RTX 5090 vs Apple M3 Ultra for local AI
The Apple M3 Ultra fits more: up to Qwen3 235B-A22B, against Qwen3 32B on the RTX 5090. The RTX 5090 generates about 2.2× as fast on any model both can run, because it has 1792 GB/s of memory bandwidth to the Apple M3 Ultra's 819 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.
| RTX 5090 | Apple M3 Ultra | |
|---|---|---|
| Memory | 32 GB VRAM | 96 or 256 or 512 GB unified memory |
| Usable for models | All 32 GB | About 384 GB of 512 GB |
| Bandwidth | 1792 GB/s | 819 GB/s |
| Biggest model that fits | Qwen3 32B | Qwen3 235B-A22B |
| Best everyday pick | Qwen3 32B, ~49 tok/s | Qwen3 235B-A22B, ~22 tok/s |
Every model, side by side
Estimated generation speed at the largest memory size each was sold in (32 GB and 512 GB).
| Model | RTX 5090 | Apple M3 Ultra |
|---|---|---|
| Qwen3 4B About 4 GB | ~390 tok/s Runs great | ~180 tok/s Runs great |
| Qwen3 8B About 7 GB | ~190 tok/s Runs great | ~87 tok/s Runs great |
| Gemma 3 12B About 10 GB | ~120 tok/s Runs great | ~56 tok/s Runs great |
| Qwen3 14B About 11 GB | ~110 tok/s Runs great | ~48 tok/s Runs great |
| gpt-oss 20B About 16 GB | ~310 tok/s Runs great | ~140 tok/s Runs great |
| Gemma 3 27B About 19 GB | ~58 tok/s Runs great | ~26 tok/s Usable |
| Qwen3 30B-A3B About 21 GB | ~310 tok/s Runs great | ~140 tok/s Runs great |
| Qwen3 32B About 22 GB | ~49 tok/s Runs great | ~23 tok/s Usable |
| Llama 3.3 70B About 45 GB | Needs 45 GB | ~10 tok/s Usable |
| gpt-oss 120B About 67 GB | Needs 67 GB | ~99 tok/s Runs great |
| Qwen3 235B-A22B About 144 GB | Needs 144 GB | ~22 tok/s Usable |
| DeepSeek R1 671B About 406 GB | Needs 406 GB | Needs 406 GB |
It depends. The Apple M3 Ultra fits bigger models; the RTX 5090 runs the models both fit faster. 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.
Other matchups
- RTX 5090 vs RTX 409024 GB · 1008 GB/s
- RTX 5090 vs RTX 309024 GB · 936 GB/s
- RTX 5090 vs RTX Pro 6000 Blackwell96 GB · 1792 GB/s
- RTX 5090 vs Tesla V10016 / 32 GB · 900 GB/s
- Apple M3 Ultra vs Apple M2 Ultra64 / 128 / 192 GB · 800 GB/s
- Apple M3 Ultra vs Apple M1 Ultra64 / 128 GB · 800 GB/s
- Apple M3 Ultra vs L40S48 GB · 864 GB/s