"Future data centers of all kinds will be built as high-performance computers," said Nvidia's CEO Jensen Huang during a phone call on Monday after Nvidia revealed the high-speed network operations Mellanox for $ 6.9 billion.
If it's your vision – as it is Nvidia – there's a good reason to buy one of the world's leading HPC interconnectors. Of course, there are other reasons, Keep-Mellanox-away-from-Intel plays for one. Another one that may not have been the top of everyone's brain: to gain a foothold in Israel – even if you look closely, you know what a hotbed of technological innovation Israel is.
But the primary reason is that Nvidia's perception of strategic changes role and change of architecture, of the data center to the 2020s.
"Datacentre is the most important computer in the world today and in the future ̵
"We believe that the data in future data centers will not start and end on the server, but the calculation will extend to the network. And the network itself, the substance, becomes part of the data processing. Long-term, I think we have the ability to create data center-scale computing architectures; In the short term, Mellanox footprints in data centers are quite large. […] We will be able to handle this great market opportunity much better, he says.
Rumors are that the Mellanox bidding process, which started last year under pressure from the activist company Starboard (which acquired a 10.7 percent stake in the company in November 2017 – see our coverage at the time), was highly competitive and that Intel was a top competitor. Huang would not or could not confirm reports that Intel was in the mix. But he said the bid was competitive because of the interconnection – especially the intelligent interconnection – became more important than ever, with more of the computing load being done on the interconnect fabric. This is the so-called offloading approach that is fundamental to Mellanox's technology strategy.
Huang also praised Mellanox's networking style and their "extraordinary" software chip that has been "galvanized and integrated into a large workgroup over a long period of time."
By explaining the hot bidding process, he also pointed to Israel's favor, where Mellanox has its head office. "It's a technology center – the culture, people's spirit, wealth of technical excellence makes it a great place," he said. "It is one of the world's major AI development centers and technical development centers."
Along with its high-powered GPUs, its DGX machines, the development of NVLink and NVSwitch, and now an end-to-end interconnect portfolio, Nvidia is more than a de facto system company, a point that Huang mainly admitted, although he said he prefers the word "data center scale computing company."
"We were a GPU company and then we became a GPU system company. We became a computing company that started from the chip, now we reach out to a data center database," he says.
Nvidia sees itself as another company style from other vertically integrated suppliers. "We make the whole architecture, but we compose it in such a way that we can cooperate with the entire IT -økosystemet. We compose our technology by thinking of scalability from a GPU to thousands of GPUs and in the future millions of GPUs. We think of it from a scalability perspective and from a compatibility perspective, so that we can offer our components to the entire IT industry so that everyone can build computers and configurations that satisfy them and solve their problems, Huang says. .
"Our business model will not change even if we continue to expand into a system sense; we are not a system firm, if you want. We are truly a system architecture company," Huang said, emphasizing the distinction. "We open ourselves to work with our customers and partners, but they like it. If they want to pack our DGX servers like NetApp and Pure and DDN, we are happy about it. If they want to buy our HGX motherboard, is the system board in our DGX, basically, in Google Cloud, which Google has done, and Microsoft has done, and others have done it, we are happy about it. If you want to buy our chips and build your own systems, we are happy about it, or if you want to buy it in the option card form. What we work hard is to make sure they are all compatible, so that CUDA acceleration libraries work perfectly on top of them. "
There is a reason the cautious and cautious language, the security with regard to Nvidia commits to system activity – and it is a point that Addison Snell, CEO of Intersect360 Research shared with us. "Nvidia does not have to be careful that its desires do not threaten their important server side partners. While Nvidia can sell components and reference architectures directly to hyperscale companies and ODM, they should not take the traditional OEM channel for granted."
Really the only missing item from a complete Nvidia system stack is the general CPU. Nvidia intended to enter the CPU game that came back in 2011 with Project Denver, but threw those plans. With the arm torque high and Arm looking for partners to make their Neoverse chips, many think it's a natural next step for Nvidia.
But Huang, when asked about this, seemed less than excited by CPUs – as a market opportunity. He reiterated his belief that the company believes in serial processing (confirming "you can't parallel everything") but confirmed that while Nvidia is happy to work closely with all major CPUs – 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's side – it is more beneficial to focus engineering R&D that can deliver" x factors for improvement. "
" think if we were to pour a lot of R&D to build processors today, the x factor you get, after five years, is about 15 percent – and [the CPU makers] struggle over 15 percent! "Huang said. "And since they do such a good job fighting over 15 percent, we can invest our R&D in the areas that get big x-factors – places where people don't invest. And I believe that accelerated computing was heavily under-invested and so has our return It has been fantastic and it will continue to be fantastic and get better. "
Returning to the Mellanox agreement, it is largely held as a strategic purchase after market -watchers. "$ 10 billion capex versus $ 1 billion Nvidia suggests that Nvidia sells many more customers, has a broader sales / distribution channel and a global service infrastructure," a long-standing industrial observer we spoke to said. "By using economies of scale, simply releasing the Intermediate product line into the Nvidia portfolio, we should run the Interox business significantly. This can be the principle that they paid nearly $ 7 billion. They looked at Mellanox's leading technology, based on customer service and channels. , and estimated the opportunity over a period of time, typically three to five years. Nvidia must feel that they can operate the Mellanox business area at $ 7 billion for three to five years, as well as improve their core products. "
On the Technology site, Snell said" the combination of Nvidia and Mellanox are excited about HPC and AI, the respective leaders in their high-tech technology areas, each with important growth trends, while Nvidia's acquisition of Mellanox not only unlocks potential synergies by promoting high-bandwidth connections for accelerated systems, but also protects Intel technologies, which is a natural competitor to both companies. "