Semiconductor startup Groq, has announced its plans to release its first product sometime next year. With its roots from Google’s tensor processing unit team. According to the website, Groq’s first chip will run 400 trillion operations per second, which is much faster than Google’s most recent tensor processing unit.
By concentrating on making custom server chips for AI, the company could soon see itself taking on the likes of Nvidia that’s held the leading position in providing graphics chips for AI technology for quite some time. Groq received a $10.3 million investment from venture capitalist Chamath Palihapitiya earlier this year in which to help it achieve its goals of releasing its chip in less than two years of being in operation.
As well as taking on Nvidia, the company is looking also to compete with the likes of Google with its offering of silicon through the cloud and Intel with its new custom processor that’s about to be released.
In its defense, Nvidia announced earlier this week that most of these server manufacturers and cloud computing firms were, in fact, using graphics chips based on its very own Volta hardware.
Volta was first created out of the need to handle certain deep learning and machine learning software more efficiently. Inside these beastly graphics, chips are hundreds of unique tensor cores that have the ability to perform up to 120 trillion operations per second.
But, good performance comes at a price. According to Nvidia’s chief executive, Jensen Huang, the company spent around $3 billion on Volta’s architecture. If rival companies are to come anywhere near Nvidia they’ll need to be investing several million at the very least.
Other companies that may be on the horizon to do so include Wave Computing and Cerebras Systems. Wave Computing has invested $60 million into its hybrid-type chips while Cerebras Systems has now bumped its net worth up to $860 million, after receiving an investment of $112 million.
Chipmakers Graphcore are also scrambling to get to the pole position and with its recent $50 million investment from venture capital firm Sequoia Capital, it may just have a chance. Earlier this year the company divulged its plans to release an intelligence processing unit (IPU) that will be up to 100 times faster than Nvidia’s Pascal in completing certain tasks.
Source Electronic Design