Quick/Comparisons
The GeForce RTX 5090 takes the lead in gaming performance, boasting a perfect score of 100, while the GeForce RTX 4090 trails behind with a score of 92.
The GeForce RTX 5090 edges out the GeForce RTX 4090 in content creation, with a score of 93 to the latter's 91, thanks to its slightly stronger workstation performance.
The GeForce RTX 5090 dominates in AI and machine learning, with a score of 95, while the GeForce RTX 4090 lags behind with a score of 76, due to its significantly lower AI/ML capabilities.
92 | Gaming | 100 |
91 | Workstation | 93 |
76 | AI/ML |
Compare with something else?
Compare GeForce RTX 4090 with another →95 |
43 | Energy Efficiency | 49 |
89 | Quick Comparison Final Score | 96 |
| AD102 | Graphics Processor | GB202-300 |
| 16384 | Cores | 21760 |
| 512 / 176 | TMUs / ROPs | 680 / 176 |
| 11688.48 | Blender GPU | 15010.34 |
| 203 | Forza Horizon 5 | 244 |
| 270 | The Witcher 3 | 323 |
| 245 | Counter-Strike 2 | 265 |
| 183 | Far Cry 6 | 202 |
| 165 | Hogwarts Legacy | 207 |
| 269 | Call of Duty | 327 |
| 150 | Ghost of Tsushima | 179 |
| 204 | Cyberpunk 2077 | 229 |
| 283 | Tomb Raider | 299 |
| 219 | Average FPS | 253 |
| 256 | 1080p High | 294 |
| 219 | 1080p Ultra | 253 |
| 176 | 1440p Ultra | 207 |
| 107 | 4K Ultra | 124 |
| Low | Margin of Error | Low |
| 2.38 Gpixels/sec | Feature Matching | 2.53 Gpixels/sec |
| 1950 Gpixels/sec | Stereo Matching | 2850 Gpixels/sec |
| 48090.4 FPS | Particle Physics | 59146.9 FPS |
| OpenCL | API | OpenCL |
| 316301 | GB6 Compute Score | 419309 |
| 318.9 img/sec | Background Blur | 333 img/sec |
| 222.8 img/sec | Face Detection | 298.5 img/sec |
| 14.2 Gpixels/sec | Horizon Detection | 21.7 Gpixels/sec |
| 21 Gpixels/sec | Edge Detection | 33.7 Gpixels/sec |
| 23.9 Gpixels/sec | Gaussian Blur | 34.5 Gpixels/sec |
| 38071 | G3D Mark Score | 38955 |
| 1303 | G2D Mark | 1413 |
| 324 FPS | DirectX 11 | 341 FPS |
| 151 FPS | DirectX 12 | 178 FPS |
| 26193 Ops/s | GPU Compute | 24632 Ops/s |
| 42962 | GB6 ML Single Precision | 53861 |
| 59636 | GB6 ML Half Precision | 81080 |
| 31670 | GB6 ML Quantized | 38938 |
| 15600 | Image Classification (SP) | 18948 |
| 36657 | Image Segmentation (HP) | 55109 |
| 48114 | Image Super Resolution (Q) | 58478 |
| 78111 | Face Detection (HP) | 114685 |
| 289639 | Pose Estimation (Q) | 344729 |
| 4074 | Text Classification (SP) | 4464 |
| 6349 | Machine Translation (HP) | 8184 |
| 21186 | Object Detection (SP) | 27802 |
| 75599 | Depth Estimation (Q) | 85572 |
| 601428 | Style Transfer (SP) | 687080 |
| ONNX | Framework | ONNX |
| DirectML | Backend | DirectML |
| 444 GPixel/s | Pixel Fill Rate | 424 GPixel/s |
| 1290 GTexel/s | Texture Fill Rate | 1637 GTexel/s |
| 82.6 TFLOPS | FLOPS (FP32) | 104.8 TFLOPS |
| GDDR6X | Memory Type | GDDR7 |
| 24 GB | Memory Size | 32 GB |
| 2625 MHz | Memory Clock | 1750 MHz |
| 21000 Mbps | Effective Memory Speed | 28000 Mbps |
| 384-bit | Bus | 512-bit |
| No | ECC | No |
| 1010 GB/s | Memory Bandwidth | 1792 GB/s |
| PCIe 4.0 x16 | Interface | PCIe 5.0 x16 |
| 450 W | TGP | 575 W |
| TSMC | Manufacturing | TSMC |
| 5 nm | Fabrication Process | 4 nm |
| 609 mm² | Die Size | 750 mm² |
| 76.3 billion | Transistor Count | 92.2 billion |
| 125.29 MTr/mm² | Transistor Density | 122.93 MTr/mm² |
| 12 | DirectX | 12.2 |
| 1.3 | Vulkan | 1.4 |
| 4.6 | OpenGL | 4.6 |
| 3 | OpenCL | 3 |
| 8.9 | CUDA | 12 |
| Yes | Ray Tracing | Yes |
| DLSS 3 | DLSS | DLSS 4 |
| 1.4a | DisplayPort | 2.1b |
| Nvidia | Vendor | Nvidia |
| Discrete | Build | Discrete |
| September 20, 2022 | Released | January 7, 2025 |
| Desktop | Case | Desktop |
| Gaming | Purpose | Gaming |
| High-end | Segment | High-end |
| Ada Lovelace | Architecture | Blackwell 2.0 |
| AD102 | GPU Codename | GB202-300 |
| -Intel Core i9 14900Kor above | Recommended CPU | -Intel Core Ultra 9 285Kor above |
| 42169 | Steel Nomad Lite Score | 52638 |
| 36311 | Time Spy | 46941 |
| 187386 | Solar Bay | 235224 |
| 26128 | Port Royal | 38113 |
| 72387 | Fire Strike | 87433 |
| 85230 | Wild Life Extreme | 108736 |
| 195880 | Night Raid | 207058 |
| 2235 MHz | Base Clock | 2017 MHz |
| 2520 MHz | Boost Clock | 2407 MHz |
| 16384 | Shading Units | 21760 |
| 512 | Texture Mapping Units (TMUs) | 680 |
| 176 | Render Output Units (ROPs) | 176 |
| 128 | Compute Units (Pipelines) | 170 |
| 512 | Tensor Cores | 680 |
| 128 | Ray-tracing Cores | 170 |
| 128KB per cluster | L1 Cache | 128KB per cluster |
| 72MB shared | L2 Cache | 96MB shared |
| 2 IPC | Instructions Per Cycle | 2 IPC |