Radeon RX 9000 series

The Radeon RX 9000 series and RDNA 4 architecture were officially previewed on January 6, 2025 during AMD's CES keynote in Las Vegas. AMD were light on concrete details surrounding the RDNA 4 architecture or the Radeon RX 9000 series during their CES keynote. The Radeon RX 9000 series targets midrange performance and value rather than competing with Nvidia at the high-end like the Radeon RX 7000 series did. This is a similar approach taken by the RX 5000 series in 2019. On January 8, 2025, reports surfaced that U.S. retailer B&H would begin pre-orders for the Radeon RX 9000 series on January 23.

The Radeon RX 9070 series was revealed on February 28, 2025 in an AMD live stream event.

The Radeon RX 9060 XT was revealed on May 21, 2025 during AMD's Computex keynote.

Features

RDNA 4 architecture

The RDNA 4 architecture used by the Radeon RX 9000 series is, according to AMD, focused on improved ray tracing performance and expanded AI acceleration capabilities with an "optimized" Compute Unit design.

RDNA 4 architecture

• RDNA 4 architecture built on TSMC 4 nm process (TSMC 4N Gen 5 Display Engine)
• AMD RDNA 4 Compute Units with redesigned 3rd generation Raytracing Accelerators for improved ray tracing performance and image quality
• 2nd Generation AI Accelerators with support for FP16, INT8 operations, and sparsity acceleration enabling up to 4× FP16 and 8× INT8 throughput for AI workloads
• AMD HYPR-RX1 technology combining Radeon Super Resolution, FidelityFX Super Resolution 4, Radeon Anti-Lag 24, Radeon Boost, and AMD Fluid Motion Frames 2.1 for advanced AI-based upscaling and frame generation
• PCIe 5.0 support for high bandwidth GPU-to-CPU communication
• Display connectivity includes DisplayPort 2.1a and HDMI 2.1b with support for high refresh rates and resolutions
• AMD Infinity Cache 3rd generation with up to 64 MB cache to reduce memory latency and increase bandwidth efficiency
• Memory subsystem supports up to 16 GB GDDR6 with up to 640 GB/s memory bandwidth depending on model and interface width
• Advanced media engine optimized for ultra-fast video encoding/decoding and enhanced streaming capabilities
• No dedicated multi-GPU or NVLink equivalent support (focus on single GPU scalability)
• Double-precision (FP64) performance of RDNA 4 architecture is significantly lower than single-precision (FP32), optimized primarily for gaming and AI workloads rather than HPC use cases

FSR 4