"Future datacenters of all kinds will be built as high-performance computers," Nvidia CEO Jensen Huang said during a telephone briefing on Monday after Nvidia revealed the acquisition of the high-performance networking company Mellanox for $ 6.9 billion.
vision – as is Nvidia – is a good reason to buy one of the leading HPC interconnect companies in the world. Of course, there are also other reasons, playing to keep Mellanox-away-from-Intel for one. Another that may not have been at the top of everyone's mind: having a support point in Israel – even if you look closely at the space, you know what the hotbed of Israel's technological innovation is.
But the main reason is Nvidia's conception of the strategic role change and modification of the architecture of the data center of the years 2020.
"The datacenters are the most important computers in the world today and in the future ̵
"We believe that in the future datacenters, the calculation does not start and will end on the server, but the calculation will extend to the network. And the network itself, the fabric, will become part of the computer fabric. In the long term, I think we have the opportunity to create data-scale computing architectures: in the short term, Mellanox's footprint in the data centers is quite wide. […] We will be able to tackle this great opportunity much better market, "he said.
Rumors are that the bid process of Mellanox started last year under pressure from the activist investor company Starboard (which bought 10.7% stake in the company in November 2017 – see our cover at that time), was highly competitive and that Intel was the best contender. Huang would not or could not confirm the reports that Intel was in the mix. But he said the offer was competitive due to the interconnection – in particular the intelligent interconnection – which becomes more important than ever, with a greater amount of IT workload carried out on the interconnection fabric. This is the so-called offloading approach that is fundamental to Mellanox's technological strategy.
Huang also praised Mellanox's networking style and his "extraordinary" software stack that was "galvanized and integrated into a large number of jobs for a long time". "
Explaining the offer process, he also emphasized the attractiveness of Israel, where Mellanox is based." It is a technological center: the culture, the spirit of the people, the wealth of the technical excellence make it a fantastic place, "he said." It is one of the largest centers of development and technological development of the artificial intelligence in the world. "
As a whole, with its high-power GPUs, the its DGX machines, its development on NVLink and NVSwitch and now an end-to-end interconnection portfolio, Nvidia is more than a systems company a point that Huang conceded substantially, although he said he prefers the term " datacenter scale computing company ".
" We were a GPU company and then we became a company of GPU systems. We have become a computing company that started from the chip up, now we are extending to a datacenter processing company, "he said.
Nvidia considers itself a different style of company from other vertically integrated suppliers. "We create the entire architecture, but we model it in such a way that we can collaborate with the ; whole IT ecosystem. We compose our technology thinking of scalability from a GPU to thousands of GPUs and future millions of GPUs. We think about it from the point of view of scalability and from the point of view of compatibility, so we can offer our components to the entire IT sector, so that everyone can build computers and configurations that satisfy them and solve their problems ", said Huang.
"Our business model will not change even if we continue to expand in a system sensitivity; we are not a systems company, if you want it. We are really a system architecture company, "added Huang, underlining the distinction." We are open to cooperate with our customers and partners as they wish. If you wish to group our DGX servers like NetApp and Pure and DDN, we are pleased. If they would like to buy our HGX motherboard, in practice the system board inside our DGX, to be included in their cloud, like what Google did and Microsoft did and others did, we are pleased. If you want to buy our chips and create your systems, we are pleased or if you want to buy them as an additional card. The thing we work hard on is to make sure they are all compatible, so the CUDA acceleration libraries work perfectly on all of them. "
C & # 39; is a reason for the careful and prudent language, the coverage towards Nvidia that is committed to business of systems – and is a point that Addison Snell, CEO of Intersect360 Research shared with us." Nvidia must be careful that its aspirations do not threaten their important server-side partners. While Nvidia is able to sell reference components and architectures directly to hyperscale companies and ODMs, it should not take the traditional OEM channel for granted. "
Actually the only thing missing from a complete system stack on Nvidia is the generic CPU. Nvidia was going to get into the CPU game that dates back to 2011 with Project Denver, but discarded those plans. Arm momentum high and Arm looking for partners to make Neovers chips, many think it's a natural next step for Nvidia.
Ma Huang, when asked, seemed less enthusiastic about the CPU – as an opportunity He reiterated his position that the company believes in serial processing (recognizing "you can't parallel everything"), but stated that while Nvidia is happy to work closely with all the major CPU companies – IBM, AMD and yes Intel ("people think we're antagonistic, but that's not true") as well as Cavium / Marvell, Ampere and Broadcom on the Arm side – see more benefits in focusing their research and development on engineering in gr I give "x improvement factors".
"I think if we were pouring a lot of R & D into CPU construction today, the x factor you get, after five years, is about 15 percent – and [the CPU makers] fights more than 15 percent!", said Huang. "And since they are doing a good job fighting over 15%, we could invest our R & D in areas that have big factors x – places where people don't invest. And I think the acceleration of the ; computer science has been seriously underestimated, so our return was fantastic and will continue to be fantastic and improve ".
Returning to the Mellanox agreement, it is largely considered a strategic purchase for market-watchers.
"Nvidia at $ 10 billion capex against Mellanox at $ 1 billion suggests that Nvidia sells many more customers, has a more extensive sales / distribution channel and a global service infrastructure," a long time observer said date of the sector with which we spoke. "Through economies of scale, by simply dropping the Mellanox product line into the Nvidia portfolio, the Mellanox business line should be driven significantly. This could be the main reason they paid almost $ 7 billion. They looked at Mellanox's leading technology , taking into account their customer base and channels, and evaluating sales opportunities over a period of time, typically from three to five years. Nvidia must feel able to lead the Mellanox business unit to $ 7 billion in three -five years, in addition to improving its core products. "
Returning to the technological side, Snell noted that" the combination of Nvidia and Mellanox is intriguing for HPC and AI. Companies are the respective leaders in their technological areas to high performance and each has driven important growth trends. The purchase by Nvidia of Mellanox not only blocks potential synergies in the advancement of broadband connections for accelerated systems, but also protects technologies from Intel, which is a natural competitor for both companies. "