Quick/Comparisons
The GeForce RTX 5070 Ti's 76 score in gaming far surpasses the Radeon RX 590's 15, thanks to its Blackwell 2.0 architecture and 16GB GDDR7 memory.
The GeForce RTX 5070 Ti's 74 score in workstation tasks edges out the Radeon RX 590's 25, driven by its solid performance and large memory pool.
The GeForce RTX 5070 Ti's 66 score in AI/ML tasks outpaces the Radeon RX 590's 17, despite the Radeon's dedicated AI tensor cores, due to the GeForce's overall architecture and memory.
76 | Gaming | 15 |
74 | Workstation | 25 |
66 | AI/ML | 17 |
49 | Energy Efficiency | 28 |
74 | Quick Comparison Final Score | 19 |
| GB203-200-A1 | Graphics Processor | Polaris 30 |
| 8960 | Cores | 2304 |
| 280 / 96 | TMUs / ROPs | 144 / 32 |
| 7584.67 | Blender GPU | 487.92 |
| 166 | Forza Horizon 5 | 42 |
| 213 | The Witcher 3 | 47 |
| 219 | Counter-Strike 2 | 64 |
| 159 | Far Cry 6 | 50 |
| 135 | Hogwarts Legacy | 29 |
| 205 | Call of Duty | 42 |
| 121 | Ghost of Tsushima | 25 |
| 174 | Cyberpunk 2077 | 30 |
| 260 | Tomb Raider | 55 |
| 184 | Average FPS | 43 |
| 217 | 1080p High | 57 |
| 184 | 1080p Ultra | 43 |
| 148 | 1440p Ultra | 28 |
| 86 | 4K Ultra | 15 |
| Low | Margin of Error | Low |
| 240508 | GB6 Compute Score | 45783 |
| 297.9 img/sec | Background Blur | 69 img/sec |
| 218.6 img/sec | Face Detection | 38.2 img/sec |
| 10.1 Gpixels/sec | Horizon Detection | 1.75 Gpixels/sec |
| 13.9 Gpixels/sec | Edge Detection | 3.31 Gpixels/sec |
| 14.8 Gpixels/sec | Gaussian Blur | 3.31 Gpixels/sec |
| 2.03 Gpixels/sec | Feature Matching | 0.38 Gpixels/sec |
| 1290 Gpixels/sec | Stereo Matching | 180.5 Gpixels/sec |
| 35438.3 FPS | Particle Physics | 6238 FPS |
| OpenCL | API | OpenCL |
| 32425 | G3D Mark Score | 9293 |
| 1329 | G2D Mark | 790 |
| 304 FPS | DirectX 11 | 65 FPS |
| 126 FPS | DirectX 12 | 45 FPS |
| 18089 Ops/s | GPU Compute | 3764 Ops/s |
| 36504 | GB6 ML Single Precision | 9064 |
| 64459 | GB6 ML Half Precision | 9971 |
| 27312 | GB6 ML Quantized | 6293 |
| 14913 | Image Classification (SP) | 3307 |
| 45199 | Image Segmentation (HP) | 4895 |
| 39360 | Image Super Resolution (Q) | 9072 |
| 84188 | Face Detection (HP) | 12467 |
| 161400 | Pose Estimation (Q) | 43468 |
| 4010 | Text Classification (SP) | 1547 |
| 7973 | Machine Translation (HP) | 2504 |
| 18934 | Object Detection (SP) | 3746 |
| 67761 | Depth Estimation (Q) | 15075 |
| 335879 | Style Transfer (SP) | 96813 |
| ONNX | Framework | ONNX |
| DirectML | Backend | DirectML |
| 235 GPixel/s | Pixel Fill Rate | 49 GPixel/s |
| 687 GTexel/s | Texture Fill Rate | 222 GTexel/s |
| 43.9 TFLOPS | FLOPS (FP32) | 7.1 TFLOPS |
| GDDR7 | Memory Type | GDDR5 |
| 16 GB | Memory Size | 8 GB |
| 1750 MHz | Memory Clock | 2000 MHz |
| 28000 Mbps | Effective Memory Speed | 8000 Mbps |
| 256-bit | Bus | 256-bit |
| No | ECC | No |
| 896 GB/s | Memory Bandwidth | 256 GB/s |
| PCIe 5.0 x16 | Interface | PCIe 3.0 x16 |
| 300 W | TGP | 175 W |
| TSMC | Manufacturing | GlobalFoundries |
| 4 nm | Fabrication Process | 12 nm |
| 378 mm² | Die Size | 232 mm² |
| 45.6 billion | Transistor Count | 5 billion |
| 120.63 MTr/mm² | Transistor Density | 21.55 MTr/mm² |
| 12.2 | DirectX | 12 |
| 1.4 | Vulkan | 1.3 |
| 4.6 | OpenGL | 4.6 |
| 3 | OpenCL | 2.1 |
| 12 | CUDA | — |
| Yes | Ray Tracing | No |
| DLSS 4 | DLSS | No |
| 2.1b | DisplayPort | 1.4a |
| Nvidia | Vendor | Amd |
| Discrete | Build | Discrete |
| January 7, 2025 | Released | November 15, 2018 |
| Desktop | Case | Desktop |
| Gaming | Purpose | Gaming |
| Mid-range | Segment | High-end |
| Blackwell 2.0 | Architecture | GCN 4.0 |
| GB203-200-A1 | GPU Codename | Polaris 30 |
| -Intel Core Ultra 5 245Kor above | Recommended CPU | -Intel Core i5 8500or above |
| 29989 | Steel Nomad Lite Score | 4611 |
| 27815 | Time Spy | 4778 |
| 136964 | Solar Bay | — |
| 19613 | Port Royal | — |
| 67382 | Fire Strike | 16527 |
| 57734 | Wild Life Extreme | 9735 |
| 189734 | Night Raid | 62048 |
| 2295 MHz | Base Clock | 1469 MHz |
| 2452 MHz | Boost Clock | 1545 MHz |
| 8960 | Shading Units | 2304 |
| 280 | Texture Mapping Units (TMUs) | 144 |
| 96 | Render Output Units (ROPs) | 32 |
| 70 | Compute Units (Pipelines) | — |
| 280 | Tensor Cores | No |
| 70 | Ray-tracing Cores | No |
| 128KB per cluster | L1 Cache | 16KB per cluster |
| 48MB shared | L2 Cache | 2MB shared |
| 2 IPC | Instructions Per Cycle | 2 IPC |
Compare with something else?
Compare GeForce RTX 5070 with another →