Quick/Comparisons
The GeForce RTX 4090 outperforms the GeForce RTX 3070 Ti by a significant margin, thanks to its superior Ada Lovelace architecture and 24GB GDDR6X memory. Its 92 score is a testament to its exceptional gaming capabilities.
The GeForce RTX 4090 takes the lead in content creation, boasting a 91 score, while the GeForce RTX 3070 Ti trails behind with a 49 score. This is largely due to the 4090's stronger workstation performance and more extensive memory capacity.
The GeForce RTX 4090 excels in AI/ML compute, with a 76 score, outpacing the GeForce RTX 3070 Ti's 39 score. This is attributed to the 4090's capable AI tensor cores and more advanced Ada Lovelace architecture.
46 | Gaming | 92 |
49 | Workstation | 91 |
39 | AI/ML |
Compare with something else?
Compare GeForce RTX 3070 with another →76 |
35 | Energy Efficiency | 43 |
46 | Quick Comparison Final Score | 89 |
| GA104 | Graphics Processor | AD102 |
| 6144 | Cores | 16384 |
| 192 / 96 | TMUs / ROPs | 512 / 176 |
| 3746.83 | Blender GPU | 11688.48 |
| 104 | Forza Horizon 5 | 203 |
| 143 | The Witcher 3 | 270 |
| 198 | Counter-Strike 2 | 245 |
| 125 | Far Cry 6 | 183 |
| 85 | Hogwarts Legacy | 165 |
| 120 | Call of Duty | 269 |
| 81 | Ghost of Tsushima | 150 |
| 102 | Cyberpunk 2077 | 204 |
| 161 | Tomb Raider | 283 |
| 124 | Average FPS | 219 |
| 150 | 1080p High | 256 |
| 124 | 1080p Ultra | 219 |
| 93 | 1440p Ultra | 176 |
| 54 | 4K Ultra | 107 |
| Low | Margin of Error | Low |
| 136096 | GB6 Compute Score | 316301 |
| 197 img/sec | Background Blur | 318.9 img/sec |
| 128.1 img/sec | Face Detection | 222.8 img/sec |
| 6.56 Gpixels/sec | Horizon Detection | 14.2 Gpixels/sec |
| 8.81 Gpixels/sec | Edge Detection | 21 Gpixels/sec |
| 6.47 Gpixels/sec | Gaussian Blur | 23.9 Gpixels/sec |
| 1.26 Gpixels/sec | Feature Matching | 2.38 Gpixels/sec |
| 609.3 Gpixels/sec | Stereo Matching | 1950 Gpixels/sec |
| 18205 FPS | Particle Physics | 48090.4 FPS |
| OpenCL | API | OpenCL |
| 23260 | G3D Mark Score | 38071 |
| 1061 | G2D Mark | 1303 |
| 192 FPS | DirectX 11 | 324 FPS |
| 91 FPS | DirectX 12 | 151 FPS |
| 11452 Ops/s | GPU Compute | 26193 Ops/s |
| 22516 | GB6 ML Single Precision | 42962 |
| 36377 | GB6 ML Half Precision | 59636 |
| 14215 | GB6 ML Quantized | 31670 |
| 8615 | Image Classification (SP) | 15600 |
| 24659 | Image Segmentation (HP) | 36657 |
| 20631 | Image Super Resolution (Q) | 48114 |
| 43834 | Face Detection (HP) | 78111 |
| 98347 | Pose Estimation (Q) | 289639 |
| 3162 | Text Classification (SP) | 4074 |
| 4911 | Machine Translation (HP) | 6349 |
| 11498 | Object Detection (SP) | 21186 |
| 34719 | Depth Estimation (Q) | 75599 |
| 196833 | Style Transfer (SP) | 601428 |
| ONNX | Framework | ONNX |
| DirectML | Backend | DirectML |
| 170 GPixel/s | Pixel Fill Rate | 444 GPixel/s |
| 340 GTexel/s | Texture Fill Rate | 1290 GTexel/s |
| 21.7 TFLOPS | FLOPS (FP32) | 82.6 TFLOPS |
| GDDR6X | Memory Type | GDDR6X |
| 8 GB | Memory Size | 24 GB |
| 1188 MHz | Memory Clock | 2625 MHz |
| 19000 Mbps | Effective Memory Speed | 21000 Mbps |
| 256-bit | Bus | 384-bit |
| No | ECC | No |
| 608.3 GB/s | Memory Bandwidth | 1010 GB/s |
| PCIe 4.0 x16 | Interface | PCIe 4.0 x16 |
| 290 W | TGP | 450 W |
| Samsung | Manufacturing | TSMC |
| 8 nm | Fabrication Process | 5 nm |
| 392 mm² | Die Size | 609 mm² |
| 17 billion | Transistor Count | 76.3 billion |
| 43.37 MTr/mm² | Transistor Density | 125.29 MTr/mm² |
| 12 | DirectX | 12 |
| 1.3 | Vulkan | 1.3 |
| 4.6 | OpenGL | 4.6 |
| 3 | OpenCL | 3 |
| 8.6 | CUDA | 8.9 |
| Yes | Ray Tracing | Yes |
| DLSS 2 | DLSS | DLSS 3 |
| 1.4a | DisplayPort | 1.4a |
| Nvidia | Vendor | Nvidia |
| Discrete | Build | Discrete |
| June 10, 2021 | Released | September 20, 2022 |
| Desktop | Case | Desktop |
| Gaming | Purpose | Gaming |
| Mid-range | Segment | High-end |
| Ampere | Architecture | Ada Lovelace |
| GA104 | GPU Codename | AD102 |
| -Intel Core i7 12700Kor above | Recommended CPU | -Intel Core i9 14900Kor above |
| 15402 | Steel Nomad Lite Score | 42169 |
| 14880 | Time Spy | 36311 |
| 69900 | Solar Bay | 187386 |
| 8884 | Port Royal | 26128 |
| 37238 | Fire Strike | 72387 |
| 31911 | Wild Life Extreme | 85230 |
| 134012 | Night Raid | 195880 |
| 1575 MHz | Base Clock | 2235 MHz |
| 1770 MHz | Boost Clock | 2520 MHz |
| 6144 | Shading Units | 16384 |
| 192 | Texture Mapping Units (TMUs) | 512 |
| 96 | Render Output Units (ROPs) | 176 |
| 48 | Compute Units (Pipelines) | 128 |
| 192 | Tensor Cores | 512 |
| 48 | Ray-tracing Cores | 128 |
| 128KB per cluster | L1 Cache | 128KB per cluster |
| 4MB shared | L2 Cache | 72MB shared |
| 2 IPC | Instructions Per Cycle | 2 IPC |