Q&A with AMD CEO Dr. Lisa Su

Computex 2024 Press Roundtable

Companies mentioned: AMD 0.00%↑, MSFT 0.00%↑, NVDA 0.00%↑, INTC 0.00%↑

You'll probably hear this a lot, but the trade show known as Computex, in this year 2024, is a big one. A number of key companies in our industry are holding official keynotes (and a few unofficial ones) to talk about their latest and greatest end-to-end AI solution. There's so much AI here, I'm not sure I recognize them as letters of the alphabet anymore. Nonetheless, there's substantial innovation going on! At this year's Computex, AMD's CEO Dr. Lisa Su held the official keynote of the event.

During the Keynote, Dr. Su presented AMD's end-to-end AI solution and updates, covering consumer hardware (desktop and notebook), enterprise (CPUs and AI accelerators), and the updates to the software ecosystem as well as partners. This year's Computex is all about AI PC, and the adoption of NPUs powerful enough to run Microsoft's copilot+.

From AMD's consumer line, we saw the announcement of its new Ryzen AI 300-series notebook processors, featuring its next-generation Zen 5 cores, next-generation XDNA2 AI engine hardware, and updated RDNA3 graphics in a monolithic die. AMD is promoting this chip and line of chips as the most powerful AI engines available in notebook processors to be available this year - and to signify the importance of AI, they changed the branding from Ryzen to Ryzen AI. These chips will be available in July. AMD also announced its next-generation Zen 5-based desktop processors, with a headline 16% IPC average uplift across all workloads. More details on these announcements in a separate piece.

From AMD's enterprise line, the company showcased its new MI325X, which features double the memory of the MI300X and a transition from HBM3 to HBM3e. Also in the mix is a commitment to significant year-on-year performance uplift with a distinct annual cadence. After the MI325X this year, the MI350 series with an updated architecture will come in 2025, followed by a move to 3nm in 2026. You can read about these announcements in this following article. AMD also previewed it’s next generation EPYC processors, codenamed Turin, but only gave the core count of the top CPU.

After the keynote, a select group of the press were invited to attend a 30 minute open Q&A session with CEO Dr. Lisa Su. As always with these open press Q&A, the direction of the questions is unfocused as each person will have their areas of focus. Presented here are cleaned versions of the questions and answers for readability.

Q: How has AI helped you personally in your role?

A: AI is touching all of our lives. Personally, I am an avid user of GPT, Co-Pilot too. I'm very focused on AI being used within AMD. As much as we talk about customer AI, we also prioritize AI for making our company better. Things like building better chips, faster chips, we want to put AI though the development pipeline, as well as marketing, sales, HR, all of those. It is going to be AI everywhere.

Q: NVIDIA made it clear to investors it plans to shorten development cycles to a yearly cadence, Now AMD too. How and why are you doing this?

A: It's what we see in the market. AI is #1 priority for us as a company. We've harnessed the development capability of the whole company, increased investments. The annual cadence is there because the market requires newer products, with more capabilities. The product portfolio is there to address a wide set of workloads. Not all customers will use everything, but each year will be the next big thing, and be most competitive. It's about investment, ensuring that the hardware/software systems are part of it, and we're committed to it (AI) as our largest strategic opportunity

Q: What is the importance of being in Taiwan, Taiwan’s role in the global semi-con sector, and also how important is China, with regard to the chip makers and the military drills recently seen?

A: Taking a step back, the semiconductor ecosystem is global. When we think about product development, it's global. Taiwan is very important to the semiconductor ecosystem, and we do a lot of manufacturing here, with TSMC, with back-end and substrate partners too. Also we have partners building the ecosystem, ODM partners, as well as contract manufacturers. Bottom line, it's really important to be global, and Computex is a great place where people come together. It's a special week. The overall global ecosystem, we are very much a believer about global semiconductor supply chain, customer supply chain, that's how we operate.

Q: PC World - Strix Point (Ryzen AI 300) has a sharp jump in TOPs. TOPs cost money. How do you see TOPs scaling compared to CPU/GPU?

A: Nothing is for free! Especially in power and cost-constrained designs. What we're seeing is that AI will be everywhere. The current CoPilot+ PCs and Strix at 50+ TOPs will start at the top of the stack. But it (AI) will go throughout our product stack. At the high-end, we will scale the TOPs, because we’re big believers in the more local TOPs  you have, the more capable the AIPC is going to be, it adds value when put on chip, and It helps offload some of the compute from the cloud.

Q: Last week you said AMD will make 3nm with GAA chips. Samsung Foundry is the only foundry doing 3nm GAA - so will AMD choose Samsung Foundry for this?

A: Referring to keynote at imec last week. What we were talking about is that AMD will always use the most advanced technology. We will use 3nm. We will use 2nm. We didn't say the vendor for 3nm or GAA. Our current partnership with TSMC is very strong - we talked about 3nm products we're doing now.

Q: CRN Poland – A question about sustainability. AI means more consumed power. As a chip vendor, is it possible optimize power consumption by devices which use AI?

A: Energy efficiency is as important as performance for all we do, especially for AI. We look at how we can improve energy efficiency with each gen going forward - we've said that we would improve the energy efficiency by 30x from 2020 to 2025, we're on track to exceed that. We’re currently looking to get to 100x over next 4-5 years for energy efficiency. So yes, it’s possible to be focused on energy efficiency and we must be focused on energy efficiency because it will be a limiter for future compute.

Q: Adrian Branco - We’ve had CPU, then GPU, now NPU. One, how do you see the scalability of NPU? And two, what is going to be the next big chip, something Neuromorphic?

A: You need to have the right engine for each workload. A CPU is very good for traditional workloads. The GPU has been great for gaming and graphics tasks. The NPU is helping for AI specific acceleration. As we go forward, and we look at specific new acceleration technologies, we will see some of those going forward - but in the end it is application driven.

Q: Paul Alcorn - You initially upset the status quo from Intel by moving up in core count compared to them. But you have topped out at 16 cores (on the consumer side) for a few generations now. Is that sufficient for the consumer and gaming markets? Or should we expect increases in the future?

A: I think our strategy is that we have to increase performance continually. For gaming in particular, the gaming software developers have not necessarily used all the cores from time to time. There is no physical reason we couldn’t go past 16 cores. The key is that we're going at a pace that the software guys can and do utilise it.

Q: With regards to the desktop, do you see the need for more of an efficiency geared NPU accelerator?

A: We see the NPU having an impact in desktops. We are constantly evaluating the product segments that can utilise this. You'll see future desktop products with NPU to expand portfolio.

Q: You surprised everyone with time to market - last year in July with the first AIPC. Coming back to TOPS, you proudly showed you’re better than the current competition. So what’re your thoughts on beating the competition?

A: Our goal is to be very competitive always. For AI PC this is our 3rd generation. The XDNA architecture is something we've innovated. As we've gone forward, we will continue to innovate with it. We think 3rd Gen Ryzen AI will be a very powerful accelerator. We will continue to drive more TOPs, and drive to get the NPU into more mainstream price points.

Q: George Cozma, Chips and Cheese – You’ve spoken about getting ROCm on more consumer focused GPUs, what about APUs for wider device accessibility?

A: ROCm on AMD is important. The focus with ROCm has been on Data Center GPU side, and that's where a lot of effort is, that's where performance is. We want people to be able to use ROCm on all devices. We've started to expand with ROCm on Radeon, and over time we'll see ROCm on APUs as well.

Q: Gartner - How long will the X670/X870 chipset be used in future devices?

A: Jack Huynh (SVP of Computing and Graphics) - We work closely with ASMedia and as long as the end-user demands that chipset, we have no issues sustaining it.

Q: Are developers/customers willing to pay a premium to pay for AMD AI infrastructure vs competitors (for cloud)?

A: The way to think about AMD’s value proposition in Data Center is that our goal on perf/$ or TCO (Total Cost of Ownership) that we're offering a better value than the competition. We heard from Satya Nadella that Microsoft has done a lot of work with us on MI300 and all of our AI Data Center infrastructure. On LLMs, there's optimization to be done. But today, the leading price/perf for GPT based workloads is with AMD. It is about ensuring the TCO is better with AMD.

Q: It is said that in 2045, the AI can transcend human being. Why can AI beat the champion of Go?

A: AI is a great technology. It will only get better with time. It's also not perfect - we need to continue to make it better, in terms of capabilities, training, models, data, also how we utilize it. It will be excellent on many things. There’s a lot of focus now on getting to AGI, general intelligence. We're not there yet. Our goal at AMD is to provide foundational hardware and software capabilities so researchers can reach those milestones.

Q: Terrence (DaPoets) - Viewers like upgrading when it makes sense. Users like slotting in cards - should we be able to slot in an AI card, and do you think it’s likely we will?

A: I do think PCIe cards will have a lot of capability for AI going forward. That will be one way people upgrade. We should think about AI being everywhere - our notebooks, our desktops, our enterprise environments. You need the right kind of AI in those form factors. We need to make sure we have the right engines, the right type of AI engine. Good performance for the form factor. An AI engine in desktop is coming as well. The important thing is that ISV ecosystem is there as well. Work needs to be done to optimize for all of these different workloads.

Q: Dave, PC Gamer - Where do you see AI in PC Gaming? Will it just be on the development side, or will gamers see it?

A: You’re going to see AI everywhere. AI for gaming, certainly for developers, but as we go forward, game developers will want to use more AI. Not everything needs to be rendered, some things can be done with partial AI. We’ll see more of that in future systems.

Q: We heard about the performance of the Ryzen AI 9 HX 370, but we didn't hear about power consumption. Is there something you can share?

A: The overall Ryzen 9 HX 370 is going to be excellent in power consumption. Expect all day battery life, these are the expectations that people have grown to expect from notebooks

Q: Can you share more details AI research centre in Taiwan that AMD is investing in? It seems the Tainan and Kaohsiung government are working on this and hoping you might choose them as the location?

A: Overall, Taiwan is very important for R&D - we have 1000+ people here to do significant amounts of our R&D for all of our product portfolio, including PC and Data Center and supply chain. We look at how we partner in the region, so we’re continuing to look what we do here in Taiwan. No new info on our R&D Center.

Q: AMD is the only company to provide all different unit to satisfy all the markets. AMD also loves open source and open standards. What is your opinion on democratized AI? And the importance of Open AI (as in open source, not the company OpenAI)

A: I completely agree with your comments – a few key points about our philosophy – first, you need the best engines. whether It's CPUs, GPUs, NPUs, FPGAs, ASICS. We can do all of these engines at AMD. We work closely with our partners to think about their use cases. We feel strongly about open ecosystem, our view is that no one company can do everything. So many different use cases – whether you’re talking about AI, PC, edge, cloud, it all requires something different. An open ecosystem allows everyone to contribute, so that you can get the best solution overall.

Q: You said that AI PC will play an important role this year. What does it mean for the PC market total? Or market share for AMD for end-user and Data Center?

A: Our belief is that AI will be in all our products. It's critical for Data Center, edge, client. That's what is unique about our business, all aspects of the business require AI. What does it mean for the market - we've said that AI Data Center will grow to be a $400B industry by 2027, it's a huge market opportunity. Demand for GPUs will continue to grow with models. It's a large focus. For AI PC, we're excited, the form factor hasn't seen much change over 5+ years, and this is the biggest change in the PC form factor in the last decade. The AI PC will bring new capabilities to the PC, which will allow the PC market to grow in the next few years. We’re also excited for the opportunity for PC upgrade cycle to shorten with AI PC capabilities. It's just the beginning. At Computex this year you’ll see a lot about AI PC, you’re going to see a lot more as we go into the second half of 2024 and into 2025

Q: ROCm is going to build an open community, but there are a lot of ISAs outside of x86, such as ARM or RISC-V. I’d like to ask if AMD are working on getting ROCm to run well on platforms other than x86. Is AMD going to improve GPU perf on these platforms?

A: Overall, we're very open to using our GPUs in any platforms. Right now, most of the work is being done on x86, like EPYC + Instinct. If customers would like to pair our Instinct accelerators with ARM or other architectures in the future, we’re certainly open to that.

Q: Ian Cutress, More Than Moore: Currently in AI we are where we were in the early 2000s in GPUs: lots of players, lots of software vendors and software styles, but there is no unified system. It took a while before DirectX came along, and then Vulkan. With the fact AI is being split over CPU,GPU,NPU, FPGA, ASIC, do you think we’ll get a “DirectX” Moment for AI?

A: I don’t think there is any desire to push our own agenda. I’d say, from our perspective, that an open ecosystem is the best ecosystem. What you’re seeing is that due to the algorithms changing so fast, the hardware is also changing so fast, and that you can’t develop for a moving goalpost. Up until very recently, CPUs were doing the majority of the AI, now we’re seeing GPUs and other accelerators come in. I do think in time things will settle out. I won’t say there are too many things, but that there are many different workloads, and we’re at the place where we want to place AI across the product portfolio. Every part of the market will require some type of AI, and we’re trying to innovate as quickly as possible.

Q: What is the most important advice you have for women in tech?

A: It's an exciting industry. We need the most talented men and women in this industry. What I tell people, men or women, young or old, engineer or not – is to run towards problems. There are so many interesting problems to solve. As an engineer, I love it. As we solve those problems, it's great for your career, and very good for the industry.

More Than Moore, as with other research and analyst firms, provides or has provided paid research, analysis, advising, or consulting to many high-tech companies in the industry, which may include advertising on the More Than Moore newsletter or TechTechPotato YouTube channel and related social media. The companies that fall under this banner include AMD, Applied Materials, Armari, Baidu, Facebook, IBM, Infineon, Intel, Lattice Semi, Linode, MediaTek, NordPass. NVIDIA, ProteanTecs, Qualcomm, SiFive, Supermicro, Tenstorrent, TSMC.