The NVIDIA GeForce RTX 5090 D is a graphics card designed specifically for the China market, launching on January 30, 2025. It features a 4 nm process and is based on the GB202 graphics processor, supporting DirectX 12 Ultimate. The card includes 21,760 CUDA cores and 32 GB of GDDR7 memory, with a 512-bit interface.
While it delivers similar gaming performance to the RTX 5090, its AI performance is limited to 2,375 TOPS, a reduction of approximately 29.15% from the original 3,352 TOPS. The GPU boasts a base clock of 2017 MHz and a boost clock of 2407 MHz. This card is specifically hampered in terms of Ai and Crpyto-Mining performance to be in compliance with US export restrictions placed on technology sold in the Chinese market.
Nvidia GeForce RTX 5090 D Specs
The Nvidia GeForce RTX 5090 D. It’s a high-end graphics card from Nvidia, and here’s a summary of its key specifications based on current information:
Key Specs:
- GPU: GB202 (Blackwell architecture)
- CUDA Cores: 21,760
- Boost Clock: 2.41 GHz
- Memory: 32GB GDDR7
- Memory Interface: 512-bit
- Memory Bandwidth: 1,792 GB/s
- Total Board Power (TDP): 575W
Key Differences from the RTX 5090:
The key distinction is that the RTX 5090 D has significantly reduced AI capabilities due to restrictions for the Chinese market, while keeping the same specifications for gaming performance and memory as the standard RTX 5090. Essentially, the RTX 5090 D has the same hardware but limited AI features to follow export regulations.
Other Features:
- 4th-gen Ray Tracing Cores
- PCIe 5.0 interface
Important Notes:
- Information about the RTX 50 series is still emerging, so some specifications might change slightly upon official release.
- The RTX 5090 D appears to be a variant targeted at specific markets, potentially with a focus on gaming performance over AI capabilities.
Comparing the RTX 5090 D, 5080, 5070 Ti, and 5070
| Feature | RTX 5090 D | RTX 5080 | RTX 5070 Ti | RTX 5070 |
|---|---|---|---|---|
| GPU Engine Specs | ||||
| CUDA Cores | 21760 | 10752 | 8960 | 6144 |
| Shader Core | Blackwell | Blackwell | Blackwell | Blackwell |
| Tensor Cores (AI TOPS) | 5th Gen, 2375 TOPS | 5th Gen, 1801 TOPS | 5th Gen, 1406 TOPS | 5th Gen, 988 TOPS |
| RT Cores (TFLOPS) | 4th Gen, 318 TFLOPS | 4th Gen, 171 TFLOPS | 4th Gen, 133 TFLOPS | 4th Gen, 94 TFLOPS |
| Boost Clock (GHz) | 2.41 | 2.62 | 2.45 | 2.51 |
| Base Clock (GHz) | 2.01 | 2.30 | 2.30 | 2.16 |
| Memory Specs | ||||
| Standard Memory Config | 32 GB GDDR7 | 16 GB GDDR7 | 16 GB GDDR7 | 12 GB GDDR7 |
| Memory Interface Width (bit) | 512 | 256 | 256 | 192 |
| Display Support | ||||
| Max Digital Res & Refresh | 4K 480Hz/8K 120Hz (with DSC) | 4K 480Hz/8K 120Hz (with DSC) | 4K 480Hz/8K 120Hz (with DSC) | 4K 480Hz/8K 120Hz (with DSC) |
| Standard Display Connectors | 3x DisplayPort, 1x HDMI | 3x DisplayPort, 1x HDMI | 3x DisplayPort, 1x HDMI | 3x DisplayPort, 1x HDMI |
| Multi-Monitor Support | 4 | 4 | 4 | 4 |
| HDCP | 2.3 | 2.3 | 2.3 | 2.3 |
| Technology Support | ||||
| NVIDIA Architecture | Blackwell | Blackwell | Blackwell | Blackwell |
| Ray Tracing Support | Yes | Yes | Yes | Yes |
| NVIDIA DLSS | DLSS 4 | DLSS 4 | DLSS 4 | DLSS 4 |
| NVIDIA Reflex | Reflex 2 (Coming Soon) | Reflex 2 (Coming Soon) | Reflex 2 (Coming Soon) | Reflex 2 (Coming Soon) |
| NVIDIA Broadcast App | Yes | Yes | Yes | Yes |
| PCIe Gen 5 Support | Yes | Yes | Yes | Yes |
| Resizable BAR Support | Yes | Yes | Yes | Yes |
| NVIDIA App Support | Yes | Yes | Yes | Yes |
| NVIDIA Ansel Support | Yes | Yes | Yes | Yes |
| NVIDIA FreeStyle Support | Yes | Yes | Yes | Yes |
| NVIDIA ShadowPlay Support | Yes | Yes | Yes | Yes |
| NVIDIA Highlights Support | Yes | Yes | Yes | Yes |
| NVIDIA G-SYNC Support | Yes | Yes | Yes | Yes |
| Game Ready Drivers | Yes | Yes | Yes | Yes |
| NVIDIA Studio Drivers | Yes | Yes | Yes | Yes |
| NVIDIA Omniverse Support | Yes | Yes | Yes | Yes |
| RTX Remix Support | Yes | Yes | Yes | Yes |
| DirectX 12 Ultimate | Yes | Yes | Yes | Yes |
| NVIDIA GPU Boost | Yes | Yes | Yes | Yes |
| NVIDIA NVLink (SLI) | No | No | No | No |
| Vulkan/OpenGL Support | Yes (1.4/4.6) | Yes (1.4/4.6) | Yes (1.4/4.6) | Yes (1.4/4.6) |
| NVIDIA Encoder (NVENC) | 3x Gen 9 | 2x Gen 9 | 2x Gen 9 | 1x Gen 9 |
| NVIDIA Decoder (NVDEC) | 2x Gen 6 | 2x Gen 6 | 1x Gen 6 | 1x Gen 6 |
| AV1 Encode/Decode | Yes/Yes | Yes/Yes | Yes/Yes | Yes/Yes |
| CUDA Capability | 12.8 | 12.8 | 12.8 | 12.8 |
| VR Ready | Yes | Yes | Yes | Yes |
| Card Dimensions | Varies by Manufacturer | 304mm | Varies by Manufacturer | 242mm |
| Length | Varies by Manufacturer | 137mm | Varies by Manufacturer | 112mm |
| Width | Varies by Manufacturer | 2 Slots | Varies by Manufacturer | 2 Slots |
| Slot Width | Varies by Manufacturer | Founders Edition + others | Varies by Manufacturer | Founders Edition + others |
| SFF-Ready | ||||
| Power & Thermal | ||||
| Max GPU Temp (°C) | 90 | 88 | 88 | 85 |
| Total Graphics Power (W) | 575 | 360 | 300 | 250 |
| Recommended System Power (W) | 1000 | 850 | 750 | 650 |
| Power Connectors | 4x 8-pin PCIe or 1x 600W+ PCIe Gen 5 | 3x 8-pin PCIe or 1x 450W+ PCIe Gen 5 | 2x 8-pin PCIe or 1x 300W+ PCIe Gen 5 | 2x 8-pin PCIe or 1x 300W+ PCIe Gen 5 |
Key improvements in this table:
- English Translation: All terms are now in clear English.
- Concise Formatting: The table is structured for easier readability.
- Combined Related Info: Related specs (like clock speeds or memory) are grouped together.
- Clarifications: Added notes where necessary (like “with DSC” for display resolutions or “Varies by Manufacturer” for card dimensions).
Key Takeaways
- The RTX 5090 D offers unprecedented graphics processing power with 21,760 CUDA cores
- It supports DirectX 12 Ultimate, ensuring compatibility with next-generation gaming features
- The card’s 32 GB of GDDR7 memory provides ample bandwidth for demanding applications
Graphics Processing Unit
The GeForce RTX 5090 uses NVIDIA’s GB202 GPU, built on TSMC’s 4 nm process. This Blackwell 2.0 architecture chip packs 92.2 billion transistors into a 744 mm² die. The impressive transistor density of 123.9 million per square millimeter showcases NVIDIA’s advanced design capabilities.
GPU Frequencies
The RTX 5090 D boasts impressive clock speeds. Its base clock runs at 2017 MHz. The boost clock reaches 2407 MHz. Memory clocks at 2209 MHz, delivering 28 Gbps effective speed.
| Clock Type | Frequency |
|---|---|
| Base | 2017 MHz |
| Boost | 2407 MHz |
| Memory | 2209 MHz |
These high frequencies contribute to the GPU’s performance in gaming and compute tasks. Overclocking may push speeds even further for enthusiasts seeking maximum power.
Memory Specifications
The RTX 5090 D boasts 32GB of GDDR7 memory. This high-capacity RAM connects via a 512-bit bus. The memory bandwidth reaches an impressive 1.79 TB/s.
Rendering Capabilities
The graphics card boasts impressive rendering specs. It features 21,760 shading units for complex visual effects. 680 TMUs handle texture mapping tasks. 192 ROPs process final pixel output.
170 streaming multiprocessors contain 680 Tensor Cores for AI acceleration and 170 RT Cores dedicated to ray tracing. Each SM has 128 KB of L1 cache, with 88 MB of L2 cache shared across the GPU.
Theoretical Computing Power
The RTX 5090 D boasts impressive theoretical performance figures. Its pixel rate reaches 462.1 GPixel/s, while the texture rate stands at 1,637 GTexel/s.
For AI tasks, the GPU delivers 104.8 TFLOPS in FP16 and FP32 precision. Double precision (FP64) calculations run at 1.637 TFLOPS.
These numbers suggest potent graphics rendering and AI capabilities. Gamers can expect high frame rates and stunning visuals. AI researchers will benefit from the card’s processing power for machine learning workloads.
Board Design and Connectivity
The RTX 5090 D features a dual-slot design. It measures 304 mm in length and 137 mm in width. The card’s height is 48 mm. It uses a 16-pin power connector and has a TDP of 575 W. NVIDIA recommends a 950 W power supply for systems using this GPU.
Output options include:
- 1x HDMI 2.1b
- 3x DisplayPort 2.1b
The card uses PCIe 5.0 for system connectivity. Its board number is PG145.
Graphics Features
The RTX 5090 supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. It uses Shader Model 6.8 and CUDA 10.1. NVIDIA DLSS 4 enhances image quality. G-Sync reduces screen tearing.
| Feature | Version |
|---|---|
| DirectX | 12 Ultimate |
| OpenGL | 4.6 |
| Vulkan | 1.4 |
| CUDA | 10.1 |
NVIDIA Studio Drivers optimize performance for creative apps. The GPU excels in ray tracing and AI-powered tasks.
GB202 GPU Specifications
The GB202 GPU boasts advanced features for high-performance computing. It includes 4th generation ray tracing cores and 5th generation tensor cores. The chip supports 9th generation NVENC and 6th generation NVDEC for video encoding and decoding. It also features VP13 PureVideo HD technology and the latest VDPAU Feature Set M.
Retail Models of the GeForce RTX 5090 D
Several manufacturers offer custom versions of the GeForce RTX 5090 D graphics card. These models share core specifications but differ in cooling solutions and physical dimensions.
AX Gaming’s RTX 5090 D X3W features a triple-slot design and measures 358 mm in length. Colorful offers three variants: the Advanced OC, Neptune OC, and Vulcan OC. All three use triple-slot coolers. The Vulcan W OC model also employs a triple-slot design.
GALAX presents the RTX 5090 D 1-Click OC, a triple-slot card measuring 326 mm. Inno3D produces two models: the Night X3 (335 mm, triple-slot) and the Super Ice Dragon (334 mm, quad-slot).
ZOTAC rounds out the lineup with its RTX 5090 D SOLID, another triple-slot design.
All these models share identical clock speeds:
| Specification | Clock Speed |
|---|---|
| GPU Base Clock | 2017 MHz |
| Boost Clock | 2407 MHz |
| Memory Clock | 2209 MHz |
Buyers can choose based on brand preference, cooling requirements, and case compatibility. The quad-slot Inno3D Super Ice Dragon may offer superior cooling but requires more space.
