"Future data centers of all types will be built like high-performance computers," said Jensen Huang, CEO of Nvidia, during a telephone meeting on Monday after Nvidia announced it would acquire Mellanox, a high-performance networking company, for $ 6.9 billion.
If that's your vision – as Nvidia does – it's a good reason to buy one of the world's leading HPC interconnect companies. Of course, there are other reasons to play Keep-Mellanox-away-from-Intel for one. Another, which may not have been on everyone's lips: to gain a foothold in Israel – if you closely observe the space, you know what a haven of technological innovation Israel is.
But the main reason is Nvidia's idea of the changing strategy role and changed architecture of the data center of the 2020s.
"Data centers are now and in the future the most important computers in the world ̵
"We believe that in future data centers, the computation will not start and end at the server, but the computation will extend into the network, and the network itself will become part of the network In the long run, I believe we have the ability to create computing architectures for data centers, and in the short term, Mellanox's presence in data centers is quite large. […] We will be able to better address this great market opportunity "He said.
There are rumors that the Mellanox bidding process in the Last year, under pressure from the activist investment firm Starboard (which had bought 10.7 percent) started to share in the company in November 2017 (see our then-coverage), was extremely competitive and Intel was a top contender. Huang wanted or could not confirm any reports that Intel was in the mix. However, he said the offer was competitive as the interconnect – especially the intelligent interconnect – became more important than ever as it required more computational effort on the interconnect fabric. This is the so-called offloading approach, which is fundamental to Mellanox's technology strategy.
Huang also praised Mellanox's networking style and its "extraordinary" software stack, which has been "integrated into a variety of applications over a long period of time".
In explaining the hot bidding process, he also referred to the attractiveness of Israel, where Mellanox is headquartered. "It's a technology center – culture, the spirit of the people, the wealth of technical excellence make it a great place," he said. "It's one of the world's largest AI development centers and technology development centers."
Along with its powerful GPUs, DGX machines, NVLink and NVSwitch development, and now an end-to-end connectivity portfolio. Nvidia is more than a de facto system company a point that Huang essentially admitted, although he said that he prefers the term "Datacenter Scale Computing Company". "
" We were a GPU company and then we became a GPU system company. We have become a computer company that has risen from the chip. Now we are expanding into a data center company.
Nvidia sees itself as a different corporate style from other vertically integrated vendors. "We build the entire architecture, but compose it to work with the entire IT ecosystem. We're composing our technology by thinking about scalability from one GPU to thousands of GPUs and the next million GPUs. We think about it from a scalability and compatibility point of view, so we can offer our components to the entire IT industry so that anyone can build computers and configurations that appeal to them and solve their problems, "said Huang
" Our business model is going to grow not change, even if we are increasingly becoming systemic; We are not a system house, if you like. We're really a system architecture company, "Huang added. "We are open to partnering with our customers and partners as they wish. If you want to bundle our DGX servers like NetApp and Pure and DDN, we'll be happy to hear from you. If you want to buy our HGX motherboard, basically the system board in our DGX, to get it into the cloud, as Google did and Microsoft and others did, we're excited about it. If you want to buy our chips and build your own systems, we look forward to it or if you would like to buy them in the form of an additional card. We work hard to make sure they're all compatible, so the CUDA acceleration libraries work perfectly on all of them. "
There's a reason for the derogatory and cautious language, backing for Nvidia's commitment to systems business – and Addison Snell, CEO of Intersect360 Research, shared with us. "Nvidia has to be careful that his efforts do not threaten their important partners on the server side. While Nvidia can sell components and reference architectures directly to hyperscale companies and ODMs, they should not take the traditional OEM channel for granted.
The only element missing from Nvidia in a complete system stack is the general-purpose CPU. Nvidia had the intention to enter the CPU game with Project Denver, but rejected them. With Arm Momentum High and Arm looking for partners to build his Neoverse chips, many think that this is a natural next step for Nvidia.
But when Huang was asked about it, he did not seem to be so enthusiastic about the CPUs – as a market opportunity. He reiterated his stance that the company believes in serial processing (recognizing "you can not parallel everything"), but reiterated that Nvidia is happily working with all the major CPU companies – IBM, AMD and yes Intel ("people think we are antagonistic, but it's just not true.)) as well as Cavium / Marvell, Ampere and Broadcom on the arm side – it sees more benefit in that research and development focuses on engineering that can deliver "x improvement factors"
"I think if we were to invest a whole range of research and development activities in the construction of CPUs today, the x-factor that you receive after five years is about 15 percent – and [the CPU makers] fighting over 15 percent! "said Huang," and since they are doing so well and fighting over 15 percent, we could invest our R & D into those areas where there are big ones There are factors in which people do not invest. I think that accelerated computing has been heavily underinvested and our return there has been fantastic and will continue to be fantastic and better.
When we return to the Mellanox deal, it is largely held by the market as a strategic buy-watcher.
"Nvidia's $ 10 billion from Mellanox at $ 1 billion indicates that Nvidia has sold many more customers, a more complete distribution and distribution channel, and a global service infrastructure," said a longtime industry observer with whom we spoke to have . "The economies of scale of simply adding the Mellanox product line to the Nvidia portfolio should significantly drive the Mellanox business. This could be the main reason why they paid nearly $ 7 billion. They looked at Mellanox's leading technology, considering its customer base and channels, and estimated the opportunities over a period of typically three to five years. Nvidia needs to feel it can increase the Mellanox business to $ 7 billion in three to five years and improve its core products. "
On the technology side, Snell said," The combination of Nvidia and Mellanox is intriguing to HPC and AI. The companies are leaders in their high-performance technology areas and have left important growth trends behind. The acquisition of Mellanox by Nvidia not only creates potential synergies for the development of high-bandwidth links for accelerated systems, but also protects Intel's technologies, which are a natural competitor to both companies. Share this: