Quick/Comparisons
The GeForce RTX 4070 Ti outperforms the GeForce GTX 980 Ti by a significant margin of 47 points, thanks to its superior architecture and 12GB GDDR6X memory.
The GeForce RTX 4070 Ti takes the lead in content creation with a 39-point advantage, driven by its efficient Ada Lovelace architecture and generous 12GB VRAM.
The GeForce RTX 4070 Ti excels in AI/ML tasks with a 44-point lead, leveraging its dedicated tensor cores and advanced architecture for improved performance.
19 | Gaming | 66 |
27 | Workstation | 66 |
13 | AI/ML |
Compare with something else?
Compare GeForce GTX 980 with another →57 |
1 | Energy Efficiency | 44 |
21 | Quick Comparison Final Score | 64 |
| GM200 | Graphics Processor | AD104 |
| 2816 | Cores | 7680 |
| 176 / 96 | TMUs / ROPs | 240 / 80 |
| 536.2 | Blender GPU | 6639.7 |
| 52 | Forza Horizon 5 | 141 |
| 64 | The Witcher 3 | 182 |
| 95 | Counter-Strike 2 | 216 |
| 63 | Far Cry 6 | 148 |
| 37 | Hogwarts Legacy | 114 |
| 47 | Call of Duty | 171 |
| 33 | Ghost of Tsushima | 106 |
| 44 | Cyberpunk 2077 | 142 |
| 69 | Tomb Raider | 245 |
| 56 | Average FPS | 163 |
| 71 | 1080p High | 194 |
| 56 | 1080p Ultra | 163 |
| 39 | 1440p Ultra | 128 |
| 20 | 4K Ultra | 75 |
| Low | Margin of Error | Low |
| 49950 | GB6 Compute Score | 219163 |
| 134.8 img/sec | Background Blur | 371.2 img/sec |
| 59.8 img/sec | Face Detection | 235 img/sec |
| 2.37 Gpixels/sec | Horizon Detection | 7.51 Gpixels/sec |
| 3.79 Gpixels/sec | Edge Detection | 10.6 Gpixels/sec |
| 3.46 Gpixels/sec | Gaussian Blur | 12.4 Gpixels/sec |
| 0.22 Gpixels/sec | Feature Matching | 2.03 Gpixels/sec |
| 191.9 Gpixels/sec | Stereo Matching | 1120 Gpixels/sec |
| 4183.6 FPS | Particle Physics | 30420.2 FPS |
| OpenCL | API | OpenCL |
| 13642 | G3D Mark Score | 31568 |
| 850 | G2D Mark | 1209 |
| 106 FPS | DirectX 11 | 288 FPS |
| 55 FPS | DirectX 12 | 116 FPS |
| 6100 Ops/s | GPU Compute | 18221 Ops/s |
| 7711 | GB6 ML Single Precision | 32405 |
| 7464 | GB6 ML Half Precision | 48989 |
| 5338 | GB6 ML Quantized | 24894 |
| 3173 | Image Classification (SP) | 12496 |
| 4916 | Image Segmentation (HP) | 33378 |
| 7407 | Image Super Resolution (Q) | 37180 |
| 8201 | Face Detection (HP) | 62919 |
| 27221 | Pose Estimation (Q) | 171125 |
| 1675 | Text Classification (SP) | 3630 |
| 2607 | Machine Translation (HP) | 6420 |
| 4123 | Object Detection (SP) | 15804 |
| 11660 | Depth Estimation (Q) | 58647 |
| 55570 | Style Transfer (SP) | 331379 |
| ONNX | Framework | ONNX |
| DirectML | Backend | DirectML |
| 103 GPixel/s | Pixel Fill Rate | 209 GPixel/s |
| 189 GTexel/s | Texture Fill Rate | 626 GTexel/s |
| 6.1 TFLOPS | FLOPS (FP32) | 40.1 TFLOPS |
| GDDR5 | Memory Type | GDDR6X |
| 6 GB | Memory Size | 12 GB |
| 1753 MHz | Memory Clock | 1313 MHz |
| 7000 Mbps | Effective Memory Speed | 21000 Mbps |
| 384-bit | Bus | 192-bit |
| No | ECC | No |
| 336.6 GB/s | Memory Bandwidth | 504 GB/s |
| PCIe 3.0 x16 | Interface | PCIe 4.0 x16 |
| 250 W | TGP | 285 W |
| TSMC | Manufacturing | TSMC |
| 28 nm | Fabrication Process | 5 nm |
| 601 mm² | Die Size | 294 mm² |
| 8 billion | Transistor Count | 35 billion |
| 13.31 MTr/mm² | Transistor Density | 119.05 MTr/mm² |
| 12 | DirectX | 12 |
| 1.3 | Vulkan | 1.3 |
| 4.6 | OpenGL | 4.6 |
| 3 | OpenCL | 3 |
| 5.2 | CUDA | 8.9 |
| No | Ray Tracing | Yes |
| No | DLSS | DLSS 3 |
| 1.2 | DisplayPort | 1.4a |
| Nvidia | Vendor | Nvidia |
| Discrete | Build | Discrete |
| June 2, 2015 | Released | January 3, 2023 |
| Desktop | Case | Desktop |
| Gaming | Purpose | Gaming |
| High-end | Segment | Mid-range |
| Maxwell 2.0 | Architecture | Ada Lovelace |
| GM200 | GPU Codename | AD104 |
| — | Recommended CPU | -Intel Core i7 14700Kor above |
| 5945 | Steel Nomad Lite Score | 23618 |
| 5855 | Time Spy | 22751 |
| — | Solar Bay | 108422 |
| 3250 | Port Royal | 13994 |
| 19504 | Fire Strike | 53484 |
| 11830 | Wild Life Extreme | 45397 |
| 70105 | Night Raid | 171258 |
| 1000 MHz | Base Clock | 2310 MHz |
| 1076 MHz | Boost Clock | 2610 MHz |
| 2816 | Shading Units | 7680 |
| 176 | Texture Mapping Units (TMUs) | 240 |
| 96 | Render Output Units (ROPs) | 80 |
| 22 | Compute Units (Pipelines) | 60 |
| No | Tensor Cores | 240 |
| No | Ray-tracing Cores | 60 |
| 48KB per cluster | L1 Cache | 128KB per cluster |
| 3MB shared | L2 Cache | 48MB shared |
| 2 IPC | Instructions Per Cycle | 2 IPC |