AMD details how it built a product line-up with just two RDNA 4 dies — Flexible design and asymmetric harvesting enables production of multiple models without new silicon

Graphics processing units (GPUs) are designed in a way that their performance can be scaled up or down during the design phase, or even after tape-out. AMD’s 9000-series, built on RDNA 4 (also known as Navi 4), is no exception. AMD confirmed this during one of its Hot Chips 2025 presentations, and demonstrated how it can cut down the design of its GPUs to produce more SKUs.

Building a product family using two GPU designs

All GPUs contain a massive number of similar components, including Arithmetic Logic Units (ALUs), caches, special-purpose hardware units, controllers, and physical layers (PHYs). So, if a defect occurs during production, the faulty part could be disabled, and the processor may still function correctly. However, it does not mean total flexibility. Some units reuse transistors between them, and others rely on data paths that are integral to neighbouring components. AMD claims it has designed RDNA 4 to be more adaptable than its predecessors, thanks to a feature it calls asymmetric harvesting.

By selectively disabling components, adjustments can be made to memory systems and asymmetric resource allocation. This allows the company to produce higher-end, mid-tier, and specialized GPUs from the same base design, or even the same die.

You may like

• AMD’s desktop GPU market share hits all-time low despite RX 9070 launch, Nvidia extends its lead [Updated]

• RX 9060 XT 16GB graphics cards with faster clocks than 8GB models appear to be in development

• AMD quietly announces Radeon RX 9060 alongside new Adrenalin driver

By using this strategy, AMD spawned a smaller Navi 44 (Radeon RX 9060-series) out of the bigger Navi 48 (Radeon RX 9070-series) design by reducing the number of shader engines (SEs), Infinity cache, GDDR6 controllers, and PHYs, but leaving things like the command processor, display engines, media engines, security processor, and other specific things intact. By reusing photomasks from Navi 48, AMD saved on manufacturing costs. In addition to this, AMD built the Radeon RX 9070 and RX 9070 GRE from the full-fat Radeon RX 9070 XT by disabling certain elements, which essentially increased yields and enabled them to hit pricing targets. Such an approach also shortened the relevant GPU’s time-to-market, as fewer unique silicon designs required tape out, validation, and production.