December 3, 2021
AI

Cerebras Completes Collection F Funding, One other $250M for $4B Valuation


Each occasionally, a startup comes together with one thing out of left area. Within the AI {hardware} technology, Cerebras holds that title, with their Wafer Scale Engine. The second technology product, constructed on TSMC 7nm, is a full wafer packed to the brim with cores, reminiscence, and efficiency. By utilizing patented manufacturing and packaging strategies, a Cerebras CS-2 encompasses a single chip, larger than your head, with 2.6 trillion transistors. The associated fee for a CS-2, with acceptable cooling, energy, and connectivity, is ‘just a few million’ we’re instructed, and Cerebras has clients that embrace analysis, oil and gasoline, prescription drugs, and protection – all after the distinctive proposition {that a} wafer scale AI engine offers. Right this moment’s information is that Cerebras continues to be in full startup mode, ending a Collection F funding spherical.

The brand new Collection F funding spherical nets the corporate one other $250m in capital, bringing the full raised by way of enterprise capital as much as $720 million. In talking to Cerebras forward of this announcement, we had been instructed that this $250 million was for successfully 6% of the corporate, bringing the valuation of Cerebras to $4 billion. In comparison with Cerebras’ final Collection E funding spherical in 2019, the place the corporate was valued at $2.4 billion, we’re taking a look at about $800m additional worth yr on yr. This spherical of funding was led by Alpha Wave Ventures, a partnership between Falcon Edge and Chimera, who’re becoming a member of Cerebras’ different buyers reminiscent of Altimeter, Benchmark, Coatue, Eclipse, Moore, and VY.

Cerebras defined to me that it’s greatest to get a funding spherical out of the way in which earlier than you really need it: we had been instructed that they already had the following 2-3 years funded and deliberate, and this extra funding spherical offers some extra on high of that, permitting the corporate to additionally develop as required. This encompasses not solely the following generations of wafer scale (apparently a 5nm tape-out is round $20m), but in addition the brand new reminiscence scale-out techniques Cerebras introduced earlier this yr. At present Cerebras has round 400 staff throughout 4 websites (Sunnyvale, Toronto, Tokyo, San Diego), and is trying to increase to 600 by the tip of 2022, focusing so much on engineers and full stack improvement.















Cerebras Wafer Scale
AnandTech Wafer Scale

Engine Gen1
Wafer Scale

Engine Gen2
Enhance
AI Cores 400,000 850,000 2.13x
Manufacturing TSMC 16nm TSMC 7nm
Launch Date August 2019 Q3 2021
Die Dimension 46225 mm2 46225 mm2
Transistors 1200 billion 2600 billion 2.17x
(Density) 25.96 mTr/mm2 56.246 mTr/mm2 2.17x
On-board SRAM 18 GB 40 GB 2.22x
Reminiscence Bandwidth 9 PB/s 20 PB/s 2.22x
Cloth Bandwidth 100 Pb/s 220 Pb/s 2.22x
Price $2 million+ arm+leg

Up to now Cerebras’ clients have been, within the firm’s personal phrases, from markets which have historically understood HPC and are trying into the boundary between HPC and AI. This implies conventional supercomputer websites, reminiscent of Argonne, Lawrence Livermore, and PSC, but in addition business enterprises which have historically relied on heavy compute reminiscent of prescription drugs (AstraZeneca, GSK), medical, and oil and gasoline. A part of Cerebras roadmap is to increase past these ‘conventional’ HPC clients and introduce the know-how in different areas, such because the cloud – Cirrascale lately introduced a cloud providing based mostly on the newest CS-2.

Developing quickly is the annual Supercomputing convention, the place extra clients and deployments are more likely to be introduced.

Associated Studying

Leave a Reply

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