At this year’s CES, AMD’s CEO Dr. Lisa Su gave the pre-show keynote, and she came with an overabundance of announcements and guests. The theme of the keynote was high-performance and adaptive computing to help solve problems. On-stage guests included representatives from HP, Intuitive Surgical, Lenovo, Magic Leap, and Microsoft. It also included a former NASA astronaut for a bonus.
Dr. Su made seven key product announcements spanning energy efficient PCs to extreme processors designed for supercomputers. With all the announcements and guests, it’s no wonder the keynote ran significantly longer than planned. It seemed that every group in AMD got a piece of the keynote.
A PC Update
Despite the current slump in PC sales, most of AMD’s announcements focused on PCs – both desktop and notebooks. Of particular interest was AMD’s decision to add a dedicated AI core to its new Ryzen 7040 series notebook processors. The AMD Ryzen 7040 is designed to fit into ultrathin notebooks with 15W to 45W TDPs (thermal design power) and is made using TSMC’s advanced 4nm process. The Ryzen 7040 has AMD’s most advanced Zen 4 CPU cores matched with the latest RDNA3 GPU architecture.
The Ryzen 7040’s embedded AI technology was developed by Xilinx, which AMD acquired last year. But even prior to the acquisition, AMD had engaged with Xilinx to license the technology. Xilinx developed the AI engine for its Versal ACAP adaptive computing product. The details of the Ryzen AI engine, as AMD is now calling it, are still to be revealed, but Versal AI Engines are 2D arrays of VLIW vector-vector and matrix-matrix compute engines. While AI workloads can be run on AMD’s CPUs and integrated GPUs, AMD claims that running these workloads on a dedicated AI engine is more power efficient, which is why the company has introduced it’s first in a notebook processor. Microsoft’s EVP & Chief Product Officer Panos Panay gave a great endorsement for Ryzen AI and how Microsoft plans to use it initially for Teams and for other features in the future. We encourage AMD to make access to Ryzen AI through open tool chains and APIs in addition to Microsoft’s Studio tools.
It’s great to see AMD really embracing the pervasiveness of AI in computing. AMD has GPUs and FPGAs/ACAP for AI acceleration, and now is incorporating AI technology into mainstream notebook processors. Intel will also incorporate an integrated AI accelerator in its Meteor Lake processors later this year.
AMD also brought its RDNA 3 architecture to mobile discrete GPUs by announcing the Radeon 7000M series. These new mobile GPUs can power 1080p gaming at the highest game settings and support graphically intensive content creation applications. When you combine the AMD Radeon RX 7000M series graphics with Ryzen 7000 Series processors, AMD SmartShift RSR technology intelligently distributes image rendering, upscaling, and presentation demands between APU and GPU resources to optimize performance. This capability is expected to be available in the first half of 2023.
Desktops are also getting a processor upgrade. The new Ryzen 7000 series X3D desktop processors are equipped with up to 144MB of cache and sport up to 16 cores and 32 threads. The new processors are a bit weird in that only one of the two CPU cluster chiplets in the package get the stacked memory, giving the CPU an asymmetric bias. For example, the Ryzen 9 7950X3D has one 8 core CPU chiplets without the 3D cache that can clock to 5.7GHz with boost, but the other 8 core chiplet in the processor with the stacked 3D cache have more cache memory but a lower boost clock. It will be interesting to see how well the OS thread manager and application developers handle such an asymmetric configuration.
The other PC-related announcements include a high-power Ryzen 7045HX notebook processor design for larger laptops, aimed at gamers and creative professionals. This processor uses the chiplet design from the desktop part to deliver as many as 16 cores.
There was also a low-key announcement of AMD Ryzen 7000 Series desktop processors coming soon. Built on the Zen 4 architecture and featuring a 65W TDP, this new lineup of Ryzen 7000 Series processors is optimized for both efficiency and performance. There are some cost-optimized AM5 motherboards coming before the spring as well.
AI in the Data Center
There were three announcements on how AMD is working to advance AI for data center applications: AMD Vitis Medical Imaging Libraries, Alveo V70 AI accelerator, and the Instinct MI300 high-performance computing accelerator.
AMD announced its Vitis Medical Imaging libraries to bring premium medical imaging products to market faster by reducing development times. These software libraries run on AMD Versal SoC devices with AI Engines to deliver high-quality, low-latency imaging for medical applications.
AMD also displayed its XDNA adaptive AI architecture in a discrete PCIe add-in card with the Alveo V70. Based on AMD XDNA with AI Engine architecture, the Alveo V70 extends pervasive AI from edge to cloud.
Xilinx FPGAs presently are used in the Perseverance Mars rover, so AMD hosted an interview with former NASA astronaut Dr. Cady Coleman. Dr. Coleman was promoting NASA’s agenda on human space travel to the Moon and eventually Mars with the Artemis program. But the broader message was about promoting STEM education and inspiring more people, including women, to follow a career in science.
Dr. Su ended her keynote with a preview of the company’s next monster GPU for high-performance computing – the MI300- which is composed of 13 chiplets and uses both 2.5D and 3D packaging techniques. It has a total of 146 billion transistors. The MI300 offers not just GPU cores, but also includes three Zen 4 CPU chiplets in the package. The goal of packaging the CPU cores in the GPU cluster is to be able to run tasks closer to the GPU and eliminate transaction latency with the main host CPU. The CPU/GPU cluster will be supported by 128GB of HBM (High Bandwidth Memory) local memory in the package. This design is similar to Nvidia’s Grace-Hopper HPC processor, which combines Grace’s ARM CPUs in the same package with the Hopper GPUs. The MI300 should ship in 2H23.
The numerous introductions during Dr. Su’s keynote showed that AMD is continuing to provide leadership products in PCs and high-performance computing. Despite the present technology market’s downturn, AMD continues to invest in new, cutting-edge products and is accelerating its AI roadmap.