AMD discusses the shift toward GPU chiplet architectures to improve performance and manufacturing efficiency.
AMD recently detailed its upcoming RDNA 3 GPU architecture, slated to launch before the end of the year with a chiplet-based design. In an interview, Sam Naffziger, AMD's Senior Vice President, Corporate Fellow, and Product Technology Architect, discussed the company's approach to addressing power and efficiency challenges in modern GPU design. Naffziger, who has been at AMD for 16 years, is responsible for multiple product areas with a focus on driving higher performance per watt and improving the overall competitiveness of AMD's CPUs and GPUs. He is also one of the primary architects behind AMD's chiplet architecture, which has proven successful in the Ryzen and EPYC CPU lines and will now come to some form in AMD's RDNA 3 graphics.
The industry faces mounting power consumption pressures across next-generation GPUs. PCIe 5.0's power interface and upcoming compatible power supplies can supply up to 600W over a single 16-pin connector, signaling a broader shift toward higher-power GPUs. Current rumors suggest Nvidia's Ada architecture will push power limits higher than previous generations, with some predictions indicating 450W TBP (typical board power) and potentially as much as 600W TBP for top RTX 40-series GPUs. While AMD has not yet announced TBP figures for RDNA 3, the company is likely to follow similar industry trends.
Naffziger confirmed that total power consumption will increase in next-generation GPUs but emphasized how efficiency improvements can maximize overall performance. "It's really the fundamentals of physics that are driving this," he explained. "The demand for gaming and compute performance is, if anything, just accelerating, and at the same time, the underlying process technology is slowing down pretty dramatically — and the improvement rate. So the power levels are just going to keep going up. Now, we've got a multi-year roadmap of very significant efficiency improvements to offset that curve, but the trend is there." AMD claims a 50% improvement in performance per watt for both RDNA and RDNA 2, and is targeting another 50% improvement in performance per watt with RDNA 3. This metric can represent 50% more performance at the same power, the same performance using 33% less power, or anywhere along the performance and power curve.
AMD has leveraged expertise from its CPU design teams to achieve efficiency gains with RDNA 3. For example, running at 2.5 GHz and 1.0V instead of 1.2V would require 40% more power. The company employs a strategy of "cross-pollinating" CPU and GPU design teams, bringing the best technology from both sides to each new design. The current GPU cores are "intrinsically more power efficient," Naffziger stated. "Performance is king," he said. "But even if our designs are more power-efficient, that doesn't mean you don't push power levels up if the competition is doing the same thing. It's just that they'll have to push them a lot higher than we will." This suggests that AMD, similar to Nvidia, may increase TBPs for its top RDNA 3 graphics cards.
AMD's historical performance supports confidence in its efficiency claims. AMD's Zen 3 CPUs generally lead Intel in efficiency and performance per watt, and RDNA 2 GPUs tend to beat rival Nvidia's GPUs in efficiency depending on specific models. RDNA, launched in 2019, delivered a clear 50% or more improvement in performance per watt compared to the previous generation Vega and Polaris architectures.