Quick/Comparisons
50 | Gaming | 5 |
59 | Workstation | 10 |
41 | AI/ML |
Compare with something else?
Compare Apple M3 Ultra with another →4 |
61 | Energy Efficiency | 67 |
52 | Quick Comparison Final Score | 17 |
| Custom | Graphics Processor | GP107 |
| 10240 | Cores | 640 |
| 640 / 320 | TMUs / ROPs | 32 / 16 |
| 7360.18 | Blender GPU | 96.75 |
| — | Forza Horizon 5 | - |
| — | The Witcher 3 | - |
| — | Counter-Strike 2 | 16 |
| — | Far Cry 6 | 15 |
| — | Hogwarts Legacy | - |
| — | Call of Duty | 21 |
| — | Ghost of Tsushima | - |
| — | Cyberpunk 2077 | — |
| — | Tomb Raider | 9 |
| — | Average FPS | 10 |
| — | 1080p High | 15 |
| — | 1080p Ultra | 10 |
| — | 1440p Ultra | - |
| — | 4K Ultra | - |
| — | Margin of Error | Medium |
| 147719 | GB6 Compute Score | 15572 |
| 204.5 img/sec | Background Blur | 40 img/sec |
| 126 img/sec | Face Detection | 18.9 img/sec |
| 6.53 Gpixels/sec | Horizon Detection | 0.59 Gpixels/sec |
| 10.7 Gpixels/sec | Edge Detection | 0.88 Gpixels/sec |
| 9.5 Gpixels/sec | Gaussian Blur | 0.97 Gpixels/sec |
| 1.05 Gpixels/sec | Feature Matching | 0.15 Gpixels/sec |
| 560.3 Gpixels/sec | Stereo Matching | 45.1 Gpixels/sec |
| 25267.8 FPS | Particle Physics | 1531.6 FPS |
| OpenCL | API | OpenCL |
| — | G3D Mark Score | 2820 |
| — | G2D Mark | 238 |
| — | DirectX 11 | 20 FPS |
| — | DirectX 12 | 15 FPS |
| — | GPU Compute | 1251 Ops/s |
| 21356 | GB6 ML Single Precision | 2424 |
| 23530 | GB6 ML Half Precision | 2615 |
| 21706 | GB6 ML Quantized | 1827 |
| 8094 | Image Classification (SP) | 949 |
| 24595 | Image Segmentation (HP) | 1407 |
| 25190 | Image Super Resolution (Q) | 2693 |
| 38337 | Face Detection (HP) | 2453 |
| 111021 | Pose Estimation (Q) | 8112 |
| 2786 | Text Classification (SP) | 1170 |
| 4339 | Machine Translation (HP) | 1183 |
| 7668 | Object Detection (SP) | 768 |
| 37488 | Depth Estimation (Q) | 3878 |
| 244483 | Style Transfer (SP) | 20820 |
| Core ML | Framework | ONNX |
| GPU | Backend | DirectML |
| 442 GPixel/s | Pixel Fill Rate | 15-23 GPixel/s |
| 883 GTexel/s | Texture Fill Rate | 30-47 GTexel/s |
| 28.3 TFLOPS | FLOPS (FP32) | 1.2-1.9 TFLOPS |
| System Shared | Memory Type | GDDR5 |
| — | Memory Size | 2 GB |
| 6400 MHz | Memory Clock | 1752 MHz |
| — | Effective Memory Speed | 7000 Mbps |
| 1024-bit | Bus | 64-bit |
| No | ECC | — |
| 819.3 GB/s | Memory Bandwidth | 56 GB/s |
| Custom | Interface | PCIe 3.0 x4 |
| 140 W | TGP | 15-30 W (configurable) |
| TSMC | Manufacturing | Samsung |
| 3 nm | Fabrication Process | 14 nm |
| — | Die Size | 132 mm² |
| 184 billion | Transistor Count | 3.3 billion |
| — | Transistor Density | 25 MTr/mm² |
| — | DirectX | 12 |
| — | Vulkan | 1.2 |
| — | OpenGL | 4.6 |
| — | OpenCL | 1.2 |
| — | CUDA | 6.1 |
| Yes | Ray Tracing | No |
| No | DLSS | No |
| — | DisplayPort | 1.4 |
| Apple | Vendor | Nvidia |
| Integrated | Build | Discrete |
| March 5, 2025 | Released | February 10, 2020 |
| Desktop | Case | Laptop |
| Professional | Purpose | Gaming |
| High-end | Segment | Entry-level |
| Apple M GPU | Architecture | Pascal |
| Custom | GPU Codename | GP107 |
| -Apple M3 Ultraor above | Recommended CPU | -Intel Core i7 1165G7or above |
| 19491 | Steel Nomad Lite Score | 967 |
| — | Time Spy | 1235 |
| 73690 | Solar Bay | — |
| — | Port Royal | — |
| — | Fire Strike | 4234 |
| 44759 | Wild Life Extreme | 2706 |
| — | Night Raid | 17008 |
| — | Base Clock | 747-1354 MHz (configurable) |
| 1380 MHz | Boost Clock | 936-1468 MHz (configurable) |
| 10240 | Shading Units | 640 |
| 640 | Texture Mapping Units (TMUs) | 32 |
| 320 | Render Output Units (ROPs) | 16 |
| 1280 | Compute Units (Pipelines) | 5 |
| — | Tensor Cores | No |
| — | Ray-tracing Cores | No |
| — | L1 Cache | 48KB per cluster |
| — | L2 Cache | 512KB shared |
| 2 IPC | Instructions Per Cycle | 2 IPC |