A stealth startup of former Google engineers is engaged on a synthetic intelligence chip that might outshine their former employer’s much-vaunted TensorFlow tech. Dubbed Groq, the clandestine firm has shared little of its plans or roadmap, although latest filings recommend greater than $10m in funding has already been secured. Greater than four-fifths of the unique Google TPU workforce at the moment are working there.
The TPU, or Tensor Processing Unit, is Google’s homegrown AI chip. The corporate started engaged on it a number of years in the past, within the hope of constructing neural networking far more sensible at scale. Whereas machine studying isn’t new – Intel is engaged on enabling it with CPUs, whereas NVIDIA is coming at it from the other way, with GPUs, amongst others – it’s computationally very intensive.
That’s an issue while you’re operating an operation on the dimensions that Google is. Neural networking and synthetic intelligence are seen as important to enhancing every part from search outcome accuracy, to computational picture processing, by to spam and malware detection. AI additionally has a house within the Google Assistant, which is at coronary heart of standalone units like Google Dwelling, but additionally options on Android smartphones and extra.
The outcome was the TPU, powering TensorFlow, a scalable system that opens neural networking as much as everybody. It was developed partially by Jonathan Ross, a former Google engineer who’s now working with investor Chamath Palihapitiya on the startup identified in the present day to be Groq. Additionally on the roster are a claimed eight of the primary ten individuals from the TPU challenge.
“It’s too early to speak specifics,” Palihapitiya instructed CBNC, “however we predict what they’re constructing might turn into a basic constructing block for the following technology of computing.” Two SEC filings from final 12 months confirmed each the identify of the corporate and its preliminary funding. In keeping with these paperwork – which identify Palihapitiya, Ross, and former Google X engineer Douglas Wightman – Groq raised $10.3m.
Whereas taking over Google – to not point out Intel, AMD, and NVIDIA, amongst others – in growing a chip is a large problem, the potential upshot if Groq’s expertise proves profitable is big. Earlier this 12 months, on a report into the TPU’s efficiency in real-world trials, Google claimed it had proved 10x extra environment friendly at processing than utilizing GPUs. That has implications for energy consumption, chip value, and extra.
NOW READ: Google has big plans for its Tensor neural networking tech
How far we’re from seeing the fruits of Groq’s labors is unclear, and it appears possible that, if the corporate’s schemes are profitable, they’ll be a tempting goal for acquisition by a few of the greater names in chip improvement. AI is massive enterprise for the time being: solely this week, Amazon opened up the substitute intelligence that powers its Alexa assistant to third-party builders and gadget producers. ARM is positioning its low-power cores as splendid for AI, and Bosch and NVIDIA are doing one thing related for the brains of self-driving autos.