Significant increase in graphics processing for Advanced Research Computing

Our Advanced Research Computing (ARC) High Performance Computing service has recently increased its available Graphics Processing Unit (GPU) resources by purchasing five state-of-the-art servers, helping researchers at Oxford by greatly increasing capacity and performance.

What are Graphics Processing Units (GPUs)?

GPUs are a type of vector processor. First seen in arcade machines in the 1970s, they moved into PCs in the 1980s to enable quick and efficient drawing of graphics on screens. Recent technology developments have made them programmable which has opened up their processing power for general computing purposes. GPU resources are commonly used in machine learning and artificial intelligence research, but they are also used in a wide range of other computational problems such as molecular dynamics or computational finance.

The technical details

We’ve added five NVIDIA DGX-1V Max-Q servers with 40 state-of-the-art 32GB Volta GPUs to our existing offering of NVIDIA 16GB V100, P100, K80 and K40 GPUs.

The DGX-1V Max-Q server is a version of the NVIDIA DGX-1V designed for maximum power efficiency, using a lower clock speed version of the V100 GPU. Benchmarking results (here shown for GROMACS, a molecular dynamics code, and TensorFlow, a machine learning framework) show performance of the V100-LS GPUs in the Max-Q systems to be nearly on par with 'standard' 32GB V100 GPUs, and outperforming the 16GB V100 GPUs.

gromacs benchmarks graph

 

 
Tensorflow comparison diagram

 

More about ARC and how to get an account

All GPU resources are part of the ARC Arcus-HTC system and are accessible to the ARC user community. See our ARC website for information about ARC and how to obtain an ARC user account. 

For more information on using GPUs on ARC, including a detailed list of available resources, visit the IT Help website.