AMD’s upcoming RDNA 5 / UDNA GPUs are set to deliver significant performance improvements, with the flagship model potentially featuring up to 96 compute units (CUs), a 50% increase over the top RDNA 4 GPU’s 64 CUs. According to recent leaks from Kepler_L2, we might see a variety of configurations, ranging from an ultra-powerful 96 CU model to a more budget-friendly 32 CU variant.
The expected memory bus for these GPUs includes a massive 384-bit interface for the flagship, ensuring faster data throughput, alongside 256-bit and 128-bit memory buses for mid- and entry-level models, respectively. This variety of options hints at improved performance across the board for both gamers and professionals.
The top-tier RDNA 5 GPU is predicted to feature a 384-bit memory bus, enhancing VRAM capacities and enabling denser VRAM dies for higher memory volumes, possibly up to 32 GB. This could put AMD in direct competition with NVIDIA’s future products, potentially even surpassing the performance of NVIDIA’s 80-series GPUs. While this flagship won’t target the ultra-enthusiast market, it still positions itself as a strong contender with its value-focused approach.
For mid-range users, the 64 CU version with a 256-bit bus promises a solid uplift from the RDNA 4, with better ray-tracing and AI capabilities. The entry-level 32 CU configuration might raise some concerns due to the 128-bit memory bus, but by 2026, when these GPUs are expected to launch, denser VRAM modules should mitigate this limitation. These GPUs, particularly in the mid- and entry-level categories, should offer decent uplifts in performance, with both AMD and NVIDIA expected to tackle the 8 GB VRAM barrier.
Although these specifications remain speculative at this stage, with mass production slated for Q2 2026, AMD’s RDNA 5 GPUs, along with NVIDIA’s “SUPER” series and Intel’s upcoming Big Battlemage lineup, could make the start of 2026 an exciting time for PC gamers.