a system with this sort of performance would rank in the top 40 of the most powerful supercomputers on the planetiii!
Here are a few specific advantages to consider:
- A GPU is a contained piece of hardware, which facilitates new installation or upgrades
- GPUs can be installed in most workstations
- There is no equivalent CPU upgrade for the price (approximately $0.77/GFlop based on NVIDIA GeForce® GTX Titan GPU)
- Low power consumption
- No complicated network required for cluster support, for GPUs installed in the same workstation
- Supercomputing contained within a single workstation – the compute power remains accessible to the end user
To learn about Barracuda VR on GPUs click here.
i NVIDIA, CUDA, GeForce, Quadro and Tesla are registered trademarks of NVIDIA Corporation. Barracuda and Barracuda VR are registered trademarks of CPFD Software, LLC. Windows is a registered trademark of Microsoft Corporation.
ii http://www.nvidia.com/object/why-choose-tesla.html, accessed September 6, 2013
iii http://www.top500.org/list/2003/06/, accessed September 6, 2013