GPUs, Xeon Phi, and other Architectures

Harnessing software and technology
Code Porting to GPUs, Xeon Phi and other Architectures
Performance and cost benefits

NAG can assess your code to identify the likely value of porting to NVIDIA or AMD GPUs, Intel Xeon (e.g., Skylake) or Xeon Phi (Knight's Landing), IBM POWER, or ARM family architectures. Where performance or cost benefits are possible, NAG can port your code to GPUs or other manycore processors using appropriate standards and languages - e.g. CUDA, OpenCL or OpenMP. We can also advise on appropriate alternative algorithms that may be better suited for use on manycore processors or GPUs.

Software Porting and Tuning

The NAG Software Porting and Tuning Service can help you to move your application to new hardware or software technology, whether a single processor or a multi-thousand node supercomputer. We can enable your application to exploit the potential cost/performance benefits of GPUs, manycore processors or other technologies. NAG expertise and experience has been proven through nearly five decades of developing and supporting high quality software products across a wide variety of systems, and through porting and tuning a range of user applications in our HPC services.

Your application running on new technology

NAG's expertise, proven processes and tools, work with you to reliably port your applications in a maintainable fashion to new processors, systems or software technology with reduced risk. NAG understands the technical issues at the heart of reliably porting, tuning and verifying software.

 

Technical Report Repository
Custom editions of NAG software

NAG has the infrastructure to port the NAG Library and customer specific numerical routines to your architecture of choice. We have developed a range of tools for checking and improving numerical code to make the porting exercise and future maintenance more reliable. As well as CPU implementations, appropriate NAG routines can be ported to GPU and coprocessor systems. We regularly tune our high-performance numerical library, by request, for specialist and custom computer systems.