🏠IT之家•Stalecollected in 6h
AMD Next-Gen RDNA 5 GPUs Delayed to 2027
💡Understand the shifting GPU hardware landscape to plan your local AI training infrastructure effectively.
⚡ 30-Second TL;DR
What Changed
RDNA 5 GPU launch window projected for Q2-Q3 2027 or early 2028.
Why It Matters
The delay in consumer GPU cycles suggests a longer lifespan for current hardware, impacting AI developers who rely on local consumer-grade GPU clusters for training and inference.
What To Do Next
Re-evaluate your hardware upgrade roadmap for local AI development, as the next major performance leap in consumer GPUs is delayed by at least a year.
Who should care:Developers & AI Engineers
Key Points
- •RDNA 5 GPU launch window projected for Q2-Q3 2027 or early 2028.
- •The lineup will focus on mainstream market segments with various CU configurations.
- •NVIDIA's potential RTX 50 SUPER refresh may delay the arrival of the RTX 60 series.
🧠 Deep Insight
Web-grounded analysis with 29 cited sources.
🔑 Enhanced Key Takeaways
- •RDNA 5 is rumored to feature significant architectural enhancements, including a fully functional dual-issue VALU pipeline for Wave32, new neural arrays, and "Radiance" ray-tracing cores, aiming for substantial improvements in shader and ray tracing performance.
- •Contrary to an initial focus on mainstream, RDNA 5 is also expected to target the ultra high-end market, potentially featuring up to 96 Compute Units and a 384-bit GDDR7 memory bus, representing a 50% increase in CUs over the current top RDNA 4 GPUs.
- •The RDNA 5 architecture, internally codenamed GFX1310, is anticipated to be fabricated on TSMC's N3E process node, with rumors hinting at production by Q2 2026.
- •NVIDIA's next-generation GeForce RTX 60 series, based on the "Rubin" architecture and built on a TSMC 3nm FinFET node, is rumored to offer a 2x increase in path-tracing performance and a 30-35% gain in pure rasterization over the RTX 50 series.
- •NVIDIA is reportedly still planning an RTX 50 "Super" refresh for 2026, which could include models like the RTX 5060 Super with 12GB VRAM, RTX 5070 Super with 18GB, and RTX 5070 Ti Super/RTX 5080 Super with 24GB, utilizing 3GB GDDR7 chips.
📊 Competitor Analysis▸ Show
Competitor Analysis: AMD RDNA 5 vs. NVIDIA RTX Series
| Feature/Category | AMD RDNA 5 (Rumored) | NVIDIA GeForce RTX 50 Series (Blackwell) | NVIDIA GeForce RTX 60 Series (Rubin - Rumored) |
|---|---|---|---|
| Architecture | RDNA 5 (GFX1310) | Blackwell | Rubin |
| Process Node | TSMC N3E | TSMC 4N | TSMC 3nm FinFET |
| Launch Window | Q2-Q3 2027 or early 2028 | January 2025 (initial models) | Late 2027 or early 2028 |
| Target Market | Mainstream to Ultra High-End | Entry-level to Enthusiast | High-End to Enthusiast |
| Max Compute Units/SMs | Up to 96 CUs (rumored high-end) | Up to 170 SMs (RTX 5090) | Up to 192 SMs (RTX 6090) |
| Memory Bus | Up to 384-bit GDDR7 (rumored high-end) | Up to 512-bit GDDR7 (RTX 5090) | Up to 512-bit GDDR7 (RTX 6090) |
| Max VRAM | Up to 24GB GDDR7 (rumored high-end) | Up to 24GB GDDR7 (RTX 5090) | Up to 32GB GDDR7 (RTX 6090) |
| Ray Tracing Performance | Significant improvements with "Radiance" cores (30%+ per CU rumored) | Fourth-gen RT Cores | 2x increase in path-tracing over RTX 50 series (rumored) |
| Rasterization Performance | Significant boost from dual-issue VALU pipeline | Successor to RTX 40 series | 30-35% gain over RTX 50 series (rumored) |
| AI Capabilities | New neural array, FSR Diamond project for ML-based upscaling/frame generation | Fifth-gen Tensor Cores, DLSS 4 | Sixth-gen Tensor Cores, DLSS 5 (2x performance over DLSS 4 rumored) |
🛠️ Technical Deep Dive
- Dual-Issue VALU Pipeline for Wave32: The RDNA 5 architecture (GFX1310) is designed with a fully functional dual-issue VALU pipeline for Wave32. This allows vector operations (VOPs) to be issued simultaneously to the GPU's X and Y arithmetic logic unit (ALU) lanes. This design expands the range of fused multiply-add (FMA) and other VOP instructions eligible for dual-issue, relaxing register constraints and enabling compilers and shader code to perform more intensive floating-point work in Wave32 mode. This is expected to significantly enhance FP32 compute utilization and boost game shader performance.
- Neural Array: RDNA 5 is anticipated to include a new neural array. This component is intended to improve communication between compute units and shader arrays, aiming to enhance throughput, reduce overhead, and boost performance in ray tracing and upscaling tasks.
- Radiance Ray-Tracing Cores: The architecture is rumored to feature new "Radiance" ray-tracing cores, which represent AMD's most significant ray tracing revamp since RDNA 2. Forecasts suggest at least a 30% increase in ray tracing capabilities per Compute Unit.
- Universal Compression: This improvement aims to reduce wasted bandwidth by compressing more data as it moves through the GPU and into VRAM, potentially optimizing memory usage even with high-speed GDDR7.
- Process Node: RDNA 5 is expected to be fabricated on TSMC's N3E process node.
- Compute Unit & Memory Configuration (High-End Rumors): Higher-end RDNA 5 GPUs are rumored to feature up to 96 Compute Units and a 384-bit GDDR7 memory bus. This configuration could potentially offer 24 GB of VRAM and over 1300 GB/s of bandwidth, which is more than double that of the RX 7900 XT.
🔮 Future ImplicationsAI analysis grounded in cited sources
AMD's RDNA 5 delay could intensify competition in the high-end GPU market upon its eventual release.
The rumored advanced architectural features and increased CU counts suggest AMD is aiming to be highly competitive, potentially challenging NVIDIA's high-end offerings, but the delay means NVIDIA has more time to solidify its position with current and upcoming architectures.
The synchronized delays from both AMD and NVIDIA for their next-gen architectures indicate broader industry challenges.
Factors such as ongoing DRAM shortages and the high demand for GPUs in the AI sector are likely impacting manufacturing capacity and component availability, pushing back consumer GPU launch timelines for both major players.
RDNA 5's focus on enhanced AI and ray tracing capabilities signifies AMD's strategic shift towards more competitive performance in these critical modern gaming and compute workloads.
The introduction of neural arrays and improved Radiance ray-tracing cores suggests AMD is directly addressing areas where NVIDIA has historically held a strong lead, aiming to close the performance gap.
⏳ Timeline
2019-07
AMD launches RDNA 1 architecture with Radeon RX 5000 series, replacing GCN.
2020-11
AMD launches RDNA 2 architecture with Radeon RX 6000 series, featuring hardware ray tracing and Infinity Cache. RDNA 2 also powers PlayStation 5 and Xbox Series X/S.
2022-11
AMD announces RDNA 3 architecture, promising over 50% performance-per-watt uplift and chiplet design.
2022-12
Radeon RX 7000 series, based on RDNA 3, is released.
2025-01
AMD previews RDNA 4 architecture and Radeon RX 9000 series at CES.
2025-03
AMD launches Radeon RX 9000 series, based on RDNA 4, targeting mainstream and upper-mainstream segments.
📎 Sources (29)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
- techpowerup.com
- noobfeed.com
- hothardware.com
- wccftech.com
- techpowerup.com
- overclock3d.net
- tweaktown.com
- tomshardware.com
- pcmag.com
- digg.com
- youtube.com
- wikipedia.org
- tomshardware.com
- videocardz.com
- hardforum.com
- tomshardware.com
- wikipedia.org
- grafickekarty.sk
- newegg.com
- wikipedia.org
- notebookcheck.net
- tomshardware.com
- wikipedia.org
- tomshardware.com
- club386.com
- hothardware.com
- pcgamer.com
- wikipedia.org
- wikipedia.org
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: IT之家 ↗