Arm and NVIDIA: Simon Segars and Jensen Huang

What used to be face-to-face Arm TechCon has turned into a virtual conference under the name Arm DevSummit.

In the unlikely event that you missed the news, NVIDIA  announced on September 13 that it intends to acquire (most of) Arm for $32B from Softbank, the Japanese company that currently owns it. It is likely that regulatory approval will take some time (and that means that the companies will have to operate independently for some time).

This post covers the two opening events of the conference, Simon Segar's keynote (Simon is the CEO of Arm) and a "fireside chat" with Jensen Huang (Jensen is the CEO of NVIDIA) hosted by Rene Haas, who is president of Arm's Intellectual Property Group.

Simon's Keynote

The conference opened with Simon Segars, Arm's CEO. And he opened by briefly discussing that NVIDIA is in the process of acquiring Arm.

Let me try and clear some of that up. I’m a huge fan of the two companies becoming one. Imagine Arm’s global deployment of CPU technology combined with NVIDIA's technology.

Simon continued with what some partners around the world are doing with Arm processors. I'm going to skip that and just talk about a few special projects that Simon covered in his keynote.

Simon moved to power in general, always one of Arm's strengths, of course, but now stretching all the way from the world's fastest supercomputer (see my post Japanese Arm-Powered Supercomputer Takes the TOP500 Crown) to tiny IoT motes.

Next, Project Triffid. This is something straight out of Arm's research labs and taped out at 28nm. It is a chip with extremely low power. In fact, so efficient that it can be powered by a single flash from an RFID reader. It is intended to track things as small as food packages.

For the server end of things, there is Arm SystemReady, where Arm is working with RedHat, Microsoft, and VMware, along with dozens of Arm partners. It is a common software stack with OS, hypervisor, and middleware. It is part of Project Cassini (cloud readiness). The idea is that stuff "just runs". It is also tied in with a secure boot option PSA, Platform Security Architecture.

Simon talked briefly about a program working with teenagers known as Gen Arm2Z, and a specific project PlantPal for smartGarden.

Every Arm keynote has the total number of Arm-based chips that have shipped. I remember when it eventually got all the way to a billion after about 20 years. This year the number is 180 billion.

Simon and Jensen's Fireside Chat

As I said at the start of this post, the fireside chat was hosted by Rene Haas, who runs Arm's IP business (on the left). He introduced Simon (center) and Jensen (right) as "my current boss and presumably my future boss's boss".

He started by asking Jensen "Why did you make this decision?"

Jensen explained:

We would like to create a computer company for the age of AI, and unite the world leader in AI computing and the most popular CPU company in the history of computing. We will bring our technology to the Arm ecosystem. Together we will create solutions that are great for customers whether in the cloud, edge, autonomous machines, and more.

The next question was a softball for Simon to get things going. "How do you see the two companies working together?"

Simon replied:

Computing is becoming so much more pervasive than today, moving from CPUs running regular code to AI where software is writing software, and it is a completely different programming paradigm. We will be able to put tools into the hands of developers so that they can create great products. After all, hardware without software is useless.

Rene said to Jensen that there have been a lot of questions about Arm's business model and whether it will change going forward. He replied:

I love Arm’s business model. The genius of Arm was inventing this super low-power CPU architecture combined with this business model. That vision took three decades to realize. I love the ecosystem. We have every intention of protecting it, nurturing it, and growing it. I want to add to the business model. Arm has the ability to take very difficult designs and turn into soft IP products anyone can use. I wish we could do that. I’d love to take NVIDIA's IP and expand our vision of accelerated computing give more access to others.

What about what Arm is currently doing in the GPU space?

Obviously, NVIDIA has world-class GPUS, but we don't see any change with Arm. Different people need different products.

Much of the audience for DevSummit are software developers, many from the open-source community. How will this affect them?

Software developers want several things. First, a computing platform with amazing capability. They need a richness of tools and libraries so they don't have to reinvent everything. But mostly they want a platform with large reach that is vibrantly growing. We are going to combine two of the most vibrant ecosystems in the world. The largest is Arm. One that is developing very fast with the growth of AI is the NVIDIA ecosystem. Developers will love it. A great platform with vibrant growth.

Simon, what about Moore's Law ending? What are your thoughts?

This issue about Moore’s Law slowing has been with us for a long time. There are new transistor structures being worked on now. As we go forward, one part of the futures is putting different die together in different processes in sophisticated packaging. These are the innovations that will keep semiconductor performance improving generation after generation. We have programs so these advanced technologies can become mainstream.

Both companies have talked about AI. How does that change with a combined Arm/NVIDIA?

Jensen: This is exciting. We know there is a whole new world of computing about to happen. Some call it edge. Some call it IoT. We can create autonomous intelligent computers, driving a car, moving crates in a warehouse, fast checkout, keeping traffic flowing. Almost as if the city or the store or the warehouse is smart. The software is called AI and no human knew how to write that software. Now that the software is possible, we're reminded that software creates new markets. For the first time, that growth is going to come from software written by machines. For the first time, it is possible to imagine a world where we are creating autonomous systems around the world with software written by other computers. That's why the combination of Arm and NVIDIA makes so much sense. No other computing platform has the reach of Arm.

Simon: Yeah, I think it’s delivering on that vision. AI grew up with huge datasets living in the cloud. If you try and take all the data at the edge and put it in the cloud, it is massively inefficient. We need to worry about where the data lives and how to handle it there.

Jensen, you had an announcement planned.

Yes, I was going to announce it at "my" show, but Simon said why don't you come to "my" show. Last year we put the first stick in the ground since it took us several years to realize that Arm would be everywhere and we need to make the commitment to bring our architecture to the platform with CUDA. There are four pillars to this. First, NVIDIA will bring GPU as well as DPU and all system software to the Arm platform. CPU/GPU/DPU to create a general computing platform. Second, we'll bring domain-specific acceleration libraries to Arm: HPC (quantum chemistry, fluid dynamics, 1000 solvers accelerated with CUDA), AI and all the deep learning algorithms from training to inference, and the data analytics platform. The third pillar is formed by the four major compute platforms: HPC, cloud, edge, and personal computing (all our graphics and imaging). Finally, the fourth pillar is to create an ecosystem. We selected three partners as a starting point: Fujitsu, who have built the worlds fastest supercomputer based on Arm, MPure, and Marvell with their Thunder architecture.

Jensen, you announced the Center for AI Excellence in Cambridge a few weeks back. What is it?

As you know, Cambridge is the home for the beginning of computing [EDSAC] and the beginning of DNA [Crick and Watson]. Today it is a hotbed of genomic AI research. What the ecosystem needs is a supercomputer. So we are going to build a brand-new supercomputer, the fastest supercomputer in the whole of the UK devoted to genomic research. Five companies are already signed up to join us. We have created a brand new computational drug discovery stack, along with all kinds of tools for drug discovery and other things.

Last question. Talk about some of the questions about the regulatory process, what it will be like, and the confidence level that it will be approved.

Jensen: I'm sure it will go through. Once we explain our plans to the regulators around the world, they know that these are two complementary companies. Combined, we will create new innovations. We love the business model and we will continue to nurture this trusted platform so that the ecosystem can grow.

Simon: This is a deal about expanding and putting this technology into the hands of people who will build really cool stuff with it. But it will take a while since it is the right thing for regulators to look at this thoroughly. Everyone will realize that this is great for expanding the market.

Thank you both.

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.