The GPU market is abuzz with excitement as fresh leaks give us glimpses of AMD’s RDNA 5 graphics cards. With the poor reception of RDNA 4, hopes are high for AMD’s latest generation, and the preliminary information hints that RDNA 5 may be a game-changer. With new Compute Units rearchitected, huge core counts, and potential changes in memory configuration, this generation could finally put AMD into direct contention with NVIDIA’s highest-end GPUs.
Table of Contents
1) Flagship Navi 5X: 96 Compute Units, 384-bit Bus
According to rumors, the top-end SKU of the RDNA 5 lineup, which is most likely named Navi 5X, will have:
- 96 Compute Units (CUs) or more
- Memory bandwidth is greatly increased by a 384-bit memory bus.
- Possible GDDR7 memory support, though final specifications are still pending
If accurate, this setup would put AMD in a direct competitive position against NVIDIA’s high-end GPUs, a market it lost during the RDNA 4 cycle.
2) A Radical Redesign of Compute Unit Structure
The most silent rumor regarding RDNA 5 is the radical redesign of the Compute Unit (CU) structure. AMD has maintained 64 cores per CU for years, but several leaks indicate that RDNA 5 will have 128 cores per CU.
This redesign of the architecture allows for AMD to fit many more cores into fewer CUs, increasing efficiency and brute throughput. If true, the top-of-the-line RDNA 5 GPU would contain 96 CUs with more than 12,000+ cores—a mind-boggling figure compared to Radeon GPUs available at the moment.
| GPU Tier | Compute Units | Cores per CU | Total Cores |
|---|---|---|---|
| Flagship | 96 | 128 | 12,288 |
| High-End | 40 | 128 | 5,120 |
| Mid-range | 24 | 128 | 3,072 |
| Entry | 12 | 128 | 1,536 |
The Navi 48 and Navi 44 are the two dies that the AMD RDNA 4 GPUs are currently available in. The top Navi 48 GPU die has a maximum capacity of 64 compute units and a minimum capacity of 48 compute units. The entry-level SKU, the Navi 44 GPU, has 32 compute units overall and can be reduced to 28 compute units.
AMD significantly altered how it positioned the RDNA 4 series in the market. Compared to the previous RDNA 3 generation, which had some trouble keeping up with supply in the early months due to its chiplet nature, the lineup only offered two dies and chose a monolithic route, which made it more cost-effective to manufacture and resulted in better supply. Although using chiplets had financial advantages, doing so also necessitated sophisticated packaging, which caused minor production delays.
3) Memory & Bandwidth: Expected Improvements
- High-End Models: Possible 384-bit to 512-bit memory bus, rivaling NVIDIA’s RTX 5090.
- Mainstream GPUs: Adoption of LPDDR5X memory, balancing bandwidth with efficiency.
- Infinity Cache Upgrades: Larger and faster cache, reducing reliance on external memory.
These enhancements will ensure that the RDNA 5 GPU can efficiently feed its massive core arrays, preventing performance-related bottlenecks.
4) Performance Expectations
Although no official benchmarks exist, the rumored specifications allow for some reasonable predictions.
- 3× Shader Performance vs RDNA 4—thanks to doubled CU core density.
- Better Ray Tracing—redesigned CUs could improve AMD’s weaker RT performance.
- AI Acceleration—potential UDNA integration may allow AMD to challenge NVIDIA’s Tensor Cores.
- Efficiency Gains – AMD is likely to use TSMC’s 3nm or 4nm process, improving performance-per-watt.
This would finally give Radeon a real chance at competing with NVIDIA in both gaming and AI workloads.
5) Competing with NVIDIA’s RTX 50 Series
NVIDIA currently dominates in ray tracing and AI thanks to DLSS, Tensor Cores, and CUDA. However, AMD may be closing the gap with RDNA 5.
- Rasterization: AMD could lead in raw performance, thanks to 12,000+ cores.
- Ray Tracing: Still uncertain, but redesigned CUs may offer a big leap.
- AI Features: If UDNA is real, AMD could bring FSR 4.0 or AI-driven upscaling that rivals DLSS.
- Ecosystem: NVIDIA’s CUDA advantage will remain strong, but AMD may expand ROCm for developers.
If AMD executes well, the RDNA 5 flagship could challenge the RTX 5090, offering gamers real competition at the top end of the GPU market.
Also Read: The NVIDIA N1X PC chip comes with 20 CPU cores and 6144 CUDA cores, according to a leak!
6) Expected Launch:
Although AMD has not formally disclosed the release dates, industry insiders have recommended the following launch:
- 2026 – platform validation, developer support, and early engineering samples.
- 2027 – The flagship desktop GPU launches first.
- 2027-2028 – Midrange and mobile GPUs will follow.
7) Conclusion:
AMD is preparing a strong and competitive GPU lineup in all segments, ranging from entry-level to enthusiast-class, if these rumors of early UDNA (RDNA 5) are correct. With next-gen GDDR7 memory or LPDDR5X memory, the high-end Navi 5X, featuring 96 CUs and a 384-bit bus, could revolutionize the GPU space.
AMD’s next-generation RDNA 5 GPUs are looking to be a radical leap. With 128 cores per CU, up to 12,000+ cores on the top model, broader memory buses, and a potential unified UDNA architecture, this generation may redefine the Radeon GPU’s capabilities.
For gamers, that translates to higher frame rates, more intense ray tracing, and more intelligent AI capabilities. For pros, it might mean more compute prowess. And for the industry overall, it might mean finally putting real competition back at the very top of the GPU industry.
As per usual with leaks, these should all be approached with some skepticism. But if only half of them are correct, AMD’s RDNA 5 GPUs could give us what fans have been clamoring for: a real flagship Radeon that can take on NVIDIA’s top guns.
As AMD nears production and eventual release, more is on the horizon.
