Desktop supercomputing

Ordinary PCs could be used to solve complex engineering and science problems, ‘supercomputer- style’, by harnessing the power of graphics processing, a specialist in visual computing has claimed.

According to NVIDIA, the low-cost breakthrough allows its second-generation Tesla-10 platform to work with CUDA, a C-compiler that translates programs for parallel processing.

Used extensively in the high-performance computing (HPC) community, for applications such as modelling for the oil and gas industry, medical imaging or weather forecasting, CUDA allows complex algorithms to be computed using the parallel architecture of a computer’s graphics processor unit (GPU) rather than the central processor unit (CPU).

A key advantage of the technology is that where traditional platforms processing the same software do not scale up performance on a multicore platform, adding more GPUs can improve the performance linearly.

Andy Keane, general manager of Tesla Computing Products, said: ‘There are two different styles of computing. Calculating everything one after the other, and breaking a problem into thousands of pieces and computing them all at the same time — that is parallel computing.

‘Many different types of technology can do this, including the GPU processor. When it’s producing an image on the screen, it breaks it into a thousand pieces and processes it in parallel. What’s new is we use the same type of architecture to break a science problem down into pieces and compute them all in parallel.’

Keane explained that while some traditional computing applications such as spreadsheets and word processing do not work well with GPU processing, science applications are very well suited.

Applications that have components of sequential and serial processing, such as transcoding digital video from one format to another divide the work between the CPU and the GPU to give about 20 times the performance of just using the CPU alone.

A provider of 3D seismic analysis software and services to the global oil and gas industry, ffA, is using the platform to model potential reservoirs with seismic survey data. technical director of industry Steve Purves said: ‘Tesla will dramatically speed up our image processing routines and algorithms, and help us build a different type of software tool centred around the user.

‘The geological models will be more accurate, based on the latest data, and available sooner. This will lead to a quicker exploration cycle in the oil and gas industry. We will be able to integrate new seismic data immediately, so modelling will move out of the back room and closer to the drilling.

‘The oil and gas industry is being explored heavily now, and younger, more technically savvy users will be using technology like this to help find reserves.’

Other applications which could benefit from Tesla’s platform include molecular simulation to study diseases, computer chip design, financial risk analysis, computer animation, cancer diagnosis and drug design.

Berenice Baker