A brand new AI startup has introduced what it says is the biggest semiconductor chip ever constructed.

Cerebras Techniques has taken the wraps off its Wafer Scale Engine (WSE), which packs in 1.2 trillion transistors, as a brand new form of laptop optimised solely for AI and deep studying features.

This places it a long way forward of different chips at the moment in the marketplace, with contemporary launches from different chip leaders akin to AMD and Nvidia packing in between 20 and 30 billion transistors.

  • AMD Processors: the most productive AMD CPUs in 2019
  • What’s AI? The whole thing you wish to have to understand
  • Is AI the brand new personalized touch?

Huge

Cerebras says that its {hardware} will get advantages AI processing features, that are continuously held again via a loss of obtainable compute capability.

The corporate says trying out a unmarried new hypothtesis or coaching a brand new style is each dear and time-heavy, continuously taking weeks and even months on currently-available methods. 

Cerebras believes its new chip can lend a hand meet the “monumental computational calls for” of deep studying, being purpose-built to permit for higher computation than every other {hardware} in the marketplace nowadays. 

Not like many different high-powered chips in the marketplace nowadays, that are in truth a choice of chips piled on best of a silicon wafer, the WSE is a unmarried chip on a unmarried wafer, taking into consideration a lot sooner switch speeds and processing capability.

The WSE is greater than 56X higher than the biggest graphics processing unit, containing 3,000X extra on-chip reminiscence and in a position to reaching greater than 10,000X the reminiscence bandwidth, with high-speed reminiscence with regards to every core making sure that cores are at all times occupied doing calculations. 

Throughout its 46,225 sq. millimeters of silicon, (making it relatively smaller than a standard pill) the chip is in a position to ship 400,000 AI-optimised cores, and its specialized reminiscence structure guarantees every of those cores operates at most potency. 

The WSE supplies 18 gigabytes of rapid, on-chip reminiscence allotted a number of the cores in a single-level reminiscence hierarchy, one clock cycle clear of every core. 

There is not any information but on when the {hardware} will probably be obtainable for wider use, with Cerebras merely pronouncing updates will probably be coming quickly.

  • The most efficient laptops for programming in 2019