Berkeley Lab, Intel to Collaborate in Updating Scientific Codes for Manycore Architectures

Lawrence Berkeley National Laboratory has been named an Intel Parallel Computing Center (IPCC), a collaboration with Intel aimed at adapting existing scientific applications to run on future supercomputers built with manycore processors. Such supercomputers will have millions of processor cores, but today’s applications aren’t designed to take advantage of this architecture.

Most scientific applications, such as those used to study climate change, combustion, astrophysics, materials, etc., are designed to run on parallel systems, meaning that the problem is divided into smaller tasks so more of the calculations can be done simultaneously to reduce the time to solution for the scientists. With the growing use of manycore processors, such as Intel’s Xeon Phi processors, which can have 60 or more cores in each processor, applications will need to have even more parallelism. Unless applications are modernized, they will not be able to take advantage of the greater computing performance promised by manycore processors.

The Berkeley Lab IPCC will be led by Nick Wright of the National Energy Research Scientific Computing Center (NERSC), and Bert de Jong and Hans Johansen of the Computational Research Division (CRD).

“Although manycore processors will significantly increase supercomputing performance, that’s only part of the equation,” said Wright, who leads NERSC’s Advanced Technologies Group. “To fully capitalize on this capability, we need to modernize the applications our user community uses to advance scientific discovery. Intel Parallel Computing Centers such as ours are helping to support the community to attack this problem.”

Optimizing applications for manycore is important for NERSC, which announced in April that its next-generation supercomputer will be a Cray XC supercomputer using Intel’s next-generation Xeon Phi manycore processor, which will have more than 60 cores. NERSC is working with its 5,000 users to help them adapt their codes to the new system, which is expected to be delivered in 2016.

The Berkeley Lab IPCC will focus on increasing the parallelism of two widely used applications: NWChem and CAM5, the Community Atmospheric Model. NWChem is a leading application for computational chemistry and CAM5, part of the Community Earth System Model, is widely used for studying global climate. Modernizing these codes to run on manycore architecture will enable the scientific community to pursue new frontiers in the fields of chemistry, materials and climate research. Because both NWChem and CAM5 are open source applications, any improvements made to them will be shared with the broader user community, maximizing the benefits of the project.

“Enabling NWChem to harness the full power of manycore processors allows our computational chemistry and materials community to accelerate scientific discovery, tackling more complex scientific problems and reducing the time researchers have to wait for simulations to complete,” says de Jong, who leads CRD’s Scientific Computing Group and is a lead developer of the NWChem software. “Advances made by our IPCC will be shared with the developer community, including lessons learned and making our code available as open source.”

The goal is to deliver enhanced versions of NWChem and CAM5 that at least double their overall performance on manycore machines of today. The research and development will be focused upon implementing greater amounts of parallelism in the codes, starting with simple modifications such as adding or modifying existing components and going as far as exploring new algorithmic approaches that can better exploit manycore architectures.

“The open-source scientific community truly depends on CAM components running effectively at NERSC. And climate scientists have always been early adopters of cutting-edge architectures,” says Johansen, a computational science researcher at Berkeley Lab. “With more performance and more parallelism, scientists can accelerate their simulations and more accurately represent atmospheric dynamics. This collaboration with Intel will help climate science developers leverage NERSC’s and Intel’s network of resources and manycore expertise.”

Berkeley Lab is an ideal collaborator for this project. The lab is home to NERSC, the U.S. Department of Energy’s most scientifically productive supercomputing center with more than 5,000 users running about 700 different applications. CRD is home to fundamental research programs in computer science, applied mathematics, and computational science where researchers investigate future directions in scientific computing and work to develop new tools and technologies to fully exploit the increasing power of supercomputers.

NERSC staff regularly conduct training and outreach programs to help users adapt to new systems, as is currently being done to familiarize users with NERSC’s new flagship supercomputer “Edison,” a Cray XC30 supercomputer that uses Intel Xeon “Ivybridge” multi-core processors.

Additionally, the work will be part of the NERSC’s Application Readiness program to help prepare users for the expected 2016 delivery of “Cori,” a Cray XC supercomputer designed using Intel’s next-generation Xeon Phi processor named “Knights Landing”, which will have more than 60 cores per processor. For the manycore-based Cori, the simplicity of the cores, the amount of memory per core and optional software-managed on-chip memory are expected to require new programming and performance tuning techniques, which will be developed in this collaboration with Intel and then shared with the rest of the NERSC community.

Berkeley Lab is the first Department of Energy laboratory to be named an IPCC. Other IPCCs are located at leading universities and research institutions around the world.