AMD’s forthcoming RDNA 5 / UDNA GPUs could deliver a substantial leap in performance, according to insights from leaker Kepler_L2. The flagship model is expected to boast up to 96 compute units (CUs), a significant enhancement over its predecessors.
Rumors suggest AMD is gearing up for three main configurations within this new lineup:
1. A high-end version with 96 CUs featuring a 384-bit memory bus.
2. A mid-tier option offering 64 CUs and a 256-bit memory bus.
3. An entry-level model with 32 CUs coupled with a 128-bit memory bus.
The top model’s increase to 96 CUs represents a 50% jump from the previous 64 CUs found in the RDNA 4 series. This could potentially result in impressive performance, especially in ray tracing and AI capabilities. NVIDIA rivals might see a run for their money as AMD aims to challenge even the next-gen 80-tier products from its competitors.
While the flagship model might not dethrone NVIDIA’s ultra-enthusiast GPUs, it promises an impressive value proposition, mirroring the configuration seen in the Navi 31 with its 384-bit interface. The mid-range 64 CU version follows suit with similar improvements, thanks to refined CUs and upgraded RT cores.
The 32 CU variant aims to cater to the entry-level market. Despite its smaller 128-bit memory bus, advances in architecture and memory technology could still provide noticeable performance boosts. The expectation is that by 2026-2027, denser VRAM modules will be common, improving capacity and tackling long-standing limitations in entry-level graphics cards.
While these configurations are based on speculative insights, they hint at an exciting new chapter for AMD’s gaming GPUs. The new architecture, geared for mass production by Q2 2026, coincides with NVIDIA’s “SUPER” series and Intel’s anticipated GPU releases. As we approach 2026, PC gamers should prepare for an exhilarating shift in the GPU landscape.
