During CES 2021, Advanced Micro Devices CEO Lisa Su spoke with a group of journalists about a wide range of issues, including AMD’s product road map, the core counts for its processors, launches of the PlayStation 5 and Xbox Series X/S, and the company’s relationships with Apple.
I was part of the group Q&A that makes up this interview.
At Su’s CES keynote, AMD showed off some impressive Ryzen 5000 mobile processors and teased the performance of its 3rd Gen Epyc server chips.
Su has had a very good run in the past five years, after AMD introduced the Zen architecture that made its processors far more efficient at processing instructions per clock. This gave AMD a performance advantage over Intel for the first time in years. Then Intel began stumbling with its technology transitions (leading to the appointment of Pat Gelsinger as CEO this week).
Market analyst firm Mercury Research noted that AMD has gained market share in client computing chips for 12 straight quarters, ending in Q3 2020. For that quarter, AMD’s desktop share was 20%, up two percentage points from a year earlier and up seven percentage points from two years ago. In Q3, AMD’s mobile share was 20.2%, up six percentage points from a year earlier.
For the turnaround, the Semiconductor Industry Association awarded Su its Robert N. Noyce Award, the chip industry’s highest honor.
Here’s an edited transcript of the conversations.
Lisa Su: High-performance computing goes through so many different threads. This year, there’s a large PC thread. There’s also the overall gaming thread, as well as the datacenter, cloud, high-performance computing thread. It’s also a big year for Zen 3 and our launch of the Ryzen 5000 mobile processors, as well as a preview of our third-generation Epyc rollout.
Question: What’s your stance when it comes to competition from non-x86 designs, especially in the server space? We’ve seen the launch of Amazon’s Graviton2, Ampere’s Altra, the eight-core offering that sits right at the top of that market. Arm has positioned itself with substantial generational gains and is quickly appearing to be a substantial threat to x86. How does AMD position itself if it remains solely targeted on x86?
Su: This whole discussion recently about ARM and x86 and other compute architectures is a great validation of how important computing is throughout the ecosystem. That’s the first overarching principle. As it relates to our road map and our focus, our focus is around high-performance computing, pushing the envelope, ensuring that we get significant generation-to-generation improvements that are not only CPU core-related, but also broader, system-related.
What we’re seeing with the current situation is that as you see the cloud, datacenter, and the overall expansion of the high-performance computing market, you’re also seeing some more ability to target specific workloads. The fact that there are custom designs out there, or let’s call it purpose-built designs, is an indication of how much computing is growing. You’re going to keep seeing that. We believe in that too. That’s part of why we have our semi-custom solutions, which are working on customization.
The overarching thing is x86 is a very strong ecosystem, and we’re going to continue to invest heavily in that performance road map. But we also see a larger opportunity to customize solutions with our customers that are looking for specific workloads. That’s always been part of our strategy, and it’s somewhat being shown out that it’s a strong strategy as well for very workload-specific compute.
Question: Do you have any semi-custom options in the server space?
Su: There are opportunities for us in the semi-custom space as it relates to — obviously game consoles have been the largest piece of it. But there are opportunities outside of game consoles, including in the datacenter.
Question: I was thinking about manufacturing capacity, internally owned. It seems like when you have it, you don’t want it, and when you want it, you don’t have it. I remember days long ago when AMD’s market share was capped because it couldn’t make any other manufacturing plant. There wasn’t as much contract manufacturing out there. What is the situation now? Do you have concerns that your market share, at 22% and growing, is going to be capped because you don’t have that access to capacity?
Su: First of all, we’re very happy with our manufacturing strategy and our manufacturing partners. It’s been a competitive advantage for us, for the choices we’ve made. The broader question — there is a tightness in the supply chain, particularly around some of the consumer products, around PCs and gaming. That’s a result of the overall demand environment and not necessarily any issues from a manufacturing standpoint.
As it relates to semis, we’re putting significant capacity online to satisfy that demand. It’s an opportunity, and we love that opportunity. But it takes some time to catch up. That’s what we’re seeing. Do we think that there’s a market share cap? We don’t think so at all. It’s just a matter of making sure that we forecast what’s needed well and work closely with our partners to put that capacity online.
Question: Is AMD also hindered, like many others, by substrate shortages? That seems to be a pervasive theme for the last few months.
Su: You’ve seen some reports of substrate shortages, and we also see tightness in the substrate market. Again, this is more a function of the demand. It’s outstripped overall worldwide capacity. You do see that more capacity is being invested in and coming online, including AMD investments. But it takes time to get those online. The industry is overall reacting quite broadly — this is across the industry — to ensure that we put more capacity online, and I expect that will continue to happen through 2021.
Question: Can you comment on when we can expect supply to improve? What kind of a timeline are we looking at?
Su: We’re shipping a lot of parts. Volumes are continuing to increase across gaming, graphics, and CPUs. We expect that to continue to happen through 2021. There will be tightness certainly through the first half of the year. But we continue to ship more into our OEM partners, as well as our channel partners, to increase overall supply. We completely understand why consumers want more, and so it’s very high on the priority list to have supply catch up to demand.
Question: Is it fair to say that AMD is prioritizing output to OEMs at this point, for prebuilt and that part of the market?
Su: I wouldn’t say that. I would say we’re trying to satisfy as much of the customer set as we can. That means some real-time prioritization. But I would not say we’re prioritizing prebuilt versus DIY.
Question: With Ryzen 4000, the Pro variations were announced alongside the H and U series. For 2021, are you planning to release Ryzen Pro variants? Also, what is AMD’s strategy to counter Intel’s growing investment in its vPro platform?
Su: We have Ryzen Pro variants of the 5000 series. One of the points we made in the keynote is our overall design wins for the Ryzen 5000 series, which are 50% higher than last year — 150 designs. There’s significant growth in commercial designs in there, as well as gaming designs. We’ve been working closely with our partners on security solutions. We feel good about the commercial quality and security that we have with the current Ryzen 4000 and going into Ryzen 5000. You’ll see a number of designs from the top OEMs coming out throughout 2021 with Ryzen 5000.
Question: Are there any new initiatives as far as trying to enable the wider ecosystem with the Ryzen Pro platform?
Su: Yes, there are a couple, but the team will talk more about them as we get closer to announcing some of those partnerships.
Question: Some vendors like Asus have suggested that import tariffs could increase the price of GPUs. With such high demand for the latest generation, GPU prices are already rising in third-party markets. Do you have a plan to mitigate an explosion in pricing for gaming GPUs? And do you think that mining is playing a role in some of the price increases?
Su: As we came into the new year, there were some expirations or changes in tariff policies. We’ve spent quite a bit of time to ensure that we have a flexible supply chain. I don’t think that’s a significant issue as it relates to AMD. More broadly, we’re very committed to trying to keep the GPU pricing as close to SUP as possible from an overall fairness standpoint. One thing we’ve done, for example–typically when we start our GPU launches, we have our own Radeon-branded MBAs, and then that phases out to go into AIB versions. We’re actually not phasing out our Radeon 6000 series MBAs for the purpose of trying to ensure that. As stock becomes available, we’ll offer it on AMD.com at SUP, and we’ll encourage our partners to do that as well.
There are some COVID-related logistics and other commodity components that have increased pricing. Some of that is what’s blowing through. These are things we’re living with in the current situation, and hopefully as we get to a more normal environment in the second half of the year, we’ll see some improvements.
Question: Your competitors have been integrating AI-specific optimizations into hardware for a short while now. They see it as a huge opportunity for growth. AMD doesn’t appear to have that right at this moment. Can you give us some color on your thoughts about AI and hardware moving forward?
Su: You’ll see us talk more about that in the coming years. We’re certainly making investments in AI. We have our CDNA architecture on the datacenter GPU side that we launched in mid-November last year. You’ll see more upgrades to that. You’ll also see us add more AI capability in our CPUs and GPUs. That is an investment area for AMD, and you’ll see more as you bring those out in new products.
Question: You’ve had some very big dual launches in game consoles two generations in a row now. What was different about launching this time? Did you learn some things from the last time around that were helpful? Are you happy with the holiday season and how the console launches went?
Su: We’re thrilled with how the console launches went. You’ve heard separately from Sony and Microsoft, their discussions about the size of the launches and the reception of the products. From our standpoint, if you think about it, with just the amount of new hardware that had to come into place — millions of units of both consoles, or all three consoles — that needed to ship, it came together very nicely.
As far as what we’ve learned, there is higher demand than we thought, and we’re trying to put more capacity in place for that. But we’re very happy with the launches, with the partnerships with both Sony and Microsoft. They have somewhat different strategies, but we’ve partnered very well with both of them. This is a big cycle. That says a lot about how much technology we’ve been able to integrate into the console form factor.
Question: It feels like we’re teetering on extreme frustration with getting Ryzen 5000, with getting Radeon 6000. Is there anything you’d want to say to people who are so exasperated that they can’t get it?
Su: I do want to be very specific. The main thing I want to say to our fans and our enthusiasts is, I get it. I completely understand that there’s a huge desire for more Ryzen 5000 and Radeon 6000 hardware. We’ve shipped a lot into the channel. It takes some time for it to work itself through, and that was some of the logistics I was talking about. There will be more. You’ll continue to see refreshes as we go into the first quarter and the first half. It’ll still be tight, but there is a lot of product coming to the market. We appreciate that there’s so much interest and desire for these products. We look forward to getting more into the hands of our users.
Question: In the pre-briefing last week, an AMD representative talked about how the company is undergoing an evolution in how it positions Epyc, with more of a focus on business value. Could you talk about what this evolution means for how AMD will position and sell Epyc processors in the future, and whether that will result in anything like prevalidated or bundled solutions, like what Intel has been doing for a little while?
Su: We’re very excited about Milan. Since CES is a bit more of a consumer slant, we showed just a preview of the Milan processing capability. We like how it’s showing up with customers. We started shipping in volume in the fourth quarter. We’re talking about expanding the depth and breadth of where Epyc plays. Traditionally, we’ve been strong in cloud datacenters, and you’ll continue to see strength in the cloud, but we’re increasingly broadening the enterprise, large enterprise focus of Epyc. That means more business solutions. You’ll see when we do the larger launch for Epyc later this quarter. You’ll see some of that with our partners and how we’re putting together more full solution capabilities. You’ll see us continue to broaden our focus on the large enterprise, the overall enterprise market.
Question: By solutions, do you mean things that are pre-validated, a mix of software and hardware?
Su: We’re talking about pre-validated solutions with the partner ecosystem, yes. A mix of hardware and software.
Question: Is that different from how AMD had been approaching the market previously?
Su: It’s faster. If you look at our previous Epyc launches, we would launch the product and then you would do a lot of the optimization. We’re seeing much more time to market for this launch, and that’s true about OEM platforms and platform availability, as well as the pre-validated solution capability.
Question: Given the high-performance silicon for mobile, up to 16 threads in the CPU, and with discrete GPUs to follow, do you see the opportunity for a branding category or exercise, much like Nvidia’s RTX Studio, for content creators?
Su: We believe that content creation is an important category overall, whether you’re talking about CPUs or GPUs. As we’re adding all of this capability, there are opportunities for — whether it’s branding or specialized SKUs for content creators. We see all of those as opportunities. Primarily it’s been, let’s just get as much processing horsepower as we can on both the CPU and the GPU side with good PCIe Gen 4 between, so that we can use all the horsepower.
Question: Part of AMD’s success that’s helped you grab market share is the increased core counts year over year. Today you confirmed that Epyc Milan tops out at 64 cores, just like its predecessor. We’ve also seen that the desktop PC chips have stayed at a maximum of 16 cores on mainstream platforms. Given thermal power restrictions, have we reached a functional limit for core counts, or can we expect more to come in the future?
Su: If you look at what we’ve done between Zen 2 and Zen 3, as well as between the second generation of Epyc and the third and Ryzen 4000 to Ryzen 5000, we’ve focused on increasing single-threaded performance, as well as improving some of the latencies and overall systems such that we’ve gotten tremendous generation-to-generation performance in the same process technology. All of that is in 7nm, but we’ve been able to increase performance by more than 20%, depending on which metric you’re looking at.
There will be more core counts in the future. I would not say that somehow 16-core and 64-core are the limits. They will come as we scale other parts of the system as well.
Question: How have you maintained design leadership in multiple generations now, even when some of your executives have departed who started the work with things like Zen? Even with those departures, you’ve maintained leadership in design. What’s the approach that’s been successful?
Su: I’m very proud of our engineering teams. On the CPU side, Mark Papermaster, Mike Clark, and the team have done a phenomenal job with a very ambitious CPU road map. The team has executed very well. We also know, though, that we’re as good as our current architecture. As excited as we are about Zen 3 getting out into the marketplace, all of the focus is on Zen 4 and Zen 5, ensuring that those are also extremely competitive. On the GPU side, we’re very pleased with the work that David Wang and the team have done.
Our focus, from an engineering side, is to set up long-term road maps. You’ve heard me say this before. It’s a five-year road map that we’re looking at. How do we pick the right mix of — you have to take some risk. You have to take some risk to get the innovation performance where it needs to be. But you also want to be predictable in when the products are going to come out. It’s that give and take. What are the bets you make, and how do you make sure you track progress?
Our aspirations are continuing to push the envelope. Many of you have asked about our progress on GPU architecture. I’m extremely happy with the progress we’ve made with RDNA 2 in terms of performance per watt and overall performance. We have a lot of focus on RDNA 3 and beyond, to ensure that we continue to drive those leading architectural capabilities.
Question: Could some customers be a bit disappointed that Cezanne carries over older GPU tech, even though it’s stronger on the CPU side? Is there a notion that things haven’t moved as quickly as hoped, given what you’ve done recently year on year?
Su: The way to think about that is — Cezanne is a phenomenal product. We’re very happy with it. Renoir was a very good product too, but Cezanne just takes it to another level. The choice of what you put in on a given product, those choices were made quite a while ago. We feel like we made a good set of choices. I don’t think there’s any concern about that.
As we move forward, this is all about what user experience we can bring to the PC market. That’s important for all of us in the PC ecosystem, to recognize that this is an opportunity to show people what you can do and how experiences can get better. That’s what we’re thinking about when we think about the beyond-Cezanne road map and what are the next-generation user experiences we’re trying to put out there.
Question: One of the goals since the launch of Zen and Ryzen has been this regular product cadence every 12 to 18 months, with a new generation. Part of that is being able to reuse previous elements of the design to ensure time to market. We’re seeing that today with Ryzen 5000 mobile taking parts of Ryzen 4000 mobile, the graphics. Is it this philosophy that enables that fixed product cadence? And if so, is that cadence sustainable for the better-than-average performance gains when you don’t update everything every cycle? The fact that you can’t update both, is this more indicative of AMD being design resource-constrained compared to the competition?
Su: I’d put it a slightly different way. I don’t think there’s — when we think about product cadences and road map cadences, for example, one thing that was very important to us with Cezanne was shipping in production early in 2021. The reason for that is, if you think about the entire OEM cycle, we have a whole bunch of platforms that will now launch throughout the first half of 2021, and that’s a nice way to build cadence.
As it relates to a choice on, do we put the latest generation of graphics in or not, it’s really a choice. It’s a matter of timing, of where we want to be on that particular cadence. Nothing fundamental, not design resource-limited. More on the notion of what we think is needed at a given point in time. Some people might have expected that we would have left Renoir in the marketplace a bit longer because it’s a fantastic product. But we thought there would be high demand for Zen 3 in the notebook form factor, and so we prioritized that for the Ryzen 5000 series.
Question: What are your top three objectives from the board of directors, for you and for the company?
Su: For 2021, we’re very focused on overall customer adoption, particularly in the enterprise and commercial markets, for our server and PC space. We’re focused on ensuring that we have enough supply. We had a good conversation with Gordon about how some of our consumer customers would like to see more supply in the market, and we’re working through that. We’re also very excited about closing our Xilinx acquisition. That’s the next big leg in the stool of AMD.
I don’t want at all to take for granted — a lot of the folks here are tech press. There’s a lot of technology that we’re planning to roll out in 2021. But overall, when I think about where we are in the business, we’ve made very good progress over the last few years in trying to deepen our customer partnerships and get people to trust us on their most important platforms. 2021 is a year where you’ll see that a lot more. It’s something I take very seriously. We want the largest enterprises and datacenters in the world to trust AMD as their preferred supplier.
Our goal is to not see too many valleys at AMD. We’re very focused on scaling the company. That’s what I try to remind people. We have scaled the footprint of the company, and we’ll continue to scale the footprint of the company.
Question: On the Xilinx acquisition, there’s been some concern raised among various analysts that perhaps this isn’t the best timing for a merger or acquisition of this size for AMD, because of maybe becoming distracted by all of the tasks involved with that. Especially when you have been focusing on execution so well recently. What would you say in response to that? How do you plan to avoid that type of distraction?
Su: We’re continuing to stay focused on execution. I want you all to know that that’s overall job one. On the other hand, we do have a very talented management team. We have the desire to be a much bigger footprint in this industry. Xilinx is the right next step for us. I’m quite confident that we can both execute on the base AMD business, as well as bring over Xilinx. The fact that Victor Pang, the current CEO of Xilinx, will join us is part of that strategy to ensure that it’s a seamless transition.
These are the things that leaders have to do. We have to expand and scale. We have the capability to do that. I very much think we can do both.
Question: What are your thoughts on using hybrid processors in the mobile space moving forward? Some high-performance cores alongside energy-efficient cores.
Su: There is some good work going on in that space. We continue to look at what’s the right balance between big core, small core, medium core, all that stuff. My team will be willing to talk about more as we go through it. Overall, we’re focused, again, on the experience we’re trying to drive. The Zen design point is quite well-balanced right now. If you look at performance, area, power consumption, it’s a very well-optimized design. We would have to see a significant value add to add something else into that max. Those continue to be architectural concepts that we look at with each generation.
Question: AMD has partnered with Microsoft, and between yourselves, Intel, Qualcomm, you’ve all announced that the Pluton security module is coming to hardware. You’ve obviously worked deeply with Microsoft on enabling Pluton in the previous and current generations of Xbox. Where does AMD stand with Pluton? As the key partner with Microsoft, will you be first to deploy in more traditional x86 markets?
Su: We’re very committed to working with Microsoft on deploying the solutions. The first solutions will be coming over the next couple of years. We’re not getting more specific on the future road map at this point. They’re deep in design.
Question: It’s hard not to be amazed by what Apple has done with the M1. They could bring that design to other product categories, such as workstations, but also graphics. I know AMD and Apple have had a strong partnership. How does that relationship with Apple evolve now that we’re in this post-M1 era, where Apple is now a performance player in mobile players, and if they wanted to, they could try to design AMD out of the graphics partnership you have?
Su: This is an indication of just how much processing innovation there is in the market. We continue to partner with Apple. They continue to use us in their discrete graphics configs, as we’ve said before. As it relates to how we think about the overall market, though, I’ll come back to the earlier question about how we think about ARM in these various places. This is an opportunity for people to innovate more. When I say that, it’s innovation on the hardware side, as well as innovation on the system side. It’s more about that than it is about x86 or ARM.
From our standpoint, we think that there is a tremendous amount of innovation that can happen in the PC space. It requires more optimization. If you look at today’s PC model, we have a lot of choices in the ecosystem. People are able to use the same family of processors in a lot of different places. You’ll see a bit more specialization, perhaps, as we go forward, so you’re hitting some of the different workloads and use cases. That’s what I see the evolution looking like as we go forward over the next couple of years. It enables more differentiation at the system level, which is a good thing.
Register for GamesBeat’s upcoming event: