H200 vs Apple M1 Ultra for local AI
Both fit the same models we track, up to gpt-oss 120B. The H200 generates about 6.0× as fast on any model both can run, because it has 4800 GB/s of memory bandwidth to the Apple M1 Ultra's 800 GB/s.
Memory and bandwidth
Memory decides which models fit. Bandwidth decides how fast they generate, because every token reads the model's active weights once.
| H200 | Apple M1 Ultra | |
|---|---|---|
| Memory | 141 GB VRAM | 64 or 128 GB unified memory |
| Usable for models | All 141 GB | About 96 GB of 128 GB |
| Bandwidth | 4800 GB/s | 800 GB/s |
| Biggest model that fits | gpt-oss 120B | gpt-oss 120B |
| Best everyday pick | gpt-oss 120B, ~580 tok/s | gpt-oss 120B, ~97 tok/s |
Every model, side by side
Estimated generation speed at the largest memory size each was sold in (141 GB and 128 GB).
| Model | H200 | Apple M1 Ultra |
|---|---|---|
| Qwen3 4B About 4 GB | ~1060 tok/s Runs great | ~180 tok/s Runs great |
| Qwen3 8B About 7 GB | ~510 tok/s Runs great | ~85 tok/s Runs great |
| Gemma 3 12B About 10 GB | ~330 tok/s Runs great | ~54 tok/s Runs great |
| Qwen3 14B About 11 GB | ~280 tok/s Runs great | ~47 tok/s Runs great |
| gpt-oss 20B About 16 GB | ~840 tok/s Runs great | ~140 tok/s Runs great |
| Gemma 3 27B About 19 GB | ~160 tok/s Runs great | ~26 tok/s Usable |
| Qwen3 30B-A3B About 21 GB | ~840 tok/s Runs great | ~140 tok/s Runs great |
| Qwen3 32B About 22 GB | ~130 tok/s Runs great | ~22 tok/s Usable |
| Llama 3.3 70B About 45 GB | ~61 tok/s Runs great | ~10 tok/s Usable |
| gpt-oss 120B About 67 GB | ~580 tok/s Runs great | ~97 tok/s Runs great |
| Qwen3 235B-A22B About 144 GB | Needs 144 GB | Needs 144 GB |
| DeepSeek R1 671B About 406 GB | Needs 406 GB | Needs 406 GB |
The H200, for speed. They fit the same models. Price matters as much as either number: check used prices on the builds page, where every parts list has a measured or estimated speed.
Both columns use the same formula: memory bandwidth divided by the weights read per token, calibrated on our own rack, with Ollama's default 4-bit sizes. Real results are usually within about a third. Measured either one? Send us the number.