Based on current information, AMD's upcoming Radeon graphics cards are not expected to be released soon. Insights from various sources suggest that the launch of the RDNA 5 architecture GPUs will occur around the middle of 2027. Consequently, the year 2026 is anticipated to be a transition period for Radeon’s consumer graphics cards, concentrating on refining the existing product lineup rather than frequent updates.

This pace aligns with the progression of chip development. According to source @Kepler_L2, the RDNA 5 has started shipping from TSMC, utilizing TSMC's N3P node, contrary to the earlier rumors of using Samsung's process. At this stage, AMD continues to invest in TSMC's advanced processes for its high-end GPUs. AMD aims to enter full-scale production in the latter half of 2026, preparing for a broad market release in 2027.

The shift in process nodes is noteworthy; RDNA 4 primarily utilizes TSMC's N4P, an enhanced iteration of the N5. In contrast, the N3P is a performance-enhancing version of the N3. Public metrics indicate that the N3P delivers approximately 18% more performance and 36% less power consumption compared to the N5, with a markedly higher transistor density. For GPUs, these node upgrades provide benefits beyond increased frequency, such as enabling more complex front-ends, cache structures, and specialized acceleration units, which are often more vital than simply increasing power.

Architecturally, RDNA 5 is not solely dependent on process improvements; AMD has outlined several strategies for the underlying technology of next-generation GPUs, including more robust general-purpose compression to ease memory bandwidth challenges and a novel way of organizing compute units known as "neural arrays." These are intended to allow more flexible sharing of compute resources between AI and graphics workloads. Additionally, a new generation of ray-tracing hardware, dubbed Radiance Cores, is being introduced. Rather than focusing on a singular performance metric, these innovations indicate a strategic allocation of hardware resources toward bandwidth, real-time ray tracing, and AI-assisted graphics processing.

These advancements are not limited to discrete graphics cards. As revealed, RDNA 5's associated IP will also be integrated into next-generation console SoCs, including Sony’s PlayStation (codenamed Orion) and Microsoft’s Xbox (codenamed Magnus). Such cross-platform utilization will help offset R&D costs while simultaneously constraining the architectural design to be more adaptable concerning power usage and space.

Currently, details are primarily speculative. Claims have emerged suggesting that the next-generation RDNA architecture GPUs may feature over 12,000 compute cores, with 128 cores per compute unit, although this information remains unconfirmed. It is verified, however, that GFX13-related IP has appeared in the early Linux kernel code, indicating that the architecture has at least reached a defined stage. Previously circulated production schedules suggested a Q2 2026 timeframe, but it's now more plausible that this will be adjusted towards late 2026 or beyond, aligning with a mid-2027 release.

Presently, GPU manufacturers are directing resources towards the AI market, which, along with high usage of DRAM and NAND capacity, has propelled the overall cost of graphics memory, system memory, and SSDs. In this climate, accelerating consumer GPU updates would be challenging. For AMD, maintaining a stable product line and waiting for the maturation of RDNA 5’s process and architectural advantages appears to be a prudent approach before fully concluding the lifecycle of RDNA 4.