AMD RDNA 5: High-End GPU to Challenge Nvidia’s Best
▼ Summary
– AMD’s upcoming RDNA 5 GPUs may feature a new CU naming convention where each CU equals two previous-generation CUs, making the top model effectively equivalent to 192 old-style CUs despite a reported 96 new CUs.
– The lineup includes four GPUs (AT0, AT2, AT3, AT4) with respective CU counts of 96, 40, 24, and 12 under the new measure, corresponding to 192, 80, 48, and 24 under the old measure.
– Lower-end RDNA 5 models (AT3 and AT4) are rumored to use LPDDR5X laptop memory with wide memory buses (256-bit or 384-bit) to maintain bandwidth, potentially stacking memory controllers to save space.
– RDNA 5 is expected to deliver significant performance improvements in ray-tracing and path-tracing beyond just increased CU counts, targeting a full stack of competitive gaming GPUs.
– These details remain speculative but form a coherent and plausible picture of AMD’s high-end ambitions for next-generation graphics technology.
Speculation surrounding AMD’s next-generation graphics architecture continues to intensify, with fresh leaks suggesting a bold strategy to compete directly with Nvidia’s flagship offerings. Whether referred to as RDNA 5 or the newer UDNA naming convention, these upcoming GPUs appear poised to deliver significant architectural improvements and performance uplifts across the entire product stack.
Recent information from trusted industry sources indicates a revised compute unit configuration for the top-tier model, now rumored to feature 96 CUs rather than the previously leaked 192. This apparent contradiction actually stems from a fundamental change in how AMD defines compute units in its new architecture. Where previous RDNA generations used Work Group Processors (WGPs) containing two CUs, it appears RDNA 5 will simply rename WGPs as CUs. This means 96 RDNA 5 CUs would effectively equal 192 units under the old measurement system.
The proposed lineup consists of four primary dies, codenamed AT0, AT2, AT3, and AT4, with respective compute unit counts of 96, 40, 24, and 12 under the new naming convention. When compared to current generation hardware, these numbers reveal substantial potential gains. The flagship AT0 die could offer performance equivalent to 192 traditional CUs, representing a massive leap over the 64 CUs found in AMD’s current RDNA 4 flagship, the Radeon RX 9070 XT.
Beyond raw compute power, RDNA 5 is expected to deliver substantial improvements in ray-tracing and path-tracing capabilities that transcend simple CU comparisons. Architectural enhancements and new feature sets should provide noticeable gains in visual fidelity and performance across supported titles.
Another intriguing development concerns memory configuration for the lower-tier models. Leaked block diagrams suggest both AT3 and AT4 may utilize LPDDR5X memory typically found in laptops rather than traditional GDDR6 or GDDR7 graphics memory. To maintain adequate bandwidth, these chips might employ unusually wide memory buses, potentially 256-bit or even 384-bit configurations. This approach mirrors AMD’s strategy with its Strix Halo laptop APU, which stacks memory controllers to accommodate similar bus widths while using space-efficient LPDDR5X modules.
While these details remain unconfirmed, the emerging picture suggests AMD is preparing a comprehensive assault on the high-end GPU market with RDNA 5. After focusing on the mainstream segment with RDNA 4, the company appears ready to challenge Nvidia’s best with a full stack of competitive gaming graphics cards featuring innovative memory solutions and significantly improved ray-tracing performance.
(Source: pcgamer)
