AMD RDNA 5 Architecture Set to Double Core Count with 128 Cores per Compute Unit

AMD’s next-generation RDNA 5 gaming GPU architecture is reportedly set for a major core count overhaul. New details suggest a design featuring 128 cores per compute unit. This represents a 100% increase over the current RDNA 4 architecture. The flagship GPU could boast over 12,000 cores.

According to information from industry forums, this architectural shift signals AMD’s renewed focus on high-performance gaming graphics. The move aims to enhance competitiveness in the GPU market. These developments are based on leaks from reliable hardware sources.

RDNA 5 GPU Configurations Promise Significant Performance Leap

AMD is planning at least four different GPU die configurations based on RDNA 5. These will span the Radeon and Radeon PRO product families. The top-tier die will reportedly feature 96 compute units.

This configuration would deliver a staggering 12,288 cores. A mid-tier die with 40 compute units would offer 5,120 cores. Lower-tier options include 24-CU and 12-CU designs with 3,072 and 1,536 cores respectively.

This represents a significant departure from the current RDNA 4 lineup. The existing architecture uses only two dies with a maximum of 64 compute units. According to Wccftech, this expansion suggests AMD is returning to a broader product portfolio.

AMD’s Strategic Return to High-Performance GPU Market

The RDNA 5 architecture appears to mirror AMD’s successful RDNA 2 strategy. That generation featured multiple dies scaling from high to entry-level configurations. It was also AMD’s last truly competitive generation against NVIDIA’s best offerings.

The reported specifications suggest AMD is making a serious play for the high-end market. The company has focused on more mainstream segments with recent releases. This change could signal a renewed commitment to enthusiast-grade performance.

Performance improvements will likely extend beyond raw core counts. AMD is expected to enhance ray tracing capabilities and AI acceleration. These areas have become increasingly important for modern gaming workloads.

Industry observers note that core count alone doesn’t determine final performance. Architectural efficiency, clock speeds, and memory subsystems all play crucial roles. AMD will need to balance all these elements to compete effectively.

The RDNA 5 architecture represents AMD’s most ambitious GPU project in years. With potentially over 12,000 cores, the flagship configuration could dramatically shift the competitive landscape. This development marks a potential turning point for high-performance gaming graphics.

Info at your fingertips

What is the core count increase in RDNA 5 versus RDNA 4?

RDNA 5 is rumored to feature 128 cores per compute unit. This doubles the 64 cores per compute unit in RDNA 4. The flagship die could thus offer over 12,000 cores.

When will AMD release RDNA 5 GPUs?

No official release date has been announced. Industry speculation points to a potential 2026 launch timeframe. AMD typically reveals new architectures at major tech events.

How will RDNA 5 compete with NVIDIA’s offerings?

The increased core count suggests AMD is targeting high-end performance segments. Enhanced ray tracing and AI capabilities are also expected. This could make them more competitive with NVIDIA’s RTX series.

Will RDNA 5 use chiplet design like RDNA 3?

Leaks suggest a return to monolithic die designs for better economics. RDNA 3 utilized chiplet designs which faced some manufacturing challenges. The approach may vary across different product tiers.

What memory configuration will RDNA 5 use?

Flagship models may feature a 384-bit or 512-bit memory bus. This would support 24-32GB of VRAM for high-resolution gaming. Lower-tier models will likely use narrower memory interfaces.

Will RDNA 5 focus on gaming or AI workloads?

The architecture will primarily target gaming performance. However, improved AI acceleration is expected for features like upscaling. Professional variants will likely offer enhanced compute capabilities.