Comparing The Cerebras Wafer Scale Engine with the largest GPU — Credit: Cerebras website

Will this monster chip conquer the AI world?

Slightly bigger than an iPad, the semiconductor device will turbocharge the Neural networks of the Future

Over the years, the trend has been to pack more computing power in chips which got smaller with time. Most standard chips these days can sit on top of your fingertips. Most of these chips are manufactured by using a single silicon wafer, which is later cut to separate them from each other.

With the introduction of Artifical Intelligence and the need for super-fast processing & training of these futuristic systems, we need a new processor design that can accommodate the needs of these power-hungry & data-intensive applications.

A California-based chip developing startup Cerebras claims to have made a breakthrough in this regard with its innovate chip design dubbed as Wafer-Scale Engine” (WSE) & pronounced as Wise. WSE is the benchmark product designed to power & train the AI systems.

The severe limitation of the current chip infrastructure to train complex neural networks is seen as a major bottleneck towards innovating & solving problems. WSE expects to remove this hindrance to achieving industry-wide progress.

Specs & Comparison:

  • Cerbras has been funded well for the project with US$200 million coming from well-known Venture capitalists.
  • WSE is 57 times bigger than the leading AI chip “V100” from Nvidia. Just a little bigger than a standard iPad.
  • With a memory of 18 gigabytes, it dwarfs its Nvidia competitor by 3,000 times.
  • The mega AI chip has 400,000 processor cores — compare this to the most powerful CPUs have about having 30 cores & GPUs (graphics processing units), which are currently used in the AI systems for speech recognition, image processing and pattern matching having 5,000 cores.


The Industry trend for decades has been to create tinier chips which can be bundled together to create powerful processors. Cerebras believes amalgamating smaller chips creates latencies slowing down the super-fast AI systems. WSE provides a distinct advantage of processing complex machine learning algorithms in less time with lower power.

WSE provides 400,000 cores tightly linked together on a single silicon wafer providing great data-crunching capability with the ease of shifting between the processing and memory. This reduces the processing time of complex processes from months to minutes.

Also, it would be a physical space saver, considering there will only be one computer appliance to place on the rack instead of multiple processing chips.

“The new chip will spur the reinvention of artificial intelligence. It provides the parallel-processing speed that Google and others will need to build neural networks of unprecedented size.” ~ Cerebras CEO Andrew Feldman


There are a number of challenges that WSE needs to overcome before it can conquer the AI space. The first one is related to the manufacturing of such a large chip — if some impurity was to sneak into the silicon wafer being used to make the mega chip, the whole chip would need to be thrown away. Cerebras claims it has found innovative ways to avoid this problem. How this works out in mass production needs to be seen.

The second challenge is in regards to energy efficiency. The mega chip would have economic & environment implications with its power consumption. The chip that comes as part of a computer appliance will need a complex system of water-cooling to counter extreme heat produced with the chip running at 15 kilowatts of power.

“ It is hard to say just what kind of impact a company like Cerebras or its chips will have over the long term. That’s partly because their technology is essentially new — meaning that they have to find willing partners and developers, let alone customers to sign on for the ride.” ~ Charles King, principal analyst at Pund-IT

Cerebras is not the first company to produce chips powering AI systems. Google’s TPU (tensor processing unit) in 2016 & Huawei’s smartphone Kirin chips boasting of NPU (neural processing unit) in 2017 were other such endeavors. However, Not all these projects have really taken off. The success of WSE will depend on whether it can overcome its challenges & pave the way for mainstream adoption.

Stay informed with the content that matters — Join my mailing list

A devout futurist keeping a keen eye on the latest in Emerging Tech, Global Economy, Space, Science, Cryptocurrencies & more

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store