Radeon RX 9000 series
AMD graphics processing units
From Wikipedia, the free encyclopedia
The Radeon RX 9000 series is a series of consumer graphics processing units developed by AMD, based on the RDNA 4 architecture. The series is targeting the mainstream segment and is the successor to the Radeon RX 7000 series.
 An AMD Radeon RX 9060 XT | |
| Release date | March 6, 2025 |
|---|---|
| Manufactured by | TSMC |
| Designed by | AMD |
| Marketed by | AMD |
| Codename | Navi 4x |
| Architecture | RDNA 4 |
| Transistors |
|
| Fabrication process | TSMC N4P |
| Cards | |
| Entry-level |
|
| Mid-range |
|
| High-end |
|
| API support | |
| OpenCL | OpenCL 2.2 |
| OpenGL | OpenGL 4.6 |
| Vulkan | Vulkan 1.4 |
| DirectX | |
| History | |
| Predecessor | Radeon RX 7000 series |
| Support status | |
| Supported | |
Background
AMD's Q3 2024 earnings call in October 2024 confirmed that RDNA 4 would be releasing in early 2025 with CEO Lisa Su saying that the architecture "delivers significantly higher ray tracing performance and adds new AI capabilities".[1][2]
In December 2024, an AMD advertising campaign tie-in with Call of Duty: Black Ops 6 on Reddit showed a Ryzen 9 processor and what appeared to be the Radeon RX 9070 XT reference design.[3]
The Radeon RX 9000 series and RDNA 4 architecture were officially previewed on January 6, 2025 during AMD's CES keynote in Las Vegas.[4] AMD were light on concrete details surrounding the RDNA 4 architecture or the Radeon RX 9000 series during their CES keynote.[5] 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.[6] 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.[7][8]
The Radeon RX 9070 series was revealed on February 28, 2025 in an AMD live stream event.[9]
The Radeon RX 9060 XT was revealed on May 21, 2025 during AMD's Computex keynote.[10]
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.[11]
RDNA 4 architecture
- RDNA 4 architecture built on TSMC 4 nm process (TSMC 4N Gen 5 Display Engine)[12]
- AMD RDNA 4 Compute Units with redesigned 3rd generation Raytracing Accelerators for improved ray tracing performance and image quality[13]
- 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[13]
- 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[13]
- PCIe 5.0 support for high bandwidth GPU-to-CPU communication[13]
- Display connectivity includes DisplayPort 2.1a and HDMI 2.1b with support for high refresh rates and resolutions[13]
- AMD Infinity Cache 3rd generation with up to 64 MB cache to reduce memory latency and increase bandwidth efficiency[13]
- Memory subsystem supports up to 16 GB GDDR6 with up to 640 GB/s memory bandwidth depending on model and interface width[13]
- Advanced media engine optimized for ultra-fast video encoding/decoding and enhanced streaming capabilities[13]
- No dedicated multi-GPU or NVLink equivalent support (focus on single GPU scalability)[13]
- 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[13]
FSR 4
FidelityFX Super Resolution 4 (FSR 4) is AMD's first machine-learning upscaling solution that is able to leverage the second-generation AI accelerator cores in the RDNA 4 architecture.[14] AMD stated that due to requiring hardware acceleration, FSR 4 was limited to the Radeon RX 9000 series. Call of Duty: Black Ops 6 was first title to integrate FSR 4 upscaling support.[14]
As of August 2025, over 75 games include support for FSR 4.[15]
Products
Desktop
| Radeon RX | 9060[16][17] | 9060 XT[18][19][13][20] | 9070 GRE[21][22] | 9070[23][24] | 9070 XT[25][26] | |
|---|---|---|---|---|---|---|
| Release date | 5 Aug 2025 | 5 Jun 2025 | 8 May 2025 (China) 1 June 2026 (Worldwide) |
6 Mar 2025 | ||
| Launch MSRP | OEM | $299 USD (8 GB) $349 USD (16 GB) |
¥4,199 CNY[a] $549 USD |
$549 USD | $599 USD | |
| GPU die | Navi 44 | Navi 48 | ||||
| Transistors (billion) | 29.7 | 53.9 | ||||
| Die size | 199 mm2 | 356.5 mm2 | ||||
| Core | Stream processors | 1792 | 2048 | 3072 | 3584 | 4096 |
| Texture mapping units | 112 | 128 | 192 | 224 | 256 | |
| Render output units | 64 | 96 | 128 | |||
| Ray accelerators | 28 | 32 | 48 | 56 | 64 | |
| AI accelerators | 56 | 64 | 96 | 112 | 128 | |
| Game frequency (GHz) Boost frequency (GHz) |
2.40 2.99 |
2.53 3.13 |
2.22 2.79 |
2.07 2.52 |
2.40 2.97 | |
| Compute units | 28 | 32 | 48 | 56 | 64 | |
| Cache | L0 | 32 KB per CU | ||||
| L1 | 128 KB per Array | |||||
| L2 | 4 MB | 8 MB | ||||
| L3 | 32 MB | 48 MB | 64 MB | |||
| Memory | Type | GDDR6 | ||||
| Size | 8 GB | 8 GB 16 GB |
12 GB | 16 GB | ||
| Clock (Gb/s) | 18 | 20.1 | 18 | 20.1 | ||
| Bandwidth (GB/s) | 288 | 320 | 432 | 640 | ||
| Bus width | 128-bit | 192-bit | 256-bit | |||
| Fillrate | Pixel (Gpx/s)[b] | 191.4 | 200.3 | 267.8 | 322.6 | 380.2 |
| Texture (Gtex/s)[c] | 334.9 | 400.6 | 535.7 | 564.5 | 760.3 | |
| Processing power |
FP16 (TFLOPS)[d] | 21.4 | 25.6 | 34.3 | 36.1 | 48.7 |
| FP32 (TFLOPS) | 21.4 | 25.6 | 34.3 | 36.1 | 48.7 | |
| AI FP16 (TFLOPS)[e] | 85 | 103 | 137 | 145 | 195 | |
| AI INT8 or FP8 (TOPS)[e] | 171 | 205 | 274 | 289 | 389 | |
| AI INT4 (TOPS)[e] | 343 | 410 | 549 | 578 | 778 | |
| Interface | Host | PCIe 5.0 x16 | ||||
| Power | 1x 8-pin | 2x 8-pin | ||||
| Displays | 1x HDMI 2.1b, 2x DisplayPort 2.1a | 1x HDMI 2.1b, 3x DisplayPort 2.1a | ||||
| TDP | 132 W | 150 W (8 GB) 160 W (16 GB) |
220 W | 304 W | ||
- Original China-only MSRP.
- Pixel fillrate is calculated as the number of render output units (ROPs) multiplied by the base (or boost) core clock speed.
- Texture fillrate is calculated as the number of texture mapping units (TMUs) multiplied by the base (or boost) core clock speed.
- Officially declared performance is 2x shown here due to sparsity.
See also
- Radeon RX 5000 series – first implementation of RDNA architecture
- Radeon RX 6000 series
- Radeon RX 7000 series – AMD's predecessor to Radeon RX 9000 series (RDNA 3 based)
- RDNA (microarchitecture)
- RDNA 4 – microarchitecture used by the RX 9000 series
- List of AMD graphics processing units
- GeForce RTX 50 series – competing Nvidia GPU generation released in a similar time-frame
- Arc B-Series – competing Intel GPU generation released in a similar time-frame
