It uses 372 Fermi GPUs and the expected CUDA-optimized applications

Jul 9, 2012 10:01 GMT  ·  By

Despite Intel’s Xeon-Phi x86 push, Nvidia’s CUDA is gaining more and more popularity. The American chip designer has just announced that UK’s Appleton Laboratory has powered up the country’s most powerful GPU supercomputer.

The new high-performance, GPU-accelerated supercomputer has been unveiled at the Center for Innovation in High Performance Computing (HPC) at the STFC Rutherford Appleton Laboratory in Didcot.

The 84-node cluster was baptized “Emerald” and uses 372 Nvidia Tesla M2090 accelerators.

Appleton Laboratory’s “Emerald” is not using the newer, dual-GPU K10 cards, but the more FP64 potent M2090 accelerators.

The M2090 cards are powered by the well-known Fermi architecture and deliver a very high 0.665 TFLOPs of peak double-precision (DP) floating point performance.

This is more than 300% of what a K10 accelerator is able to do in the FP64 applications, and likely more than half of Intel’s Xeon-Phi’s performance.

On the peak single-precision floating point performance, on the other hand, the Tesla M2090 accelerators are only able to deliver less than a third of K10’s 4.57 TFLOPs.

There is a great difference in FP64 performance between K10 and M2090, so Nvidia emphasizes that the Fermi cards are recommended for CFD, CAE, financial computing, computational chemistry and physics, data analytics, satellite imaging, weather modeling.

According to Nvidia, Kepler-based K10 cards excel at signal and image processing along with video analytics.

The “Emerald” supercomputer will accelerate research in astrophysics, bioinformatics, chemistry, engineering, genomics, life sciences, nanotechnology and physics.

Nvidia was also proud to announce on its official website that Oxford University has been named CUDA Center of Excellence (CCOE) for the all the work on parallel processing done there.

Thus, Oxford will utilize equipment and grants provided by Nvidia to support a number of research and academic programs across its mathematics, physical and life sciences divisions, including:

•Astrophysics – real-time pulsar detection application for the forthcoming Square Kilometre Array Project to deploy the world’s most powerful radio telescope

•Bioinformatics – analysis and statistical modeling of whole-genome sequencing data

•Chemistry – molecular dynamics simulations of key DNA nanotechnology mechanisms

Like we said in our professional GPU compute accelerators analysis here, there is a lot of work and research to be done in order to make full use of a GPU’s compute abilities.

It’s good to see that Nvidia’s CUDA is getting even more attention, and we certainly hope it will quash Intel’s Xeon-Phi GPU compute aspirations just to have good prices and competition on the market.