M3 Ultra 192GB
Apple Silicon
192GB VRAM · 800 GB/s bandwidth · 65.5 FP16 TFLOPS · 120W TDP · $6,999 street price
The M3 Ultra 192GB has 192GB of VRAM with 800 GB/s memory bandwidth and 65.5 TFLOPS FP16 compute. At Q4 quantization, it can comfortably run Gemma 3 4B (172 tok/s), Qwen 2.5 7B (90 tok/s), Llama 3.1 8B (85 tok/s). Models larger than ~326B parameters won't fit even at Q4. Electricity cost is approximately $13/month at 120W TDP.
What LLMs can you run?
| Model | Params | Q4 Weight | Fit | Decode |
|---|---|---|---|---|
| Gemma 3 4B | 4.0B | 2 GB | comfortable | 172 tok/s |
| Qwen 2.5 7B | 7.6B | 4 GB | comfortable | 90 tok/s |
| Llama 3.1 8B | 8.0B | 4 GB | comfortable | 85 tok/s |
| Mistral Small 24B | 24.0B | 12 GB | tight | 28 tok/s |
| Gemma 3 27B | 27.4B | 14 GB | tight | 25 tok/s |
| Qwen 2.5 Coder 32B | 32.5B | 16 GB | tight | 21 tok/s |
| Llama 3.3 70B | 70.6B | 35 GB | tight | 9 tok/s |
| Qwen 2.5 72B | 72.7B | 36 GB | tight | 9 tok/s |
| Llama 3.1 405B | 405B | 202 GB | won't fit | |
| DeepSeek R1 671B | 671B | 336 GB | won't fit |
Similar GPUs
| GPU | VRAM | BW | TFLOPS | TDP |
|---|---|---|---|---|
| M2 Ultra 192GB | 192GB | 800 GB/s | 54.4 | 120W |
| Radeon Instinct MI300A | 192GB | 10300 GB/s | 653.7 | 750W |
| Radeon Instinct MI300X | 192GB | 10300 GB/s | 653.7 | 750W |
| Radeon Instinct MI308X | 192GB | 10300 GB/s | 653.7 | 750W |
| B300 | 144GB | 4099 GB/s | 1231.8 | 1400W |
Compare with another GPU
Select another GPU to compare specs and model performance side by side.