Advanced Micro Devices (AMD) stock has risen significantly since 2017, when the company first introduced its Zen family. Zen is an x86-based high-performance core architecture that uses a multichip design to boost performance. Zen delivered a 52% IPC (instruction per cycle) improvement over its previous generation, the Excavator core. It was Zen that put AMD back into competition with Intel (INTC) in the CPU (central processing unit) space.
What AMD’s Zen is about
AMD leveraged Zen architecture to build Ryzen desktops and laptops CPUs and EPYC server CPUs. It’s extending the Zen family by building Zen on next-generation process nodes.
At Hot Chips 2019 in August, AMD CEO Lisa Su explained that a CPU’s performance can be improved by using various elements. These elements are process technology, die size, TDP (thermal design power), power management, microarchitecture, and compilers. The company is using these elements to improve CPUs in each generation.
AMD’s Zen CPU roadmap from 2017 to 2022
AMD presented its updated Zen roadmap on September 4. The roadmap comprises Zen microarchitecture’s four generations. AMD built Zen and Zen+ architecture on 14nm (nanometer) and 12nm nodes in 2017 and 2018. It built two generations of Ryzen CPUs and one generation of EPYC CPUs on Zen and Zen+ architecture. These CPUs competed with Intel’s in performance and pricing.
In July, AMD launched its Zen 2 architecture built on TSMC’s (TSM) 7nm node. So far, it has launched Ryzen 3000 desktop and EPYC Rome server CPUs, and plans to launch more Ryzen CPUs later this year or early next year. Zen 2 put AMD ahead of Intel in the process technology race. Process technology contributes 40% of a CPU’s performance, because a smaller process node increases transistor density and improves performance. A denser node consumes less power, further adding to the CPU’s performance. Zen 2 delivered a 15% IPC boost over its predecessor, Zen.
AMD’s first 7nm products, Ryzen 3000 desktop CPUs, beat Intel’s existing 14nm++ CPUs on performance and price. As Intel doesn’t have a competing product, it’s using marketing and pricing strategies to compete.
The next chapter in AMD’s Zen family story is Zen 3 architecture. The company stated that Zen 3’s design is complete. Therefore, AMD could soon start sampling products based on Zen 3 architecture and launch them around mid-2020. AMD didn’t reveal details about Zen 3, other than that it will be built on the 7nm+ node. This node is only available from TSMC.
Zen 3’s architecture may be similar to Zen 2’s and Zen’s, but have upgraded process technology, compilers, TDP, and power management. These features mean we might see modest improvements in performance and efficiency in Zen 3 products. TSMC is using EUV (extreme ultraviolet) lithography technology in up to four layers of its 7nm+ node, and Intel is using EUV lithography in its 10nm node. Therefore, AMD’s 7nm+ products could compete with Intel’s 10nm products.
AMD stated that Zen 3 would have 20% more transistors and deliver 10% more efficiency than Zen 2. The smaller transistors are, the less power they consume and the more transistors can fit inside a single die, improving its performance and power efficiency.
After completing Zen 3’s design, AMD’s design team started work on Zen 4. No details about this technology are available, other than that the company plans to release it in mid-2021 or early 2022. We do have details on TSMC’s roadmap, however. In up to 14 layers of its 5nm node, the foundry plans to use expensive EUV lithography, which reduces the cost benefits of a smaller node.
TSMC expects the 5nm node to be 1.8 times denser and 15% faster than the 7nm node. As future Zen products’ performance improvements will be modest because of process technology limitiations, AMD is using a chiplet strategy. The design connects various chips on an input and output chip and improves the core count and performance while keeping costs low.
Intel’s CPU roadmap plays catch up
Intel’s 10nm node was delayed by three years due to technical difficulties and yield issues arising from its complex design. During these three years, Intel launched two 14nm upgrades, the 14 nm+ and 14 nm++ nodes. While AMD is moving quickly on process nodes, Intel may be catching up. At the company’s Investor Day 2019, Intel chief engineering officer Murthy Renduchintala revealed the company’s road map. It plans to launch the 7nm node in 2021, the 7 nm+ in 2022, and the 7 nm++ in 2023.
Intel is changing its technology strategy. Instead of adopting a single die approach, it’s focusing on advanced packaging technologies that integrate individual chips into a heterogeneous device. It also plans to focus its technology efforts on architecture, memory, interconnectivity, security, and software.
It remains to be seen whether AMD can maintain its technological advantage over Intel. AMD investors are betting on its technological advantage and potential to gain CPU market share from Intel. If AMD’s technology falls behind and it loses market share, the stock could fall to 2016 levels, below $10. Once AMD strengthens its position in the PC and server CPU markets, investors may bet on the company’s efforts in AI and autonomous driving technology.