SUNNYVALE, CALIF. – November 16, 2021 – Cerebras Systems, whose mission is to accelerate artificial intelligence (AI) compute, today announced at SC21 a new Cerebras Software Development Kit (SDK) designed to enable creation of new wafer-scale applications across computational fluid dynamics, molecular dynamics, signal processing, and more. The SDK is in Beta now and will be available year end.
The SDK will allow a wide range of AI and HPC developers to invent and test new ideas on the CS-2 system, and its Wafer Scale Engine (WSE-2), at a more flexible, lower level. The WSE-2 is the industry’s fastest and most powerful processor. With 850,000 cores, 40 Gigabytes of on-chip memory, 20 Petabytes/s of memory bandwidth, and 220 Petabits/s of fabric bandwidth, the WSE-2 has 123 times more cores, 1,000 times more on chip memory, 12,733 times more memory bandwidth, and 45,833 times more fabric bandwidth than the leading graphics processing unit. With the SDK, and a new domain-specific programming language based on C and familiar parallel programming concepts, the vast computational resources of the WSE-2 are put directly in the hands of the developer, limited only by their imaginations.
“We are excited to engage with the developer community, especially here at SC21, to help accelerate their work at the leading edge of HPC and AI,” said Andy Hock, Vice President of Product, Cerebras. “Powered by our 850,000 core, second generation wafer-scale engine with high bandwidth on-chip memory and interconnect, our CS-2 system is uniquely-suited to help users scale existing workloads and unlock new ones. We can’t wait to see what we can build together.”
Primitives will include a library of neural network kernels and a library of BLAS kernels. Using this, developers can program standalone applications, or they can combine new pieces with existing kernels in a modular fashion, using the same tools that engineers at Cerebras use for internal low-level development. The goal of the SDK is enable users to create things which have never been done before, to explore the boundaries of our industry, and to quickly and easily test these ideas on the WSE-2.
Also at SC21, Cerebras customer Argonne National Labs announced a new ALCF AI Testbed featuring the Cerebras CS-2 to support pioneering research at the intersection of AI and high performance compute (HPC). The National Energy Technology Laboratory (NETL) also shared details on work with Cerebras to accelerate both simulations for improved cybersecurity and computational fluid dynamics to model weather and earth systems dynamics, astrophysical simulations and electrodynamics.
Cerebras has expanded its footprint beyond the U.S. with new offices in Tokyo, Japan and Toronto, Canada. Its roster of customers has grown significantly to include Argonne National Laboratory, Lawrence Livermore National Laboratory, Pittsburgh Supercomputing Center (PSC) for its groundbreaking Neocortex AI supercomputer, EPCC, the supercomputing center at the University of Edinburgh, Tokyo Electron Device, GlaxoSmithKline, and AstraZeneca.
