Home PC News Cerebras puts ‘world’s largest computer chip’ in Lassen supercomputer

Cerebras puts ‘world’s largest computer chip’ in Lassen supercomputer

Lawrence Livermore National Laboratory (LLNL) stated it has built-in Cerebras Systems‘ new product, which the company claims is the world’s largest laptop chip, into its Lassen supercomputer for the National Nuclear Security Administration.

Technicians not too long ago related the Silicon Valley-based firm’s huge, 1.2 trillion transistor Wafer Scale Engine (WSE) chip — designed particularly for machine studying and AI purposes — to the 23-petaflop Lassen. The new wafer-sized laptop chip will assist speed up AI analysis at LLNL. The National Nuclear Security Administration exams the nation’s stockpiles of nuclear weapons utilizing the IBM and Nvidia-based Sierra system, to which Lassen is the unclassified companion. It is No. 14 on the newest Top 500 List of the world’s strongest supercomputers.

Normally, chip makers slice a wafer from a 12-inch diameter ingot of silicon to course of in a chip manufacturing facility. Once processed, the wafer is sliced into a whole bunch of separate chips that can be utilized in digital {hardware}.

But Cerebras, began by SeaMicro founder Andrew Feldman, takes that wafer and makes a single, huge chip out of it. Each piece of the chip, dubbed a core, is interconnected in a classy method to different cores. The interconnections are designed to maintain all of it performing at excessive speeds so the transistors all work collectively as one.

Cerebras WSE has 1.2 trillion transistors, the fundamental on-off digital switches which can be the constructing blocks of silicon chips. Intel’s first 4004 processor in 1971 had 2,300 transistors, and a current Advanced Micro Devices processor has 32 billion transistors.

LLNL stated the pc will allow researchers to research novel approaches to predictive modeling. Users gained entry to the system in July and have begun work on preliminary AI fashions.

Early purposes embrace fusion implosion experiments carried out on the National Ignition Facility, supplies science and speedy design of latest prescribed drugs for COVID-19, and most cancers remedy (via the Accelerating Therapeutic Opportunities in Medicine, or ATOM venture).

The CS-1 system runs on the WSE chip, which consists of 400,000 AI-optimized cores, 18 gigabytes of on-chip reminiscence, and 100 petabits per second of on-chip community bandwidth. The improve marks the primary time LLNL has a high-performance computing (HPC) useful resource that has included AI-specific {hardware}. Effectively it’s the world’s first laptop system designed for “cognitive simulation” (CogSim), a time period utilized by LLNL scientists to explain the mixture of conventional HPC simulations and AI methods.

Combining the CS-1 with Lassen permits LLNL to discover heterogeneity, the place the supercomputer consists of laptop parts with totally different specializations that every contribute to engaged on a given job. This would enable operations similar to information era and error correction to run concurrently, leading to a extra built-in, environment friendly, and cost-effective resolution to scientific issues, based on lab researchers.

Above: Cerebras packs 1.2 trillion transistors on a wafer.

Cerebras says the WSE processor is optimized for deep studying. It is 56 instances bigger than the biggest graphics processing unit (GPU) and comprises 78 instances extra compute cores. The WSE has 3,000 instances extra on-chip reminiscence and greater than 10,000 instances extra reminiscence bandwidth than GPUs.

LLNL researchers stated that their preliminary work focuses on studying from as much as 5 billion simulated laser implosion pictures to optimize fusion targets for experiments on the National Ignition Facility (NIF), with the aim of reaching excessive power output and strong fusion implosions for stockpile testing.

To guarantee profitable integration, LLNL and Cerebras are working in an AI heart of excellence within the coming years. Depending on the outcomes, LLNL may add extra CS-1 techniques, each to Lassen and to different supercomputing platforms.

Most Popular

Recent Comments