AMD CTO: We Now Have A 'Razor's Edge' To Fight Against Intel

Papermaster Outlines AMD's Plan

While Intel has been outlining the company's multi-product vision for its data-centric future, archrival AMD has been riding high on the promising growth trends of its Ryzen desktop and EPYC server CPUs, as well as plans for its next-generation, 7-nanometer line of processors.

The Santa Clara, Calif.-based company has not been shy about its perceived chip manufacturing position over Intel, which has delayed mass production for its next-generation 10nm processors for multiple years. Intel's 10nm desktop CPU will now launch for holiday 2019 while its 10nm server processor won't come out until 2020. AMD, on the other hand, is expected to launch its 7nm server chip next year with the desktop version to follow sometime after that.

In an exclusive interview with CRN, AMD CTO Mark Papermaster reiterated the company's confidence in its 7nm plans, how it was able to leapfrog over the "incremental step" of 10nm and laid out the company's plans to take more market share in the server and desktop markets.

"That's what AMD has always been about — innovation — but now it's got a razor's edge to fight against larger competitors that have been dominating the industry," he said.

What follows is an interview transcript that has been edited for length and clarity.

What would you say are the biggest bets AMD is making right now?

The big bet that we're making is that high performance matters, and we made this bet really over five years ago because we had to revamp the roadmap when myself and Lisa Su, who is now CEO, had to revamp the roadmap and to get it focused on restoring AMD's place in the industry … We're one of the very few providers of high-performance, both CPU and GPU. So that's the foundation of everything that we do, to be that bankable supplier of high-performance technology, now and for the future. Computing is changing. It's becoming immersive, so how you interface with the computers, you expect high-resolution imagery, not only in flat screens but [also with virtual and augmented reality]. That drives tremendous computing, so how we interface is changing. And then the underlying algorithms that can handle this massive data that is out there being created around us, with all of these embedded [Internet of Things] devices. So whether it's your home life, your business life, you've got a massive amount of data and we provide the engines, both CPU and GPU, to provide the compute capability, both with traditional workloads and machine learning.

How did you feel hearing the news that Intel's 10nm CPU is being delayed to holiday 2019?

My view is very simple. We assumed that the competition would be in 10nm this year, and we laid all of our plans out. The take that we have at AMD is focus and execution. We are not going to change our plans. We assumed that 10nm would proceed on its original schedule from our competitor, so we're executing to all the plans that we originally put in place. I think we do have the opportunity to be positioned much stronger than we originally anticipated, but I have to say, our original plan, was to be positioned very strongly, so any delay from our competitor could simply strengthen the value that AMD brings to the market.

Why are 7-nanometer processors important to AMD?

You first have to go back and look at the FinFET [fin field-effect transistor] device, the vertical transistor. We launched that starting in 2015 and across the board in 2016. It's an excellent device, because it scales very well, so you can run it at high-performance very efficiently, and it has very good characteristics if you might be in a battery application where you need very low leakage. It's proven to be incredibly robust. So we looked at 10nm, and it was an incremental step. It didn't provide for us the kind of performance per watt of energy, it didn't provide us the amount of gain we were looking for, so we made a bet to skip that and to go to 7nm. 7nm is a full-step new node, and it allows the device to shrink, as well as [cut] the power dissipation down significantly.

Can you quantify that at all?

If you want to run at the same speed you were in the previous node, you can save half the energy. So you're really getting double the performance per watt of energy expended in a previous node. That's significant.

AMD CEO Lisa Su was talking on the last earnings call about the timeline for when the products are coming out. You have 7nm Vega later this year in the fall?

We do. We originally planned it for 2019, but Lisa Su, our CEO, announced at Computex, they're actually pulling it in to the second half of 2018.

How did AMD do that?

The manufacturing for 7nm is very solid. The [foundries are] incredibly focused on 7nm, and what we found — it was indeed a huge challenge for 7nm. Because the dimensions are shrinking, very, very small, it takes additional manufacturing steps, meaning new masking levels. So it's a lot of complexity when you go to 7nm, but what we did at AMD is the teams did a great job of creating a partnership of our design community with the foundry who owns design of that transistor with the electronic design automation industry, the EDA. They manufacture all of our computer-aided design programs, and turns out, they had to be the third leg of the stool because it was such a difficult lift. It changed even the automation tools that are required to put together these types of complex leading-edge designs.

Was there anything specific that allowed you to push up Vega's timeline?

What you saw from AMD was immense focus. We knew 7nm would be a big challenge, so we made the bet, we shifted our resources onto the new node. We didn't just dip our toe in the water. We went all in. So we shifted our design community, as well as our design enablement. We made it clear to our partners that we needed their full focus on this new technology node, and it's a good thing, because it was really tough. We ran into our share of issues, but we started early, and we did test designs. We did pre-production releases of the chips to test out that it would all work out as expected. And that our designs, the process and our CAD tools could work together. So by the time we were ready to release our products to manufacturing, there were no surprises.

What is the advantage of working with GlobalFoundries and TSMC (Taiwan Semiconductor Manufacturing Co.) as opposed to doing it on your own?

Well, before I got here, AMD had made a decision to go fabless, because the value we bring, of getting great, high-performance products to market, we need to be able to move very nimbly from a design standpoint. And to really leverage what it takes to be successful in a semiconductor foundry industry, it needs 100 percent focus. That's what the fabs do. So we're doing what we do best, creating outstanding products, and then we can partner with GlobalFoundries and TSMC, our key fab partners.

In general, what do you think people should be most excited in terms of what AMD is working on?

I think, when people think about AMD, they have to think about what AMD has done to transform itself over the last five years under Lisa Su's leadership and her management team. And that is that AMD is a company of maniacal focus on delivering customers a new and better experience, providing competition, and competition drives innovation. That's what AMD has always been about — innovation — but now it's got a razor's edge to fight against larger competitors that have been dominating the industry and provide this choice and innovation, so it's a very exciting time, and it's a very different AMD.

What gains does a customer get from having an AMD CPU and GPU?

As the only player in the industry that has the key IP assets of high-performance CPU and GPU, we're very focused on leveraging that. What we've done in PCs is we launched a new product line called Ryzen, and Ryzen in notebooks has an integrated CPU and GPU on a single die, so it's providing great energy efficiency. We're providing in the leading tier of battery life performance while being able to run triple-A games and have a leadership tier of CPU performance, so we're playing in segments that we didn't even compete in in the PC space, and we're providing a best-in-class experience … And then with the Ryzen desktop products, we've got a beast out there, so when you look at Ryzen, it goes all the way up to a Threadripper — it has a tremendous advantage of cores, up to 32 cores for Threadripper 2.

Is AMD's Ryzen Threadripper processor mostly just for hardcore consumers, or is it for the professional market also? Because I know you have Ryzen Pro as well, which is more focused on workstations.

So we started with the consumer market [with Ryzen] because there was pent-up demand for that kind of high-performance in gaming and content creation. But Ryzen Pro quickly followed and what it does is it takes that same horsepower, but it hardens it for the enterprise market. What does that mean? It adds reliability features, it adds additional testing features to be able to meet those rigorous requirements and manageability requirements in the professional market. It also brought a certification of many content creation applications. So it's taking that horsepower that we first launched in the consumer market right into that content creation market [with Ryzen Pro].

It seems like AMD's EPYC server line had a good first year.

It did. Our goal for EPYC the first year was really re-establishing that AMD was back. We know that in server, it's a long test and certification cycle, so we launched that last June, and it's been going very well. Customers are kicking the tires, and they've gone beyond that phase now. And so it's in production. We're out there with the key OEMs, we already have several data centers — hyperscale data centers — that have announced and are standing up instances with EPYC. And by all accounts, we're providing that efficiency of computing by bringing more cores per socket, more memory, more I/O, things that the competition has bounded and constrained. We're opening that up and, again, it's getting a great response from our customers. It's got us on a growth vector in servers, where we expect to be able to end the year at about 5 percent of market share and grow to double-digit next year.

With EPYC, AMD has been boasting about the ability to offer two-socket performance in a one-socket configuration. Can you tell me what that's about?

What's great about EPYC is the predominant form factor in server today is a two-socket configuration, and we have a great two-socket configuration. We give customers more CPU cores at a cost efficiency and a performance-per-dollar, what we call total-cost-of-ownership, advantage in those traditional two-socket systems. And we compete against our competitor at almost the entire full range of product performance. So we brought competition back to that predominant form factor.

We did something else that's pretty interesting, and that is, we got a disruptive offering in a single socket. So what we did in a single socket is, we allow you to run a full complement of memory and I/O, so it's one-third more the amount of memory lanes available versus competitive platforms and a really massive interconnect capability. You can run in a single-socket solution many of the applications in probably half the applications that are out there today which are running in two-socket. So it's absolutely disruptive.

What is the opportunity for GPUs in the data center?

We see three key workloads driving a spike in demand for GPUs in the data center. One is virtual desktop: to be able to provide a high-performance CPU and then the visualization with the GPU in the data center, rendering your images in the data center — all you need is a solid internet connection to your workstation and you're going to have a great experience. The second is gaming, [where] cloud is starting to become real. You've seen announcements of gaming services that have already been out there for several years are starting to get traction and new players, new entrants are coming in. The third that is really going to drive GPU is machine learning, because machine learning applies to most applications you can make. If you have data that can be trained and put into use, you're going to want a GPU. It's the most flexible device. There are specialized chips you can go do for unique applications and several companies have done that, but the algorithms are constantly changing and so the most general-purpose way to protect as algorithms change over time is a GPU.

What's the pitch to partners for adopting AMD's ecosystem of products?

The reason AMD is really seeing such a high degree of attention right now is not just that we returned our product lineup to competitiveness across the entire range of performance, but that we did it in a way that's very open. It's very easy to partner with AMD. Our platforms have stability, our platforms are designed to be able to have add-ins, so we're very partner-friendly, both for IHVs with slots, where they can add in their capabilities and drop right into an x86 ecosystem, and for gaming, we've opened up our driver stack. So if you've looked at what we've done, whether it be the [AMD Vulkan Graphics API], DirectX, you can look at our GPU usage in the Apple platform — all of those are examples that we are very motivated to partner and build a broader ecosystem out in the industry.

You previously held engineering positions at Apple and then Cisco. What attracted you to AMD?

You have to go back in time when you look at how I ended up at AMD, because it was late 2011, and AMD was in a very serious situation. It's known historically for its innovation and it's sitting on world-class IP (intellectual property) around CPU and GPU [graphics processing unit]. And my experience is on getting teams to gel, focus on products, really getting great products out from an engineering standpoint. So when one of the board members I had known most of my career called me and said, "you really have to consider AMD," at first I said, "no, it seems like a pretty big lift there, and I'm busy." And they said, "you know, you may look back on this if you didn't take it, it may be actually a chance of a lifetime to really help turn around a company." It was exactly right, because the talent at AMD is outstanding. It really understands technology, and its application in the industry. We know our markets well, so it really was a phenomenal opportunity to come here and help make an impact and marshal those resources to get out leadership technology products.

What's your main focus as far as responsibilities?

I have two jobs: I'm the CTO and I also run what's called technology engineering. From the technology engineering role, I drive product development, so it's creation of those base IPs that we use around the CPU and GPU technology, how they come together in the system on a chip, the product roadmaps, how we attack our competition. And as a CTO, I do a lot of external facing [work] as well, so it's translating that value proposition of our technology to our customers.

How would you describe the state of AMD's products back when you joined?

The emphasis seven years ago was highly weighted to mobile devices, which are very, very important. And they were important because they drove the company to focus on energy efficiency, which turns out matters in every application you have, whether you're in a laptop device, an embedded device in a piece of machinery or in an airplane or an automotive. Or, if you're in a data center, even there, power efficiency matters, because it affects how many of those CPUs or graphics engines you can pack in to a very small space. So the focus of the company that was really prioritizing, driving to mobile was very important and drove the company in energy efficiency. But what we had to do was get the roadmap to where it didn't lose any of that energy efficiency but brought the performance into the highest realms of what you can get in any commodity processor in the industry, and that's what we've done.