Home > News content

Google TPU project core members left, they set up a new company to build "next-generation AI chip"

via:博客园     time:2017/4/26 17:30:33     readed:1524

谷歌谷歌 TPU 项目核心成员离职了,他们想建“下一代 AI 芯片”

April 6, Google for the first time revealed its TPU project details and test results. In the case of Intel and Nvidia chip comparison, Google machine learning chip TPU processing speed faster than the GPU and CPU 15-30 times faster, and in energy efficiency, TPU is to enhance the 30-80 times. The results of a publication suddenly sparked hot. And Google employees left the job under the boom, Google TPU project was also "hit" and "8" Google employees have left the project, the establishment of Grop start-up companies, they want to establish the next generation of AI chip & rdquo ;.

According to CNBC, several of the leading members of Google's early TPU project left each other and created a new company with a venture capitalist, Chamath Palihapitiya, a Silicon Valley investor.

According to the known news that the company is named Groq. There is no relevant promotional materials or website outflow, the network can only find a few last October, December SEC file. The SEC file indicates that the company has raised $ 10.3 million and was incorporated in Delaware on September 12, 2016.

"We are very excited about the establishment of Groq," said Palihapitiya in an e-mail. "It's too early to talk about the details, but we think the content they're building may be the basis for the next generation of computing." & Rdquo;

Groq's SEC file shows that the company has three executives: Jonathan Ross, Douglas Wightman, Palihapitiya (founder of the investment firm Social Capital). The company's address listed by the SEC is shown as the headquarters of Social Capital.

谷歌谷歌 TPU 项目核心成员离职了,他们想建“下一代 AI 芯片”

From 2013 to 2015, Ross worked as a Google hardware engineer in Wisconsin (one of the main hotbeds of TPU), one of the leading developers of the TPU project. And Wightman was the engineer of the Google X Labs' moonshot factory, which is engaged in a number of future projects such as Project Loon (with a solar balloon connection network). Prior to joining Google, he was a co-founder of four companies.

The company (which is now Groq) now has eight of the top ten people in Google's early TPU team. According to Palihapitiya, he was starting to hear about the Google TPU project two and a half years ago. At that time they knew they had to build AI chips. "I can not help but think, what happened, so that Google and Intel began to chip competition? & Rdquo;

Palihapitiya says all start-ups are tough, especially for new chip companies. Because they need R & D costs are too high, there are arrogant challenges from mining equipment manufacturers and to bear the risk of unproven technology. In addition, but also suffer from the weight of existing chip giants. Intel, Qualcomm and Nvidia in the chip research and development strength has been very strong, and Google, Apple and Amazon are also developing their own chips.

In an interview last month, Palihapitiya said: investing in this project is simply taking risks. However, the next wave of chip innovation "can make such companies like Facebook, Amazon, Tesla, get the government in the machine before learning there is no support and help."

According to Ross, he said in his LinkedIn state that he had left Google in September last year and is currently in a state of "unemployment". And Wightman also said he was also leaving Google in the same period. Although there is no explanation where to go, but it has confirmed this investment by e-mail.

Earlier in April of this year, Google revealed the details and test results of the TPU's chip project for the first time in a 17-page report entitled "TPU Performance Analysis" in the data center. Jouppi wrote in the blog, using the TPU's AI workload, running faster than modern processors 15 to 30 times, while the efficiency is 30 to 80 times higher. The study compared the TPU to Intel and Nvidia's chips.

Ross is one of the 75 authors of the report. He also lists the inventors of the four patents, which are related to neural network processors and computations, which we can see as a computing system designed to simulate the brain. Ross in his LinkedIn release dynamic said that as the project 1/5, he and another engineer with the development of the TPU project.

With the continued fever of the TPU project, Grop may or may further expand the team. And Google TPU project will have members have left the Grop, and let us continue to track.

& Middot; Note: The picture comes from the network


China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments