Decoding Days Smart Chip: Chip Start-Ups Break Through The 10 Billion GPU Market

After more than two years of research and development, the GPU chip has entered the stage of being embedded in GPU for more than seven years.

The chip itself is made up of hundreds of millions of transistors. For example, tiantianzhixin's chip contains 24 billion transistors. The current inside the transistor flows from the beginning to the end through a gate, which is the width of the gate in terms of nanounits in the chip. 7 nm is the limit of semiconductor technology. The design of GPGPU chip in this process is still monopolized by AMD and NVIDIA.

"We dare to try this. On the one hand, we have excellent teams and have made successful chips. On the other hand, the market space is here, but there are few players. We hope to win the share as a local enterprise. " In August this year, Cai quangen, chairman of days smart chip, said in an interview with 21st century economic reporter.

The full Chinese name of GPU is general purpose graphics processor (GPU), which is a kind of general purpose computing task that is originally processed by central processing unit (CPU).

According to relevant industry data, the market scale of GPGPU chip cards in China has reached 8.6 billion yuan in 2019, and it is expected to reach 45.8 billion yuan in 2025, that is, the annual compound growth rate is about 32%.

Power of assists

Tiantianzhixin was established in Nanjing in 2015. At the beginning of its establishment, it mainly built AI oriented computing power acceleration software and hardware products. In 2018, the company formally established a chip R & D team and began to develop general, standard and high-performance AI computing chips.

The three basic conditions for the development of artificial intelligence are computational power, algorithm and data. The performance of AI computing chip directly affects the computational power. From a computational point of view, the number of transistors on the chip and the chip area directly determine the computational power. Days smart chip products are mainly aimed at the high-performance computing market represented by artificial intelligence.

We will focus on the computing core and software of the era to solve the problems of the core computing chip and the computing team.

In August this year, the State Council promulgated several policies to promote the high-quality development of integrated circuit industry and software industry in the new period, which puts forward some requirements for the high-quality development of China's integrated circuit industry and software industry.

"Looking at the whole IT system, from CPU, operating system, office suite, whole machine to server, we have some commercial alternative products, but in the field of GPGPU, it is still a blank." According to Zheng Jinshan, chief scientist of tiantianzhixin, the above-mentioned policies have pointed out the direction for the development of China's integrated circuit industry and software industry, and he also hopes that GPGPU can develop better under the policy direction.

Group attack

With the further promotion of domestic digital economy, especially the development of 5g technology, intelligent factory, unmanned driving, smart city, intelligent security and other applications are continuously increasing the market demand for GPGPU.

According to the prediction of GPGPU market in 2025 according to the above report, the market scale of artificial intelligence training application scenario in China will be 14.4 billion yuan, and that of artificial intelligence reasoning application scenario will be 28.6 billion yuan.

The unavoidable situation is that more than 90% of China's GPU market is still in the hands of international leading enterprises.

"We hope to become a cloud based GPGPU chip with independent control and world-class standard," Cai quangen said in an interview

It is understood that tiantianzhixin has set up a chip R & D team since 2018, which has attracted many senior R & D personnel from well-known chip companies AMD and NVIDIA, as well as senior R & D personnel from Oracle, IBM and other companies, which are the leaders of enterprise software.

Zheng Jinshan, the chief scientist of the company, led the Shanghai chip design team in AMD and participated in the development of several GPU chips. In 2020, Cai quangen, who was the general manager of Shanghai Unicom, became the chairman of days smart chip.

"Any new product development requires a process of customer and market awareness. Our overall marketing strategy is ecological marketing, including effective cooperation with relevant large enterprises. " When it comes to product strategy, Cai quangen revealed that in order to speed up the understanding of products by customers, the product testing process is carried out in cooperation with enterprises.

The broad market prospect of GPGPU also ushered in the attention of capital. In the round B financing in September 2019, Tian Tian Zhixin obtained hundreds of millions of yuan of investment led by Dazheng capital and Princeville capital. The following institutions include Shanghai Electric Hong Kong Co., Ltd., Bangsheng capital, etc.

"Chip is a technology intensive and capital intensive industry. We need the support of capital in our development process. We also welcome the participation of national and social capital with common development concept. " Cai quangen said.

Slow is fast

From the beginning of chip design, days smart chip cooperated with server manufacturers to define product design, so as to realize domestic substitution in data center in the future. It is reported that the company's partners in server manufacturers are Inspur and Xinhua 3, and its partner in independent and controllable supercomputing solutions is Zhongke Loongson, the leader of domestic CPU.

Behind the above ecological strategy are multi-level considerations. From the technical point of view, the communication from the design stage can make the products compatible with the global common products, and provide the possibility of seamless migration in the future product replacement process; from the perspective of talents, as the verification base of cutting-edge technology, the cooperation with universities and research institutes will not only help to track the development of technology, but also help to build the talent echelon of the company.

With the development of new digital infrastructure in China, edge computing will be applied in more fields. Gartner has defined 12 application scenarios of edge computing in an earlier report, and believes that more than 90% of enterprises will open their own unique applications in edge computing in the future.

21st century economic reporter learned that days smart chip has launched product development for edge computing applications. In addition to the 7-nanometer cloud training chip which has entered the streaming stage, the company's second chip is mainly aimed at the edge cloud, mainly reasoning.

"We don't ask for a lot of market share as soon as it comes out. We do science and technology should be honest to catch up, there is no shortcut in science. " Talking about the future, Cai quangen said, "the company has a long-term development strategy, but the first thing to do is to win more users' trust in us with products." (Editor: Lin Kun)