Quick/Comparisons
The GeForce RTX 5090's perfect gaming score is unmatched by the Nvidia RTX 3500 Laptop (Ada)'s mediocre 43, thanks to its superior performance in this area.
The GeForce RTX 5090's strong workstation performance earns it the win in content creation, outscoring the Nvidia RTX 3500 Laptop (Ada) by 40 points.
The GeForce RTX 5090's capable AI/ML compute capabilities secure its victory in AI & Machine Learning, leaving the Nvidia RTX 3500 Laptop (Ada) far behind with a 56-point gap.
100 | Gaming | 43 |
93 | Workstation | 53 |
95 | AI/ML |
Compare with something else?
Compare GeForce RTX 5090 with another →39 |
49 | Energy Efficiency | 56 |
96 | Quick Comparison Final Score | 47 |
| GB202-300 | Graphics Processor | AD104 |
| 21760 | Cores | 5120 |
| 680 / 176 | TMUs / ROPs | 160 / 64 |
| 15010.34 | Blender GPU | 4377.08 |
| 244 | Forza Horizon 5 | 93 |
| 323 | The Witcher 3 | 134 |
| 265 | Counter-Strike 2 | 191 |
| 202 | Far Cry 6 | 113 |
| 207 | Hogwarts Legacy | 80 |
| 327 | Call of Duty | 111 |
| 179 | Ghost of Tsushima | 72 |
| 229 | Cyberpunk 2077 | 89 |
| 299 | Tomb Raider | 145 |
| 253 | Average FPS | 114 |
| 294 | 1080p High | 138 |
| 253 | 1080p Ultra | 114 |
| 207 | 1440p Ultra | 84 |
| 124 | 4K Ultra | 49 |
| Low | Margin of Error | High |
| 419309 | GB6 Compute Score | 131037 |
| 333 img/sec | Background Blur | 223 img/sec |
| 298.5 img/sec | Face Detection | 145.8 img/sec |
| 21.7 Gpixels/sec | Horizon Detection | 5.87 Gpixels/sec |
| 33.7 Gpixels/sec | Edge Detection | 7.77 Gpixels/sec |
| 34.5 Gpixels/sec | Gaussian Blur | 7.2 Gpixels/sec |
| 2.53 Gpixels/sec | Feature Matching | 1.39 Gpixels/sec |
| 2850 Gpixels/sec | Stereo Matching | 458.7 Gpixels/sec |
| 59146.9 FPS | Particle Physics | 14313 FPS |
| OpenCL | API | OpenCL |
| 38955 | G3D Mark Score | 19719 |
| 1413 | G2D Mark | 712 |
| 341 FPS | DirectX 11 | 187 FPS |
| 178 FPS | DirectX 12 | 83 FPS |
| 24632 Ops/s | GPU Compute | 7623 Ops/s |
| 114685 | Face Detection (HP) | 38189 |
| 344729 | Pose Estimation (Q) | 105500 |
| 4464 | Text Classification (SP) | 3319 |
| 8184 | Machine Translation (HP) | 3845 |
| 27802 | Object Detection (SP) | 10208 |
| 85572 | Depth Estimation (Q) | 38369 |
| 687080 | Style Transfer (SP) | 196359 |
| ONNX | Framework | ONNX |
| DirectML | Backend | DirectML |
| 53861 | GB6 ML Single Precision | 20783 |
| 81080 | GB6 ML Half Precision | 33871 |
| 38938 | GB6 ML Quantized | 16429 |
| 18948 | Image Classification (SP) | 8316 |
| 55109 | Image Segmentation (HP) | 17301 |
| 58478 | Image Super Resolution (Q) | 27105 |
| 424 GPixel/s | Pixel Fill Rate | 99-144 GPixel/s |
| 1637 GTexel/s | Texture Fill Rate | 247-360 GTexel/s |
| 104.8 TFLOPS | FLOPS (FP32) | 15.8-23 TFLOPS |
| GDDR7 | Memory Type | GDDR6 |
| 32 GB | Memory Size | 12 GB |
| 1750 MHz | Memory Clock | 2250 MHz |
| 28000 Mbps | Effective Memory Speed | 18000 Mbps |
| 512-bit | Bus | 192-bit |
| No | ECC | Yes |
| 1792 GB/s | Memory Bandwidth | 432 GB/s |
| PCIe 5.0 x16 | Interface | PCIe 4.0 x16 |
| 575 W | TGP | 60-140 W (configurable) |
| TSMC | Manufacturing | TSMC |
| 4 nm | Fabrication Process | 5 nm |
| 750 mm² | Die Size | 294 mm² |
| 92.2 billion | Transistor Count | 35.8 billion |
| 122.93 MTr/mm² | Transistor Density | 121.77 MTr/mm² |
| 12.2 | DirectX | 12 |
| 1.4 | Vulkan | 1.3 |
| 4.6 | OpenGL | 4.6 |
| 3 | OpenCL | 3 |
| 12 | CUDA | 8.9 |
| Yes | Ray Tracing | Yes |
| DLSS 4 | DLSS | DLSS 3 |
| 2.1b | DisplayPort | 1.4a |
| Nvidia | Vendor | Nvidia |
| Discrete | Build | Discrete |
| January 7, 2025 | Released | March 21, 2023 |
| Desktop | Case | Laptop |
| Gaming | Purpose | Professional |
| High-end | Segment | Mid-range |
| Blackwell 2.0 | Architecture | Ada Lovelace |
| GB202-300 | GPU Codename | AD104 |
| -Intel Core Ultra 9 285Kor above | Recommended CPU | -Intel Core Ultra 7 165Hor above |
| 52638 | Steel Nomad Lite Score | 13536 |
| 46941 | Time Spy | 12966 |
| 235224 | Solar Bay | 52474 |
| 38113 | Port Royal | 7976 |
| 87433 | Fire Strike | 32977 |
| 108736 | Wild Life Extreme | 26514 |
| 207058 | Night Raid | 110011 |
| 2017 MHz | Base Clock | — |
| 2407 MHz | Boost Clock | 1545-2250 MHz (configurable) |
| 21760 | Shading Units | 5120 |
| 680 | Texture Mapping Units (TMUs) | 160 |
| 176 | Render Output Units (ROPs) | 64 |
| 170 | Compute Units (Pipelines) | 40 |
| 680 | Tensor Cores | 160 |
| 170 | Ray-tracing Cores | 40 |
| 128KB per cluster | L1 Cache | 128KB per cluster |
| 96MB shared | L2 Cache | 48MB shared |
| 2 IPC | Instructions Per Cycle | 2 IPC |