In the technological race with the USA and in the face of sanctions, China seeks to create its own processors for artificial intelligence. DeepEdge chips are based on the RISC-V architecture and the outdated 14nm process.
The Chinese company Intellifusion introduced Deep Eyes artificial intelligence devices based on DeepEdge processors, which, according to the company's statement, are 90% cheaper than American counterparts . Tom's Hardware site informs about it.
In an effort to get high-end hardware and bypass US restrictions on supplying powerful processors to China, maker Intellifusion has unveiled its DeepEyes line of AI computing devices with a claimed performance of 48 TOPS (trillion operations per second) at 1,000 yuan, or about $140. Using the older 14nm process and possibly ASIC (mining devices for cryptocurrencies such as Bitcoin) technologies is another way for China to circumvent sanctions and remain competitive in the AI market.
The Deep Eyes 2024 H1 computer uses the DeepEdge10Max processor to achieve 48 TOPS in learning performance. The upcoming 2024 H2 Deep Eyes device will have a DeepEdge10Pro processor and half the performance — 24 TOPS. Coming next year, the 2025 H1 Deep Eyes box is targeting a significant performance boost thanks to the DeepEdge10Ultra processor with 96 TOPS.
Prices for the two future models are unknown. However, if the company can maintain the 1,000 yuan price tag in future chip generations, Intellifusion could achieve its stated goal of creating 90% cheaper AI hardware that covers 90% of possible use cases. All mentioned technologies and devices are made in China, including the NNP400T Intellifusion special neural network chip used (2+8-core 1.8GHz RISC processor, GPU up to 800MHz in DeepEdge 10).
To meet Microsoft's stated requirements for a computer to run artificial intelligence, modern PCs must have at least 40 TOPS of performance. So formally, Intellifusion devices can be suitable for many AI workloads. In comparison, Snapdragon's X Elite computer chips will have a performance of 40 TOPS.
As Dr. Chen Ning, Chairman of Intellifusion, pointed out, in the next three years, 80% of companies around the world will use large models of artificial intelligence. The training cost of a large model is tens of millions of yuan, and the price of basic universal training and work computers is usually one million yuan. Most companies cannot afford such costs.
According to Tom`s Hardware, the claim that 80% of companies worldwide will use artificial intelligence is very optimistic, but is fair in terms of the start-up cost for companies , who seek to effectively use artificial intelligence. DeepEdge chips use “independent and controllable home technology” and the RISC-V architecture.