Mac Studio M3 Ultra 256GB vs NVIDIA GeForce RTX 4090
For running models locally, priced in the UK. At 64k context the first runs 42 models fully in memory, the second 22.
In short
The Mac Studio M3 Ultra 256GB has more memory for models: 256 GB against 24 GB. At 64k context the Mac Studio M3 Ultra 256GB runs 42 of the models we track fully in memory and the NVIDIA GeForce RTX 4090 22; only the Mac Studio M3 Ultra 256GB holds DeepSeek V4 Flash. On the models both run fully in memory, the NVIDIA GeForce RTX 4090 is about 1.3× as fast as the Mac Studio M3 Ultra 256GB (estimated, median across models). In the UK the Mac Studio M3 Ultra 256GB costs £8,499 at eBay (used) and the NVIDIA GeForce RTX 4090 £6,015 at Amazon: the NVIDIA GeForce RTX 4090 is £2,484 cheaper.
Updated . Speeds are estimates; how they are worked out.
Specifications
| Spec | Mac Studio M3 Ultra 256GB | NVIDIA GeForce RTX 4090 |
|---|---|---|
| Chip | Apple M3 Ultra (80-core GPU) | – |
| Memory bus | 1024-bit | 384-bit |
| Memory bandwidth | 819 GB/s | 1,008 GB/s |
| FP16 compute | 28 TFLOPS | 165 TFLOPS |
| Software | Metal | CUDA |
| Unified memory | 256 GB | – |
| GPU can use by default | 192 GB | – |
| GPU can use, tuned | 248 GB | – |
| Operating system | macos | – |
| Processor | M3 Ultra 32-core | – |
| Storage | 2000 GB | – |
| GPU | – | GeForce RTX 4090 |
| Memory | – | 24 GB GDDR6X |
| Board power | – | 450 W |
| PCIe generation | – | Gen 4 |
| PCIe lanes | – | x16 |
What each runs
Estimated decode speed with the most precise quant that fits each one est.
| Model | Mac Studio M3 Ultra 256GB | NVIDIA GeForce RTX 4090 |
|---|---|---|
| DeepSeek V4 Flash | 19tok/sUD-Q4_K_XL | won't run |
| GLM 5.3 Flash | 19tok/sUD-Q4_K_XL | won't run |
| Qwen 3.8 Flash Next | 5tok/sQ8_0 | won't run |
| Qwen 3.8 27B | 16tok/sQ8_0 | 30tok/sUD-Q4_K_M |
| MiniMax M2.7 | 12tok/sUD-Q6_K | won't run |
| Qwen 3.5 122B-A10B | 26tok/sQ8_0 | offload |
| Qwen 3.6 27B | 16tok/sQ8_0 | 26tok/sQ5_K_M |
| gpt-oss 120B | 58tok/sMXFP4 | won't run |
| Qwen3 Coder Next | 55tok/sQ8_0 | offload |
| Nemotron 3 Super 120B-A12B | 24tok/sQ8_0 | won't run |
| Gemma 4 31B | 14tok/sQ8_0 | offload |
| Mistral Small 4 119B | 36tok/sQ8_0 | offload |
| Qwen 3.6 35B-A3B | 71tok/sQ8_0 | 115tok/sUD-Q4_K_M |
| Qwen 3.5 27B | 16tok/sQ8_0 | 30tok/sQ4_K_M |
| Gemma 4 26B-A4B | 56tok/sQ8_0 | 78tok/sUD-Q5_K_M |
| GLM 4.5 Air | 11tok/sQ8_0 | won't run |
| Qwen 3.5 35B-A3B | 71tok/sQ8_0 | 116tok/sQ4_K_M |
| Qwen3 Next 80B-A3B | 66tok/sQ8_0 | offload |
| GLM 4.7 Flash | 36tok/sQ8_0 | 57tok/sQ4_K_M |
| Devstral Small 2 24B | 15tok/sQ8_0 | offload |
| Nemotron 3.5 Lightning 30B-A3B | 85tok/sUD-Q6_K_XL | offload |
| Kimi Linear 48B-A3B | 85tok/sQ8_0 | offload |
| gpt-oss 20B | 79tok/sMXFP4 | 93tok/sMXFP4 |
| Qwen3 Coder 30B-A3B | 33tok/sQ8_0 | offload |
| Qwen3 32B | 10tok/sQ8_0 | offload |
| Gemma 4 12B | 38tok/sQ8_0 | 44tok/sQ8_0 |
| Qwen 3.5 9B | 45tok/sQ8_0 | 53tok/sQ8_0 |
| Gemma 3 27B | 15tok/sQ8_0 | 28tok/sQ4_K_M |
| Mistral Small 3.2 24B | 15tok/sQ8_0 | offload |
| DeepSeek R1 Distill Qwen 32B | 10tok/sQ8_0 | offload |
| Llama 3.3 70B | 6tok/sQ8_0 | won't run |
| Ministral 3 14B | 21tok/sQ8_0 | 29tok/sQ6_K |
| Qwen3 14B | 20tok/sQ8_0 | 27tok/sQ6_K |
| DeepSeek R1 0528 Qwen3 8B | 29tok/sQ8_0 | 34tok/sQ8_0 |
| Qwen3 8B | 29tok/sQ8_0 | 34tok/sQ8_0 |
| Phi-4 14B | 18tok/sQ8_0 | 26tok/sQ5_K_M |
| Qwen 3.5 4B | 80tok/sQ8_0 | 94tok/sQ8_0 |
| Ministral 3 8B | 29tok/sQ8_0 | 34tok/sQ8_0 |
| Granite 4.2 8B | 27tok/sQ8_0 | 31tok/sQ8_0 |
| Gemma 4 E4B | 53tok/sQ8_0 | 62tok/sQ8_0 |
| LFM2.5 8B-A1B | 75tok/sQ8_0 | 88tok/sQ8_0 |
| Llama 3.1 8B | 35tok/sQ6_K | 41tok/sQ6_K |
Questions
- Which is better for local AI, the Mac Studio M3 Ultra 256GB or the NVIDIA GeForce RTX 4090?
- Choose the Mac Studio M3 Ultra 256GB to run bigger models. Choose the NVIDIA GeForce RTX 4090 for speed. Choose the NVIDIA GeForce RTX 4090 to spend less.
- Which can run bigger models, the Mac Studio M3 Ultra 256GB or the NVIDIA GeForce RTX 4090?
- At 64k context the Mac Studio M3 Ultra 256GB runs 42 of the models we track fully in memory and the NVIDIA GeForce RTX 4090 22; only the Mac Studio M3 Ultra 256GB holds DeepSeek V4 Flash.
- Which is faster for local AI, the Mac Studio M3 Ultra 256GB or the NVIDIA GeForce RTX 4090?
- On the models both run fully in memory, the NVIDIA GeForce RTX 4090 is about 1.3× as fast as the Mac Studio M3 Ultra 256GB (estimated, median across models).
- What is the price difference between the Mac Studio M3 Ultra 256GB and the NVIDIA GeForce RTX 4090?
- In the UK the Mac Studio M3 Ultra 256GB costs £8,499 at eBay (used) and the NVIDIA GeForce RTX 4090 £6,015 at Amazon: the NVIDIA GeForce RTX 4090 is £2,484 cheaper.
More comparisons
- Mac Studio M3 Ultra 256GB vs Mac Studio M3 Ultra 512GB
- Mac Studio M3 Ultra 256GB vs Mac Studio M3 Ultra 96GB
- Mac Studio M3 Ultra 256GB vs Mac Studio M4 Max 128GB
- Mac Studio M3 Ultra 96GB vs NVIDIA GeForce RTX 4090
- NVIDIA GeForce RTX 3090 Ti vs NVIDIA GeForce RTX 4090
- NVIDIA GeForce RTX 4090 vs NVIDIA GeForce RTX 5090
- NVIDIA GeForce RTX 4090 vs AMD Radeon RX 7900 XTX