1 / 5


Turbine fan design with front-to-back air duct, ideal for forced convection cooling in server enclosures and multi-card deployments.
Dual-slot thickness with multiple height specifications (115/120/125/135mm), compatible with mainstream server enclosures.
5th Gen Tensor Cores offering up to 3352 AI TOPS, perfect for large model inference, deep learning, and scientific computing.
All-metal construction using industrial-grade materials, supporting 7x24-hour high-load continuous operation.
| Category | Specification |
| GPU Model | GeForce RTX 5090 (GB202-300) |
| Architecture | Blackwell (TSMC 4N) |
| CUDA Cores | 21760 |
| Base Clock | 2017 MHz |
| Boost Clock | 2437–2467 MHz |
| FP32 Performance | 104.8 TFLOPS |
| Ray Tracing Cores | 4th Generation, 318 RT-TFLOPS |
| Tensor Cores | 5th Generation, 3352 AI TOPS |
| Memory Size | 32 GB |
| Memory Type | GDDR7 |
| Memory Interface | 512-bit |
| Memory Speed | 28 Gbps |
| Memory Bandwidth | 1792 GB/s |
| Bus Interface | PCIe 5.0 x16 |
| TDP | 575 W |
| Power Connector | 16-pin (12VHPWR) / 12V-2×6 |
| Display Connectors | 3×DisplayPort 2.1b, 1×HDMI 2.1b |
| Max Resolution | 8K@165Hz, 4K@480Hz |
| Form Factor | Dual-slot, Blower/Turbo Fan |
| Dimensions | 266.7mm (L) × 37.5mm (W) |
| Recommended PSU | 1000W+ |