The continuing increase in microprocessor performance is no longer being driven by increases in clock speed, but by the exploitation of multicore architectures. GPUs make multicore computing accessible off-the-shelf. They take the specialist capabilities of traditional graphics cards and harness them in a more versatile way, to create completely new possibilities for scientific and engineering applications.

Through products such as the NVIDIA Tesla range of GPUs, the desktop supercomputer is now a reality, with a system rated at over 4 Teraflops available for £10K. Compared to traditional desktop computing, these systems:

• allow code to be speeded up often by a factor of 100 or more, depending on the extent to which the underlying algorithms can take advantage of the multicore architecture;
• allow the development of computational intensive code that needs to be tested in a more interactive manner;
• allow problems to be tackled that would simply have been out of reach previously.

NVIDIA’s standard programming model uses extensions to C and C++. It provides high levels of hardware abstraction, meaning that implementation focuses on the underlying algorithms and associated data structures.

We are offering companies the opportunity to take their initial steps into GPU computing using our programme of Industrial Mathematics Internships. An Internship provides an ideal means of creating new software demonstrators, or upgrading existing code, to exploit the benefits of GPU technology.

For more details, contact:

Dr Vera Hazelwood (07875-401570)

Dr Caroline Edwards (07917-155785)