Nvidia executives have taken a significant step forward by strengthening the GPU manufacturer's presence in the rapidly changing data center space and making it an even more formidable competitor with Intel with the planned acquisition of 6.9 billions of dollars from network provider Mellanox Technologies.
The company announced the deal on March 11, after several days of speculation that Nvidia was interested in buying Mellanox, which had also attracted Intel's attention to the acquisitions, with information circulating that Intel had proposed $ 6 billion. Another chip maker, Xilinx, was also among the bidders. The agreement is the most important in Nvidia's history.
Nvidia founder and CEO, Jensen Huang, said during a conference call announcing that having Mellanox would increase his company's capabilities and enhance his status in a wide range of data center scenarios , high performance computing and enterprise environments in the cloud. and hyperscale infrastructure.
"This acquisition … would transform our society and transform us from a chip and systems company into a data center scale company," Huang said. "In the future, we want to optimize workloads at the data center scale, also across the entire stack, from the compute node to networking, through storage. That's why Mellanox's data center interconnect technology is important to us. "
The widespread interest in Mellanox, which offers a broad portfolio of networks, ranging from InfiniBand and Ethernet switches to adapters, silicon and software, is accompanied by the ever-changing workload of data centers with intelligence artificial intelligence (AI), machine learning and data. analytic. Data centers are becoming more data centers and computing needs, putting greater pressure on architectures to get more speed and bandwidth, as well as greater energy efficiency and cost savings.
Moore's law slows down
At the same time, said Huang, Moore's law, which has been driving innovations around processors for more than five decades, is beginning to slow down. Nvidia started ten years ago to offer its GPUs to accelerate server performance, first in HPC, then in cloud computing and data centers, and heterogeneous and accelerated environments will be essential to handle these loads. more modern work, he said.
With the growth of AI, analytics and other data-intensive workloads and computations, companies will need tens of thousands of server nodes functioning as one and the same unit. The network will be the connecting fabric that unites the environment, said Huang, who placed artificial intelligence and machine learning at the center of Nvidia's growth plans several years ago.
"The progress of the processor slows down, while the computer demand is skyrocketing," he said. "As a result, accelerated computing and accelerated networking … are a fundamental way forward. … The dynamic that is happening here is that in the future, it will not be just computer-scale server, but computer-scale computing. data center, where the network becomes an extension of the IT structure. "
The combination of Nvidia and Mellanox will be a great data center company, according to Patrick Moorhead, Senior Analyst at Moor Insights and Strategy.
"This agreement allows Nvidia to become a bigger player in data centers, especially where the market is heading, and it's a more heterogeneous IT system," said Moorhead. eWEEK. "The more heterogeneous computer science is, the more [more] more efficient networking is needed. "
Nvidia responds to performance issues
Rob Enderle, Senior Analyst of the Enderle Group, said eWEEK that the agreement "allows [Nvidia] directly solve some of the performance issues of their HPC solutions that, until now, were not part of the Nvidia product line. Nvidia solutions tend to generate massive data streams due to their visual nature and they often interfere at the interconnect layer. Mellanox's technology already addresses this problem, but closer collaboration between companies should have a significant advantage in terms of targeted tuning, which should benefit Nvidia.
"At present Nvidia is more of a parts supplier than a solution provider, but as the need for technology evolves in the data center, it has been forced to take care of other parts of the business. the ecosystem, especially the interconnection technology, Dit Enderle. "This will strengthen their portfolio and you should expect them to make new acquisitions as they discover new system bottlenecks, which have a negative impact on the performance of their offerings." . "
Nvidia and Mellanox already have a long history of collaboration. Huang said its technologies occupy more than half of the systems listed in the world's 500 fastest supercomputers, including Summit and Sierra, ranked No. 1 and 2 on the latest list published in November. Products from both systems have also been adopted by leading OEMs, including Dell EMC, Hewlett Packard Enterprise, IBM, Lenovo and Cisco Systems, as well as cloud service providers such as Amazon Web Services, Microsoft Azure, Google Cloud Platform. , Oracle Cloud and Alibaba Cloud.
At the same time, Nvidia has leveraged Mellanox technologies into its own systems, including the DGX-2 server (shown) for IA workloads, which uses Mellanox InfiniBand and Ethernet technologies.
Mass Addressable Market
Nvidia officials said a merged company would cover a market comprising more than $ 60 billion in high-speed computing and networking by 2023.
The agreement has been approved by the boards of directors of both companies and will be subject to the approval of Mellanox shareholders and regulators. Nvidia officials said they did not see much overlap in Nvidia's and Mellanox's offerings and that the combined company would continue to invest in employees and operations in Israel, where Mellanox is based. The agreement should be finalized later this year.
The changing dynamics and changing workload in modern data centers are also driving other chip makers to adapt. Intel, which still holds a leading share of the data center chip market (estimated at 95% north), continues to innovate not only on its Xeon processors, but also offers accelerators such as Field Programmable (FPGA) Gate Arrays), thanks to its $ 16.7 billion acquisition of Altera in 2015 – and its ASICs through the purchase of eASIC last year, and plans to introduce its own GPUs for the future. by 2020.
Advanced Micro Devices is preparing the next generation of Epyc server chips based on its Zen microarchitecture, expected to come out this year, and unveiled in 2018 its latest Radeon Instinct GPU accelerators. Xilinx, which has made a name for itself as an FPGA manufacturer, now offers computing acceleration platforms such as its Versal offer for artificial intelligence workloads, introduced in October 2018. She also integrated her FPGAs into accelerator cards for the data center.