$6,9 billion deal: why a GPU developer buys a network equipment manufacturer

Most recently, the deal between Nvidia and Mellanox took place. We talk about the prerequisites and consequences.

$6,9 billion deal: why a GPU developer buys a network equipment manufacturer
A photo - Cecetay β€” CC BY-SA 4.0

What's the deal

Mellanox has been active since 1999. Today it is represented by offices in the US and Israel, but operates according to the fabless model - it does not have its own production and places orders with third-party enterprises, for example TSMC. Mellanox manufactures adapters and switches for high-speed networks based on Ethernet protocols and high-speed InfiniBand.

One of the preconditions for the deal is the companies' shared interest in High Performance Computing (HPC). So the two most powerful supercomputers in the world - Sierra and Summit - use Mellanox and Nvidia solutions.

The companies are also collaborating on other developments, such as Mellanox adapters installed in the DGX-2 server for deep learning tasks.

$6,9 billion deal: why a GPU developer buys a network equipment manufacturer
A photo - Carlos Jones β€” CC BY 2.0

The second significant argument in favor of the deal is Nvidia's desire to get ahead of a potential competitor, Intel. The California tech giant is similarly involved in supercomputing and other HPC solutions, which in one way or another pits it against Nvidia. It turned out that it was Nvidia who decided to take the initiative in the struggle for leadership in this market segment and were the first to make a deal with Mellanox.

What will it affect

New solutions. High-performance computing in such areas as biology, physics, meteorology, etc. is becoming more demanding every year and operates with more and more significant amounts of data. It can be assumed that the cooperation between the Nvidia and Mellanox teams will first of all give the market new solutions that will be relevant not only to hardware, but also to the segment of specialized software for HPC systems.

Product Integration. Such transactions often allow companies to optimize operating costs by reducing the number of employees and improving the overall efficiency of business processes. In this case, we can only assume that this will be the case, but what is very likely is the integration of Nvidia and Mellanox solutions in "boxed" formats. On the one hand, this is an opportunity for clients to get quick results and ready-made technologies for solving problems here and now. On the other hand, there is a potential movement towards limiting the customization of a number of components, which may not be to everyone's liking.

Optimization of "east-west" traffic. Due to the general trend towards an increase in the volume of processed data, the problem of the so-called "east-westΒ» traffic. This is actually a "bottleneck" of the data center, which slows down the entire infrastructure, including the solution of deep learning problems. Together, the companies have every chance for new developments in this area. By the way, Nvidia has previously paid attention to optimizing data transfer between GPUs and at one time introduced a specialized technology NV Link.

What else is happening in the market

Some time after the announcement of the Nvidia-Mellanox deal, other data center equipment manufacturers, Xilinx and Solarflare, announced similar plans. One of the main goals of the first is to expand the range of use FPGA (FPGA) as part of solving problems in the field of HPC. The second one is engaged in optimizing the delays of server network solutions and just uses FPGA chips in its SmartNICS cards. As in the case of Nvidia and Mellanox, this deal was preceded by the collaboration of teams and work on joint products.

A photo - Raymond Spekking β€” CC BY-SA 4.0
$6,9 billion deal: why a GPU developer buys a network equipment manufacturerAnother high-profile deal is the purchase of HPE startup BlueData. The latter was founded by former VMware employees and developed a software platform for the "containerized" deployment of neural networks in the data center. HPE plans to build startup technologies into its platforms and increase the availability of solutions for working with AI and ML systems.

It is to be expected that through such transactions we will see new products for the data center, which in one way or another should affect the effectiveness of solving customer problems.

UPD: On According to a number of publications, one of Mellanox shareholders is suing for misinformation in the course of providing financial statements before the transaction.

Our other materials about IT infrastructure:

Source: habr.com

Add a comment