AMD is steadily expanding its graphics roadmap for upcoming APUs, with new details emerging through open-source code updates. Recent LLVM compiler patches have specifically unveiled two new graphics targets, GFX1171 and GFX1172, joining the previously identified GFX1170. All three fall under the RDNA 4m designation, indicating an intermediate solution between the current RDNA 3.5 and the future RDNA 4 architectures. This discovery is notable as it clarifies AMD’s strategy of developing multiple variants of the same architecture for its integrated GPUs, primarily intended for SoCs and APUs rather than discrete graphics cards.
Technical Insights into AMD’s New Graphics Targets
The RDNA 4m designation is not directly linked to the standalone RDNA 4 series GPUs; instead, it represents an internal evolution within the GFX11 family. Technically, these new targets share similar capabilities, including support for FP8 and BF8 formats, which are increasingly crucial for artificial intelligence workloads. Furthermore, they incorporate WMMA (Wave Matrix Multiply Accumulate) instructions, essential for accelerating matrix operations common in AI and machine learning applications. This clearly indicates AMD’s commitment to optimizing its future APUs not just for gaming, but also for advanced computational scenarios.
AMD and the ‘Medusa Point’ Rumors
Rumors suggest these integrated GPUs will be part of the Zen 6-based APUs, codenamed ‘Medusa Point’. These solutions are expected to succeed the current Zen 5 ‘Strix Point’ chips, which already feature RDNA 3.5 GPUs. However, the more significant generational leap is anticipated with the ‘Medusa Halo’ variants, which are rumored to introduce the RDNA 5 architecture and LPDDR6 memory support. In this context, RDNA 4m serves as a strategic transitional phase, enabling AMD to implement incremental improvements and maintain competitiveness without waiting for a complete architectural overhaul.