What is NVIDIA's layout of artificial intelligence behind the acquisition of Mellanox by NVIDIA?

By NVE Corporation

On the evening of March 11, Beijing time, Nvidia founder and CEO Huang Renxun issued an internal mail, announcing the acquisition of Israeli chip maker Mellanox Technologies for $6.9 billion. This is the largest acquisition by NVIDIA since its inception, with the aim of accelerating the data center chip business, one of the most successful projects initiated by CEO Huang Renxun. Mellanox is a technology and equipment provider that provides high-performance, point-to-point interconnect solutions for data center servers and storage systems, producing cloud-enabled data center chips and hardware components, including Ethernet switches, chips and InfiniBand.

Why did AI Core World Yingwei Weida acquire Mellanox?


MellanoxTechnologies is an Israeli multinational computer networking product provider using InfiniBand and Ethernet technology. Some insiders commented that in Israel, Mellanox's status is equivalent to Huawei and Ali in China.

Mellanox's core technology "RDMA", which produces cables and switches that transfer data between servers, storage systems and infrastructure equipment, is an important technology for AI data storage and transmission.

This technology will complement NVIDIA's graphics processors, and although they originated in visual effects and computer games, this technology will increasingly be used in machine learning systems.

The combined entity will serve every major provider of cloud computing services and provide more than half of the world's fastest supercomputers, including Dell, Google Owners Alphabet and IBM.

Behind Nvidia’s shot

1 NVIDIA's business is relatively simple. In addition to the traditional game hardware business, Nvidia has become the first artificial intelligence chip by deep learning this wave of artificial intelligence in recent years. Its GPU products have become the mainstream of deep learning in recent years. The architecture, and adoption by a large number of users, including cloud computing giants, has made it an important location in the data center market.

As the absolute king of the data center InfiniBand network, Mellanox has a good reputation in the field of data center network equipment and chips. For example, high-performance computing, more than half of the world's Top500 supercomputers use Mellanox's network products. Therefore, NVIDIA's acquisition of Mellanox will help expand its business line and increase users and revenue channels.

2 With the continuous development of artificial intelligence applications, AI chips are constantly evolving and developing. A simple GPU architecture is difficult to handle the increasingly complex artificial intelligence application requirements of users, and the problem of low GPU efficiency is becoming more and more prominent. Why are FPGAs with more flexible architectures beginning to break out in recent years? NVIDIA needs to create a richer solution in the data center for users.

For NVIDIA, whether it is from the development trend of artificial intelligence applications or from the perspective of providing users with a total solution, Mellanox needs to complement products and solutions to enhance its competitiveness in the era of artificial intelligence.

3 Intel's biggest competitor forced Nvidia to choose to acquire Mellanox. From the perspective of products and solutions, Mellanox is highly compatible with NVIDIA. The two have been very tacit cooperation for many years. If Merlanox is acquired by Intel, it is equivalent to copying Nvidia's back road. NVIDIA will be extremely embarrassed, short time It is difficult to find a reasonable substitute within it, and it will definitely have an important impact on its data center business. Therefore, it is not surprising that Nvidia, with a market capitalization of US$90 billion, has joined the M&A battle with $7 billion.

Why did AI Core World Yingwei Weida acquire Mellanox?

Shopping for the data center

NVIDIA is the largest manufacturer of graphics chips for computer games. It has been working to extend this technology to other areas, claiming to provide the best solution for data processing related to artificial intelligence. At present, NVIDIA's development dilemma is that the main products are too single, and all solutions for different industries are built with GPU as the core.

NVIDIA lacks an autonomous output network and platform. Except for GPU products for the end consumer market, most cloud or network-based solutions need to be built together with industry partners. It is difficult to achieve rapid strategy based on changes in market demand. Adjustment. As a result, data centers are becoming more important than ever, and the NVIDIA Accelerated Computing Platform, combined with Mellanox's acceleration network, will create the next generation of data center computing solutions.

Mellanox is very complementary to NVIDIA's product line. The combination of long-term cooperation between the two companies will create an excellent opportunity for employees to create powerful technology. With Mellanox, NVIDIA will optimize data center-scale workloads across the entire computing, network and storage stacks to deliver higher performance, higher utilization and lower operating costs.

After the acquisition, NVIDIA has a computational stack for deep learning, HPC and big data analytics. This combination will make it a success in ultra-large-scale cloud platforms and other cloud-native companies, but if you want to make it more suitable for more traditional companies, NVIDIA will have to do more.

Why did AI Core World Yingwei Weida acquire Mellanox?

Coping with changes in the 5G era

Nvidia’s acquisition is likely to be in response to the 5G era. According to authoritative data, the rapid development of 5G technology will be ushered in the next few years, and the demand for data centers will also increase substantially. As more and more data is generated, the data processing of artificial intelligence and large databases needs to be separated between multiple computers. Just using a faster processor is not enough.

To solve this problem, the future data center will be built into a single supercomputer warehouse with "tens of thousands of compute nodes." These compute nodes need to be connected to each other and work in parallel. Chips as core technology components are critical to data center construction, and NVIDIA's move will largely enable it to gain significant positions in the upcoming 5G competition.

Limitations of Mellanox's development

With the developed military engineer system, diverse entrepreneurial groups, culture, capital, and strong sense of crisis, the country continues to create technologies and ideas that surround human needs. However, limited to the market, lack of depth, many startups are difficult to rely on Israel's own growth, shortly after the establishment, they are used to exit, through the market relay, to complete its life jump. In this process, many countries around the world regard Israel as a core research and development, design, and entrepreneurial creative center. Almost all giants have R&D centers. At the same time, many countries are constantly targeting startups and projects that were born in Israel.

Mellanox has been in business for 20 years and has a relatively mature mechanism. But from its 20-year scale curve, it should also infiltrate many Israeli company styles. In the face of a growing industry boom, when many industry applications are emerging, it has indeed come to the moment of borrowing. Mellanox's changes reflect the openness and competition of Israeli companies and the trend of global industrial division of labor. Starboard looks like a barbarian at the door, but it's just a microcosm of market power, it may just come early.

Why did AI Core World Yingwei Weida acquire Mellanox?

Intel missed the opportunity

Now that Nvidia bought Mellanox for $900 million, the situation has completely reversed. Although Intel has its own horizontal expansion interconnect technology Omni-Path. But compared to InfiniBand and Ethernet, Omni-Path has a small market share in the HPC interconnect space. For the builder of deep learning hardware systems, Omni-Path is also not attractive. At the same time, Omni-Path is far behind the first two in terms of speed.

For Intel, Mellanox is both a partner and a competitor. Intel develops its own network connectivity products based on its Intel Omni-Path architecture, including host network interfaces and switches. The company recently announced that it is developing a technology called ComputeExpressLink to improve the processing performance of compute-intensive tasks for CPUs and dedicated AI accelerators and to reduce latency bottlenecks. This new technology is supported by computer and networking giants such as Cisco, HPE and DellEMC. It's no wonder that Intel is only willing to bid $6 billion, rather than continue to entangle with Nvidia.

Why did AI Core World Yingwei Weida acquire Mellanox?


Although NVIDIA and Mellanox's $6.9 billion acquisition is insignificant in the global semiconductor market, its long-term impact is still worthy of attention, especially for the robust growth and lucrative high-end computing market. For a long time, Intel dominated the server and data center CPU market. Although AMD and NVIDIA have made progress in this market, they are far from being able to compete with Intel. Although Mellanox has a low share in this market, it has a unique position. Nvidia grabbed Mellanox as a "tip weapon" and only had the ambition to call Intel. I believe that the combined company also has the technical strength to compete with Intel.

No matter who Melanox spends, it seems that it has little to do with Chinese semiconductors and Internet companies. Although China does not currently have the ability to compete in this high-end market, as a buyer of high-performance computing and network chips, Internet companies such as Ali and Baidu can still benefit from the competition, such as high-performance computers such as Lenovo and Inspur. Server vendors have more options.

Related Products