Quick/Comparisons
The Apple M3 Ultra GPU (80-core) takes the lead in gaming performance, with a 17-point gap over the Apple M3 Max GPU (40-core). Its higher score reflects its superior handling of demanding games.
The Apple M3 Ultra GPU (80-core) outperforms the Apple M3 Max GPU (40-core) in content creation, with a 17-point gap in workstation tasks. Its higher score indicates better performance in tasks like video editing and 3D modeling.
The Apple M3 Max GPU (40-core) and Apple M3 Ultra GPU (80-core) are nearly tied in AI/ML performance, with a 2-point gap. Both GPUs offer competitive performance in machine learning tasks.
33 | Gaming | 50 |
42 | Workstation | 59 |
39 | AI/ML |
Compare with something else?
Compare Apple M3 Max with another →41 |
82 | Energy Efficiency | 61 |
44 | Quick Comparison Final Score | 52 |
| — | Graphics Processor | Custom |
| 5120 | Cores | 10240 |
| 320 / 160 | TMUs / ROPs | 640 / 320 |
| 3797.82 | Blender GPU | 7360.18 |
| — | Forza Horizon 5 | — |
| — | The Witcher 3 | — |
| — | Counter-Strike 2 | — |
| — | Far Cry 6 | — |
| — | Hogwarts Legacy | — |
| — | Call of Duty | — |
| — | Ghost of Tsushima | — |
| — | Cyberpunk 2077 | — |
| — | Tomb Raider | — |
| — | Average FPS | — |
| — | 1080p High | — |
| — | 1080p Ultra | — |
| — | 1440p Ultra | — |
| — | 4K Ultra | — |
| — | Margin of Error | — |
| 94528 | GB6 Compute Score | 147719 |
| 157.6 img/sec | Background Blur | 204.5 img/sec |
| 100.9 img/sec | Face Detection | 126 img/sec |
| 3.99 Gpixels/sec | Horizon Detection | 6.53 Gpixels/sec |
| 6.04 Gpixels/sec | Edge Detection | 10.7 Gpixels/sec |
| 5 Gpixels/sec | Gaussian Blur | 9.5 Gpixels/sec |
| 0.84 Gpixels/sec | Feature Matching | 1.05 Gpixels/sec |
| 363.8 Gpixels/sec | Stereo Matching | 560.3 Gpixels/sec |
| 12226.9 FPS | Particle Physics | 25267.8 FPS |
| OpenCL | API | OpenCL |
| — | G3D Mark Score | — |
| — | G2D Mark | — |
| — | DirectX 11 | — |
| — | DirectX 12 | — |
| — | GPU Compute | — |
| 20068 | GB6 ML Single Precision | 21356 |
| 22390 | GB6 ML Half Precision | 23530 |
| 20681 | GB6 ML Quantized | 21706 |
| 8685 | Image Classification (SP) | 8094 |
| 20739 | Image Segmentation (HP) | 24595 |
| 22775 | Image Super Resolution (Q) | 25190 |
| 37676 | Face Detection (HP) | 38337 |
| 91609 | Pose Estimation (Q) | 111021 |
| 2909 | Text Classification (SP) | 2786 |
| 5274 | Machine Translation (HP) | 4339 |
| 7480 | Object Detection (SP) | 7668 |
| 38212 | Depth Estimation (Q) | 37488 |
| 190681 | Style Transfer (SP) | 244483 |
| Core ML | Framework | Core ML |
| GPU | Backend | GPU |
| 221 GPixel/s | Pixel Fill Rate | 442 GPixel/s |
| 442 GTexel/s | Texture Fill Rate | 883 GTexel/s |
| 14.1 TFLOPS | FLOPS (FP32) | 28.3 TFLOPS |
| System Shared | Memory Type | System Shared |
| — | Memory Size | — |
| 6400 MHz | Memory Clock | 6400 MHz |
| 12800 Mbps | Effective Memory Speed | — |
| 512-bit | Bus | 1024-bit |
| No | ECC | No |
| 409.6 GB/s | Memory Bandwidth | 819.3 GB/s |
| Custom | Interface | Custom |
| 60 W | TGP | 140 W |
| TSMC | Manufacturing | TSMC |
| 3 nm | Fabrication Process | 3 nm |
| — | Die Size | — |
| 56 billion | Transistor Count | 184 billion |
| — | Transistor Density | — |
| — | DirectX | — |
| — | Vulkan | — |
| — | OpenGL | — |
| — | OpenCL | — |
| — | CUDA | — |
| Yes | Ray Tracing | Yes |
| No | DLSS | No |
| — | DisplayPort | — |
| Apple | Vendor | Apple |
| Integrated | Build | Integrated |
| October 31, 2023 | Released | March 5, 2025 |
| Laptop | Case | Desktop |
| Professional | Purpose | Professional |
| High-end | Segment | High-end |
| Apple M GPU | Architecture | Apple M GPU |
| — | GPU Codename | Custom |
| -Apple M3 Maxor above | Recommended CPU | -Apple M3 Ultraor above |
| 12391 | Steel Nomad Lite Score | 19491 |
| — | Time Spy | — |
| 49656 | Solar Bay | 73690 |
| — | Port Royal | — |
| — | Fire Strike | — |
| 30787 | Wild Life Extreme | 44759 |
| — | Night Raid | — |
| 500 MHz | Base Clock | — |
| 1380 MHz | Boost Clock | 1380 MHz |
| 5120 | Shading Units | 10240 |
| 320 | Texture Mapping Units (TMUs) | 640 |
| 160 | Render Output Units (ROPs) | 320 |
| 640 | Compute Units (Pipelines) | 1280 |
| — | Tensor Cores | — |
| — | Ray-tracing Cores | — |
| — | L1 Cache | — |
| — | L2 Cache | — |
| 2 IPC | Instructions Per Cycle | 2 IPC |