What Can You Do With 400,000 SLACers?

Source: Slashdot What Can You Do With 400,000 SLACers?

Cerbras WSE Likes Big Core Counts

There is something strange going on at Hot Chips this year, brought to us by Andrew Feldman, the man that created and later sold Seamicro to AMD; the largest chip on the planet which goes by the name Cerbras.  The chip is an impressive 46,225 square millimeters and contains 400,000 Sparse Linear Algebra Cores, specifically designed for neural network computation.   How they managed to create a a single chip interconnected on a single wafer without impurities ruining at least some of those cores is a question we would like to see answered beyond the current reply of redundancy; hopefully we will be filled in at Hot Chips.

As for the numbers, you are looking at a total of 1.2 trillion transistors on the wafer/chip with 18GB of SRAM memory offering 9PB/s of bandwidth with communication handled by Swarm fabric capable of 100 petabits per second; not a familiar phrase even for HPC machines. The SLAC architecture they use is focused on AI and DEEP Learning, so measuring floating point operations is a little misleading, for instance Cerbras WSE doesn’t believe in processing zeros.

Learn more via the links at Slashdot, as well as from the erudite comments below.

The chip comes from a team headed by Andrew Feldman, who previously founded the micro-server company SeaMicro, which he sold to Advanced Micro Devices for $334 million. Sean Lie, cofounder and chief hardware architect at Cerebras Systems, will provide an overview of the Cerebras Wafer Scale Engine at Hot Chips.

Video News

About The Author

Jeremy Hellstrom

Call it K7M.com, AMDMB.com, or PC Perspective, Jeremy has been hanging out and then working with the gang here for years. Apart from the front page you might find him on the BOINC Forums or possibly the Fraggin' Frogs if he has the time.

Leave a reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Latest Podcasts

Archive & Timeline

Previous 12 months
Explore: All The Years!