AMD's future technology direction has unfolded with greater clarity recently. The most recent updates indicate that AMD will bypass the RDNA 4 architecture for its upcoming APU generations in favor of the existing RDNA 3.5 architecture. This strategic choice is substantiated by the code found in AMD's GPUOpen driver, where the RDNA 4 device ID, "GfxIp12," is specifically designated for discrete GPUs alone.

This decision suggests that RDNA 4 will be exclusively available with AMD's newest Radeon RX 9000 series graphics cards, whereas the APU range will continue to integrate the RDNA 3.5 architecture.

Medusa Point—the next generation of mobile APUs based on Zen 6 architecture—will adhere to this strategy. As the successor to the Strix Point and Kraken Point series, Medusa Point is set to incorporate more potent Zen 6 CPU cores, yet will maintain RDNA 3.5 instead of transitioning to RDNA 4 for its integrated graphics (iGPU). Although this may appear unexpected, it aligns with AMD's product approach in recent times. RDNA 3.5 has already proved its capability within the Zen 5 mobile lineup, especially for high-performance APUs like the Strix Halo, where its iGPU matches the performance of entry-level discrete graphics cards—its flagship even competing with the 4070M model.

The reasoning behind AMD's preference for RDNA 3.5 in its APUs centers on technological optimization goals; RDNA 3.5 is not merely an extension of RDNA 3, but a refined architecture that prioritizes power efficiency and integration density—ideal for the low-power demands of mobile devices. Conversely, RDNA 4 is engineered to enhance discrete GPU performance, offering elevated ray-tracing capabilities and improved AI compute support, features that may not be essential for typical APU applications like lightweight laptops and handheld gaming devices. The Strix Halo flagship, equipped with up to 40 Compute Units (CUs), excels in 1080p and even 2K gaming, reaching levels comparable to mid-range discrete GPUs. This capacity may lead AMD to conclude that introducing RDNA 4 into APUs is unnecessary at this time.

Simultaneously, RDNA 4-based discrete GPUs have demonstrated remarkable performance. Since its debut in early March, the Radeon RX 9070 series has seen sales surpassing 200,000 units, marking it as one of AMD's most successful GPUs in recent years. The RX 9070 XT model boasts 64 RDNA 4 compute units, supported by 16GB of GDDR6 memory, and achieves a peak boost frequency of up to 2.97 GHz. It delivers an over 40% average performance boost over its predecessor, the RX 7900 GRE, at 1440p resolution. Furthermore, RDNA 4 introduces the third-generation Light Trace Accelerator, supports advanced rendering techniques like path tracing, and integrates a new hardware unit optimized for AI and machine learning, enhancing its competitiveness in both gaming and creative fields.

However, Medusa Point users opting for RDNA 3.5 may face certain limitations. The most noticeable being the lack of support for AMD's latest FidelityFX Super Resolution 4 (FSR 4) technology. FSR 4 is exclusive to RDNA 4, leveraging machine learning algorithms to enhance image quality and frame rates, particularly in high-resolution gaming. Since Medusa Point won't feature RDNA 4, FSR 4 support may have to wait for a future architectural revision. Additionally, though RDNA 3.5 suffices in performance, it falls short compared to RDNA 4 in areas such as ray-tracing and AI processing, potentially disappointing gamers seeking cutting-edge tech.

Looking forward, AMD is tentatively planning the UDNA architecture, a fresh GPU blueprint that merges RDNA and CDNA architectures to compete against NVIDIA's CUDA ecosystem—potentially serving as a unifying foundation for both APUs and discrete GPUs. Market speculation hints at Medusa Halo—a successor to Strix Halo—as the debut platform for UDNA, although this shift is unlikely before 2026, ensuring Zen 6 APUs will continue revolving around RDNA 3.5 for the time being.

Details regarding Zen 6 and Medusa Point remain sparse; however, Zen 6 is projected to utilize TSMC's advanced processes—possibly 3nm or variations thereof—to achieve better power efficiency and core density. As a mobile APU, Medusa Point is expected to maintain the high-performance stature of the Strix line, targeting premium gaming and next-gen handheld devices. Rumors suggest its CPU section may ship with up to 12 cores and 24 threads, paired with 16 to 32 RDNA 3.5 compute units, along with a consistent 256-bit memory bus width akin to Strix Point. Such configurations should satisfy 1080p to 2k gaming requirements while ensuring superior thermal and durability performance in lightweight devices.

AMD's strategy exemplifies its differentiation approach between APUs and discrete GPUs—where RDNA 4 emphasizes heightened performance for desktops and premium notebooks, RDNA 3.5 continues innovating within integrated graphics. This allows APUs to balance power, cost, and performance efficiently. As Medusa Point waits in development until 2026, tech enthusiasts should observe closely how AMD's strategic design reshapes the market landscape. Nonetheless, the combination of Zen 6 and RDNA 3.5 signifies another substantial leap for AMD within mobile computing.