Will Data-storage Processing Unit be another GPU?



Opportunities for undertaking AI and ADAS businesses are limited, then what will be another rising industry? The answer may be Data-storage Processing Unit. But will DPU be as essential as GPU?

According to Jiweinet.com, even though the AI and ADAS industry are seemingly hot and popular, their opportunities are literally limited. From certain practitioners’ point of view, if the amount of staff of a company hasn’t reached at least 50, it is hard for them to compete with other leading companies considering its size of capital, talents and technology. If opportunities for undertaking AI and ADAS businesses are limited, then what will be another rising industry? The answer may be Data-storage Processing Unit. But will DPU be as essential as GPU?

Data Control chips in enterprise level is innovated

As data has increased dramatically, the increasing amount of data enlarges the request for storage. Hard Disk Drive used to be the most popular product in the market, yet nowadays SSD has been better on the aspects of consumption, noise and speed. Therefore, HDD has gradually transited to SSD.

On the Demo Day of The Fifth Arm Accelerator Camp, Mr. Yang yafei, CEO of Shenzhen Dapu Microelectronics Corporation, pointed out that the convert of consumer market has transmitted a better half but the transmission of market at enterprise level has just began. Take PCIe SSD as an example, its compound growth rate will reach 74.35% in 2020 and its volume is going to be more than 17million.

Transmission of market at enterprise level has required for the storage and speed of storages. The market is mainly dominated by leading companies such as Samsung, Intel and Western Digital. In the first season of 2018, Samsung holds 39.1% of market share, ranking the first and followed by Intel with 17.7%. Micron holds 10.9% market share. Main components of SSD are also dominated by foreign companies and factories.

When it comes to the era of AI and big data, the previous system and requirement is totally different.

Overcome the shortcoming of data transmission

 “Flash memory can be compared to a data warehouse and controller is the administrator of warehouse. As the technology progressed, speed of controller has been tens of times faster than fifteen years ago and the capacity also enlarged hundreds of times. However, the bandwidth of data transmission only increased several times which has been a large bottleneck,” said Mr. Yang.

How to speed up data processing controller is the next key point. Mr. Yang has mentioned that even though controller is of traditional CPU framework and can be accelerated by FPGA and AI etc. However, calculation, storage and Intelligence will mixed together in the future and there will be opportunities for specialized storage controller chips.

Therefore, Shenzhen Dapu Microelectronics Corporation has innovated the controller of SSD and developed DPU which specialized in data processing. It abandons traditional CPU structure and integrates machine learning, big data analysis, intelligent storage and data security together.

“There are mainly two biggest advantages of DPU. First of all, it is that it reduces the frequency of reading and wiping which prolongs the life of SSD. What’s more, it relieves the stress of CPU. After several times of tests, it prolongs the life of flash memory at three to five times,” emphasized by Mr. Yang.

At the same time, Companies including Samsung and Western Digital is negotiating with DAPU and is willing to achieve relevant patents from it by technology licensing.

Will it be another GPU?

As a chip related product, it is important to develop steadily.

Mr. Yang implied, “DPU is at the early stage so it will not develop in big steps. We are going to profit by technology licensing in consumer areas and offer DPU to corporations. Our priority is make fundamental and all-purpose calculation. DPU has big business scenario and will continuously optimize it by developing machine learning and deep neutral network technology.”

It is said that DPU mainly uses FPGA and may adopt arm core in order to cut the cost. DAPU plans to massively produce DPU in the future.

 “The volume of data will be bigger and bigger and doubles every 18 months. DPU will offer operating system with technological support in the future in order to share the workload with CPU.” Mr. Yang has already worked out the future plan.

The developing route is somewhat similar with GPU. Mr. Yang said that NVIDIA developed GPU several years ago when CPU wasn’t able to meet the need of display screen. NVIDIA was continuously developing GPU even though it is not the market’s favorite. Finally it became one of the most important devices in the AI era but will DPU be another GPU?