January 17, 2024
The patent application reportedly relates to AMD’s use of chiplets instead of monolithic dies in its GPU designs, and details how the rendering workload will be distributed across a series of chips rather than relying on one large die for processing.
While the patent application arguably complicates the rendering process, there are reportedly potential benefits to this development. The main reason for switching to chiplets is to reduce manufacturing costs for high-end hardware. According to Yahoo Finance, large GPUs are less cost-effective because “each silicon wafer yields fewer working dies”. In contrast, the contemplated chiplets are reportedly so small that a wafer can “churn out over a thousand of them”.
This news is unsurprising considering the increasing prominence of chiplets in processor design. While GPUs have been slow to migrate to chiplets, the technology has gained traction in CPU design. For example, AMD’s Zen 2 CPUs and Intel’s Meteor Lake CPU series both boast processing via chiplets.
The patent application therefore appears to be a natural next step in the evolution of GPU processing.
Authors: Olivia Moon and Arash Rouhi