On Wednesday, AMD chief executive Lisa Su sat down with a small group of reporters following her CES keynote, in which she formally announced the Radeon VII GPU and the third-generation Ryzen CPU. During our roundtable, Su disclosed that there will be more cores to come with the third-generation Ryzen chip as well as the presence of a ray tracing GPU under development within AMD.
But there were even more details and context from the roundtable—and, well, if you’re as interested as we are in what’s coming down the pike from AMD, you’re going to want to read our full transcript of the event here.
(Note: Like all transcripts, there can be crosstalk and some fumbling about. While we’ve tried to transcribe Su’s remarks verbatim, we’ve more heavily edited the questions from the journalists, where it makes sense. We’ve also used “reporter” to indicate a question asked by any reporter.) And even a 30-minute transcript equates to an enormous block of text, so use our Table of Contents links to jump from highlight to highlight.
Su: What I would say is that we’re very excited about Radeon VII. I would probably suggest that he hasn’t seen it yet. And look, there are 400 million gamers that are using Radeon. We have a pretty big installed base, and when you look at that, much of the conversation you guys have asked me time and time again, what’s the next high-end GPU from AMD? And my view of it is, we had to bring Radeon VII to market; it’s important for our fan base and for high-end gamers. We think it performs very, very well. If you look beyond, at the memory technology and what we have, I think we feel good about it. And I think you guys will see soon enough for yourself.
Reporter: How do you see the Radeon VII’s performance compared to the Nvidia GTX 2080?
Su: We’ve shown some of the data; I think some of the data is on our press release, and I think what you’ll see is very comparable in some games—we win some, we lose some. But the key about the 2080 is, yes, there’s a lot of [feature?] function; ray tracing is important. Radeon VII also does a lot of things across the content-creation workloads; you saw some of our OpenCL numbers that were really, really phenomenal, and we think that it’s very competitive.
Su: We have ray tracing, too
Reporter: Where does AMD stand on ray tracing? Visual quality could supersede the FPS wars.
Su: Well, I think hopefully we spent a lot of time today talking about our vision on gaming, okay, and our vision on gaming is very broad, and we think about on PCs, consoles, cloud, and how we deliver content that’s all those pieces. I think ray tracing is important technology; it’s something that we’re working on as well, from both a hardware/software standpoint. I think the important thing though, and why we’ve talked so much about the development community, is technology for technology’s sake is okay; technology done together with partners, and really getting the development community fully engaged, I think is really important. So you’re going to see a lot more gaming discussions from us as we go through this year, and into the future, and that’s kind of how we do it. We view it as a broad ecosystem, we don’t focus on just the one technology. [inaudible]
Reporter: When Vega first came out, you had two cards…
Su: We had a [Vega] 64, and a Vega 56.
Reporter: Vega faded from the conversation, and now it’s back in the spotlight. How is AMD going to keep it there?
Su: Well, I think the way you should look at it is, you know, again, broadly across the market. So, you know, Vega actually has done really well for us. I mean, you see it in a bunch of different places. You saw it in the discrete graphics cards, we talked about Vega 64 and Vega 56; you saw it in some workstations, certainly coming from Apple; you have seen it in smaller form factors as well. So, you know, Vega architecture is in our notebook graphics and is in the work that we do there. We showed the Google Project Screen that’s using Radeon Pro Vega, you know, as well, and we have other cloud environments where we’ve used Vega. So, you know, as architectures go, they get better with time. You know, it’s like that whole idea of more developers and software optimizations, and Vega has definitely gotten better with time.
I think, you know, Vega VII, we launched it in the fourth quarter for data center applications and in that form, it’s MI60 and MI50, and that’s going into HBC markets, that’s going into cloud, that’s going into machine learning opportunities, that’s going into virtualization opportunities. So that’s one really important segment for Vega. And then today we talked about Radeon VII for gamers. And I think that just gives you a broad view of the ecosystem.
Reporter: The 7nm Vega was not supposed to be a consumer graphics card.
Su: You said that. I never said that.
Reporter: It wasn’t supposed to be a consumer product in 2018.
Su: What I think we said was, we would do data center first, right? [To PR] That’s what we said, right?
Reporter: You didn’t say anything else after that.
Su: We didn’t say anything else after that. Actually, if I remember correctly, what I said at Computex was, you will see 7nm gaming parts from us and we always planned to bring Radeon VII to the market. That was always in the plan. You know, our goal is, you know, frankly, we want to delight our fans and so we wanted time to get the software optimizations ready and we did quite a bit of that. With every new architecture there are things you do better. We wanted to do the work that we did with Ubisoft and Capcom, you know, on those optimizations and then you know, bring it out when it’s ready.
And gaming also is you know, a high volume market and so we wanted the technology to mature a little bit as well.
Reporter: Last year, you laid out a great presentation with specific product plans through the first half of the year. This year your plans are a lot more fuzzy. Why?
Su: Look, I think what we want to do is make sure the consumers have a good idea of when to expect these products. Radeon VII very quickly, you know, you’ll see it on the shelf in February. I think with the next big things, our third-generation Ryzen, I know there’s a lot of anticipation about that.
The key point is the part looks good. Yeah, the part looks really good and we’ll put it out sometime in the middle of the year. We haven’t decided the exact month yet. You know, there’s a little bit more tuning to be done; we want to get the clocks in the right place. We want to get, you know, all the performance in the right place. But it looks good. And you’ll see it in the middle of the year. And similarly, the next-generation Epyc is very much in development. And you know, we think things look really good there. Look, you see us all the time! You’ll get more information soon.
Reporter: Any comment on Threadripper?
Su: Well, content creators are important to us. And Threadripper has done really well. Actually, I’ve been pretty pleased at how second-gen Threadripper has done, when you think about it as a sort of niche type of market, but there are people who want more performance and so as second-generation Threadripper has come into the market, it just looks like more people have adopted [it]. I think the thing is, keeping our infrastructure consistent is important. And you know, we’ll talk more about Threadripper as we go throughout 2019.
Reporter: Nvidia has ray tracing technology. Does AMD?
Su: I’m not going to get into a tit for tat, that’s just not my style. So I’ll tell you that. What I will say is ray tracing is an important technology. It’s one of the important technologies; there are lots of other important technologies and you will hear more about what we’re doing with ray tracing. You know, we certainly have a lot going on, both hardware and software, as we bring up that entire ecosystem.
Read on to find out where AMD is investing in the PC.
Where is AMD investing in the PC?
Reporter: AMD’s financials have looked really good as Ryzen has ramped. What has that extra influx of money done for you?
Su: Well, I think what you’ll see is that we’re pretty deliberate in choosing how we roll things out, right. And so, you know, as with Ryzen and Epyc and Radeon, the thought process is, let’s do things in steps. And so, you know, the conversation about, hey, you know, Vega wasn’t supposed to be in 7nm for gamers. Actually, that’s not really true; it was, but we were going to do it in steps.
And you know, the same thing in terms of, you know, how we’re bringing 7nm technology to market. We’re spending a lot more time on software and a lot more time with software developers to optimize on AMD. I think that’s, that’s really critical. I think we are doubling and tripling down on graphics. That’s one of the reasons we talked so much about gaming today. But we’re thinking about graphics in the broad sense of graphics. And so it’s, it’s PC gaming, but it’s also everything else that goes along with it, including cloud gaming, including consoles, and, you know, we think once you put that entire ecosystem together, Radeon is going to be the most powerful gaming brand out there.
Reporter: Is there anything else you can say about the specifics of the 7nm architecture? What benefits will consumers actually see?
Su: Well, I think, you know, you see a couple different things and it manifests itself in a few different ways, depending on which market you’re looking at. Like, for example, when we talk about the server market, you know, 7nm just gives you an incredible amount of density and so you can double the number of cores. You get 4X floating point; there it’s more from sort of a data scientist standpoint—you’re just going to be able to run your simulations much, much faster and get more done.
When you go to, for example, the desktop, today we showed the preview of the third-generation Ryzen and you could see one of the things that we’re trying to show is that, you know, you could get, let’s call it, very similar performance. Maybe slightly better [inaudible] and what performance at about 30% less power. What that means is you can take the less power if you want, right, and that’s useful if you’re doing small-form-factor PCs, or, you know, you can up the frequency and get more performance. Certainly we look at those tradeoffs.
Reporter: How is the 7nm process proceeding from a manufacturing standpoint?
Su: Yeah, we’re quite happy with how things are going. We now have a number of different products that are going through 7nm and I would say, you know, things look good.
Su: I think this whole notion of breaking chips up is, you know, sort of the wave of the future, right? We did it first with HBM [high-bandwidth memory] on the graphics side, and then with these chiplets that we’re using here on the Epyc and Ryzen side. I think chip stacking will continue to be important going forward and we will look at it as different ways to put these architectures together.
But I think the foundation is something that we believe very much in.
Reporter: As followup to that, Lakefield takes a “big-little” approach where a Sunny Cove processor will be combined with four Atom chips. Could you work with ARM on something similar, or is there another approach you might try?
Su: Well, I would say, anything’s possible. The question is, you know, is it necessary? I think what we’ve done with Zen is build something that is actually pretty scalable. I mean if you think about it, we scale Zen from a notebook form factor all the way up to a server form factor and we do pretty well. So I think the idea of putting different chips together, you know, chips, I/O, memory, how you partition that, how you look across all different form factors, that’s something that our team feels really passionate about.
Is there a chiplet GPU?
Reporter: On the separating chips question, is there a chiplet approach that can be applied to graphics? It’s a tough problem to solve.
Su: I think we continue to look at every generation in terms of what’s the right partitioning, right? So whether it’s splitting the GPUs—people have asked us before, you know, [whether] Mike [Rayfield, senior vice president of Radeon Technologies within AMD] could do a CPU little die, and a GPU little die, you know, those types of things. I think it’s what makes sense; you really have to look at what the interconnect is. We have our Infinity Fabric that Mark [Papermaster, AMD’s chief technical officer] has talked to you about, and how that performs relative to the various other pieces.
Reporter: Would that require a paradigm shift in how graphics hardware is approached?
Su: I think that’s all part of what needs to be kind of thought through.
More cores coming for third-gen Ryzen
Reporter: So during the keynote, somebody messaged me about the core counts on Ryzen…
Su: So it only took like 30 minutes for somebody [to ask the question]!
Reporter: So can you give us any indication where Ryzen third-gen is going to be at?
Su: If you look at the evolution of Ryzen, we’ve always had an advantage in core count, and so in this particular case we wanted to show sort of a head-to-head comparison: eight cores, 16 threads. Some people may have noticed on the package that there’s some extra room.
Reporter: Yes, we’ve already done the math.
Su: So there is some extra room on that package. And I think you might expect that we will have more than eight cores. I didn’t say how many more.
Reporter: And two memory channels will be enough?
Su: As I said, more to come.
Reporter: One of the more interesting announcements over the last year was the partnership between Intel and AMD bringing Radeon graphics to Intel CPUs. But we’ve seen only a handful of notebooks use the Kaby Lake-G part. Is that relationship going to continue?
Su: Yeah. So look, I think it was an opportunity for us to get Radeon into more applications, you know; it’s part of our “Radeon everywhere” message. And I think it appeared in some notebooks and those were pretty high-end, premium notebooks that really wanted the form factor because of the power.
Yeah, we’ll see how that goes. We don’t have anything more to say about it at this point in time, but we’ll see how that goes.
Read on to learn more about Ryzen notebooks.
Where are the Ryzen notebooks?
Reporter: We see laptop makers announcing products with AMD inside them, but the big splashy products continue to be all Intel. Why is that?
Su: You know, it’s a journey. It’s a journey, but we’re making progress. And yeah, the way I look at it is, you know, my commentary to the OEMs is we care about PCs, and we are investing in PCs on both the desktop and notebook form factor.
I would say that the first-generation Ryzen notebooks were good, some of them were very good. I think the second-generation Ryzen notebooks are going to be a lot better, in a broader set of form factors.
One of the things I get to do at this show is that they show me some of the new form factors. I was in a meeting with one OEM yesterday and they were going to triple their designs. That’s really cool. I’m happy with them. And some of them are really, really nice, and I’m looking forward to those coming out this year. But it’s a journey and you know, from my standpoint, we make good progress; we gained notebook share in the fourth quarter as we really started ramping those guys, and I think second-gen Ryzen is better than first, and third-generation Ryzen mobile will be better than second.
Reporter: I think there’s a still a perception that within notebooks, AMD is still behind Intel. What can you do to shift that perception?
Su: I think it’s a familiarity. So what’s really interesting is if you look at the different markets, right? Like you look at the desktop enthusiast market, where the people who read your columns, who are really, really savvy in tech, they love us.
Actually, truth be told, I watch our rankings very, very closely at the top e-tail websites—I didn’t go today at what Amazon says, but for the last while I think we’ve been number one. It helps that it’s a great product, at a good price, and in stock.
So yeah, so that adoption is going really well because, you know, folks are really savvy and they’re looking at the specs and they can see the value. Sometimes on the notebook side you go through so many different people, right? You go OEM, to retail buyer, to retailer, to the blue shirt at Best Buy or Dixons that has to explain it, to the guy at home who doesn’t know much about chips.
There’s a lot of people to explain that message. And that’s part of our job. Part of our job is to get more people familiar with what AMD can bring to the party. And, you know, we believe we’re making good progress, but it doesn’t happen overnight. So yes, we have work to do, but we’re making good progress.
On Nvidia opening itself to FreeSync
Reporter: Nvidia opened itself to FreeSync. Can we expect any reaction from AMD to that? What do you think about it?
Su: Where are my graphics gaming people who love talking about this? Look, we knew FreeSync was the right answer. We’ve known FreeSync was the right answer for a couple of years. The fact that others have decided that FreeSync is the right answer, I think, says that we made the right choice a few years ago.
We believe in open standard. You know, we believe in open ecosystems. That’s been a mantra. So we have no issue with our competitors about FreeSync. And we think that just means that, you know, it’s better for gamers and we did a good job.
Reporter: It seems that there’s a huge disparity in quality in FreeSync monitors. Is AMD going to be working with them to resolve this?
Su: Well, if you really think about it, what problem are you trying to solve? I mean, the way we look at it, is there’s a whole host of monitors and TVs out there—over 550—with very different price levels: entry level, to super-duper premium, and gaming monitors. And we certify; we make sure they work with FreeSync, and you will get the experience you pay for on monitors. I don’t actually see a problem.
Reporter: Nvidia’s Jensen Huang said that FreeSync doesn’t work.
Su: I don’t believe we’ve seen that. So, yeah.
[PR interjects, noting that color, flicker, and other characteristics are governed by certification. Audio is largely inaudible.]
Reporter: Some use “adaptive sync” and “FreeSync” interchangeably. Is that OK?
Su: It’s a good point. I think both definitions will be used interchangeably, and that’s okay. So adaptive sync technology, FreeSync is a free version; there are other adaptive sync technologies, and [inaudible]. Is that right?
[PR: Yes, The other thing I would add to that is DisplayPort Adaptive Sync standard makes no qualitative recommendations on the experience. (Inaudible.) All the things that gamers care about, that something we uniquely bring to the table with our certification program versus just, yep, it’s adaptive. And so that that is where the difference lies. ]
Reporter: We always found Ryzen to be a really interesting product, but it always feels that we can never get a gaming machine for review. Is it a manufacturer issue?
Su: Can we help here? We might be able to help get some for review.
And that is also one of the value propositions for Ryzen Mobile. If you look at some of the systems like the Acer [inaudible], some of the entry-level categories. There’s a couple more in development, based on the second-generation Ryzen. So we should absolutely help you get some systems for review.
Su: So actually we have a really, really good gaming monitor partnerships with a bunch of folks. Samsung is one but there’s a few others as well. You’ll see some larger monitors and capabilities from us this year.
Reporter: Is the chiplet capability affecting the semi-custom business like consoles? Is there more interest?
Su: You know, we believe that it will accelerate the semi-custom breadth of opportunity and that you can use our CPU, and our memory, and our fabric, in a custom ASIC. So yes, that’s part of the conversations we’re having.
Ray tracing needs software support!
Reporter: Nvidia is taking criticism for bringing out ray tracing hardware without a lot of software support. Is it fair to say that AMD is waiting to release ray tracing GPUs until there’s a software ecosystem in place?
Su: It’s fair to say that. I don’t think we should say that we’re waiting. I would say that we are deep into development, and that development is concurrent between hardware and software. And so for us, it’s, you know, what is the consumer going to see? The consumer doesn’t see a lot of benefit today because the other parts of the ecosystem are not ready. I think by the time we talk more about ray tracing, the consumers will see that.
Reporter: And is there a time frame for that yet?
Su: Not yet. [PR: Yes, we do have a time frame, but we’re not prepared to share it.]
Reporter: Is AMD going after the single-threaded performance crown?
Su: Our first priority is overall system performance. But we know how important single-threaded performance is. And so you will see us push single-threaded performance. And I’m not making any predictions.
Reporter: It seems that OEMs are still working more closely with Intel in terms of the design of new hardware. Is AMD also trying to improve its collaborative design?
Su: Much more so now. Much more so now. I think when we were sort of catching up with it in performance, there was more, you know, the chassis was already available, and we needed to figure out how to slot the AMD processors already in it. And now, as we look at our forward-looking roadmaps, we have some interesting things coming beyond what we talked about today, for our second generations, and we are working with a couple of folks to design that [inaudible].
Reporter: How has the supply relationship between GlobalFoundries and TSMC shifted as a result of the chiplet development agreement?
Su: What we’ve said, and not to be cagey about it, is GlobalFoundries is a good partner. They’ve done a good job for us in 14nm; we really appreciate our partnership, we’re really working on amending the supply agreement to accommodate the change in strategy that’s in place, and we’ll be talking more about that shortly.
Reporter: There’s been a race to make laptops thinner and thinner. Is there merit to that?
Su: I think, yes, there’s a lot of merit to the fact that, you know, people like sleeker form factors. They like to carry something that’s lighter, that has a cooler feel to it, but not at the expense of performance. And so, you know, our focus has always been, yes, we need to do that. Yes, that means that focus on power, and power efficiency, but it’s got to have a minimum performance.
We’re not trying to be a tablet. We’re trying to be [inaudible].