Intel Parallel Computing Centre
15 Feb 2017
Yes
-  

 

 

The Intel Parallel Computing Centre (IPCC) works across a range of disciplines to future proof software for the scientific community.

No

​​

 

To meet the increasing demands of the scientific community, HPC systems and architectures are becoming increasingly parallel and complex. To equip future HPC systems, we need to modernise the existing scientific code base. 

The Intel Parallel Computing Centre is at the forefront of this by parallelising current software and providing training for the scientific community, enabling them to take advantage of Hartree Centre new HPC machine, Scafell Pike (based on Intel Xeon Skylake and Intel Xeon Phi Knight's Landing). Parallelisation of software enables organisations to use modern Intel architectures and brings benefits such as speeding up the scientific workflow and future -proofing software.

The IPCC works across a range of disciplines to future proof software for the scientific community.

​​Weather and climate modelling with PSyclone ​

This code generation tool has been developed in the Hartree Centre to support the GungHo project, a collaboration between the Met Office, Natural Environment Research Council (NERC) and the Hartree Centre. The project combines expertise in meteorology and computational science to develop a next generation weather and climate computer model. The model is used at the heart of the Met Office's weather forecasting, taking current and past state of the weather and simulating how it may evolve over the next few hours, days and months. 

A key design principle adopted by PSyclone is the 'separation of concerns' where elements of the software dealing with natural science is kept apart from those dealing with computational science. It adapts the performance to suit the supercomputer it is running on, improving productivity by removing the need to hand code software for a specific supercomputer. The PSyclone tool enables users to apply different transformations to their code rather than re-writing, thereby reducing error and saving time and money by streamlining the process. 

weather.jpg

​​​​​Task-based Parallelism to improve DL_POLY

​Encompassing over 25 years of HPC and molecular modelling expertise the DL_POLY project delivers a unique but generic solution that can suit the needs of many modelling communities with access to molecular statics, mechanics (engineering/physics), dynamics(chemistry, organic/bio-chemistry) and Dissipative Particle Dynamics (meso-scale dynamics, chemical engineering). A thriving community is behind DL_POLY with users all over the world with more than 3000 new downloads in the last 5 years. It is the most popular molecular CCP5 badged simulation software in the UK produced and distributed by Daresbury Laboratory. 


The IPCC has been helping optimizing DL_POLY by leveraging the new OpenMP Tasks functionality. This allows DL_POLY to move away from the fork-join parallelism model and adopt a more flexible parallelisation strategy that can dynamically adapt to the computation imbalances inherent of the science modelled. The IPCC has also been looking at the vectorization and parallel IO aspects of the code.​

IBM iDataplex Iden_jpg.jpg

​Optimising multiple UK community codes

​Our IPCC has optimized multiple codes used by the scientific community. This includes:
  • DualSPHysics:  a smoothed particle hydrodynamics numerical model developed to study free-surface flows where Eulerian methods can be difficult to apply.
  • DL_MESO DPD and LBE: a general-purpose mesoscale simulation package.
  • Castep: code for calculating the properties of materials from first principles.
  • ParaFEM: a code for parallel Finite Element Method.

We have also deployed the OSPRay (CPU-based Software Defined Visualization) software stack to improve the Hartree Visualization Suite and enable in-situ visualization from our compute cluster to our Visualization facilities.

Bull Sequana Scafell Pike 2_jpg.jpg

​Training​ and community engagement​

​The Hartree Centre in collaboration with Intel organizes yearly 3-day OpenMP and Vectorization Training Courses to help our community of users to leverage the Intel products as well as Intel's Performance Profiling tools. Additionally, the Hartree Intel PCC organizes Parallel Software Workshops to share the best practices with the community. 

Hartree also offers free access for the academic community to perform proof-of-concept and performance analysis tasks on our Intel processors.​


DL14-061- (36) (1448x1900).jpg


Contact us​

If you have a project idea in mind or are interested in finding out more about one of our projects, please contact us​.​


Contact: