An interview with AMD's SVP Sam Naffziger on challenges facing the company, its coming RDNA 3 chiplet design, GPU performance and power efficiency, and more
Context & Ripple Effects
AMD has turned executive interviews into a running narrative of its efficiency roadmap: CTO Mark Papermaster used the same format to frame Zen 3's core-complex redesign and power-efficiency gains in 2020, and earlier to discuss foundry partnerships with TSMC and Samsung. Now SVP Sam Naffziger applies the template to graphics.
The substance behind this interview arrived two weeks earlier, when AMD confirmed that RDNA 3 will use a chiplet design targeting a 50%+ performance-per-watt uplift over RDNA 2, with first chips still slated for 2022. Naffziger's appearance signals AMD is ready to talk publicly about how that target gets met.
First-order effects
- AMD is committing its GPU line to the chiplet approach it proved on CPUs, putting the 50%+ performance-per-watt target against RDNA 2 as the public benchmark its next-generation cards will be judged on.
Second-order effects
- A chipleted GPU multiplies AMD's dependence on advanced packaging and leading-edge foundry supply — the TSMC relationship Papermaster discussed in 2020 now becomes a graphics-critical dependency, not just a CPU one.
Third-order effects
- If RDNA 3 lands near its stated efficiency target, chiplets stop being an AMD CPU specialty and become the default architecture across both major product lines, shifting GPU competition toward per-watt metrics and packaging capability rather than monolithic die size.
The trend: AMD is extending its chiplet playbook from Zen CPUs to RDNA GPUs, making performance-per-watt the axis on which it competes with larger monolithic designs.