M5 Pro (64GB) holds 345 of the models in our catalogue and is, in practice, a IQ4_XS card — the largest it takes is DeepSeek R1 Distill Llama 70B at IQ4_XS.
Neighbours in memory rather than in price: memory decides whether a card can do the job at all, so two cards of the same size at different prices is the comparison you are making. Ordered by memory, then bandwidth — not by the value column, which is worked out from bandwidth and price alone and therefore rewards a cheap card whatever its software stack does to that bandwidth in practice. Read it as one input, not as a ranking.
Apple M5 Pro (64GB) — 64 GB unified, 48 GB usable.
- BRAND
- Apple
- UNIFIED MEMORY
- 64 GB
- USABLE BY A MODEL
- 48 GB
- BANDWIDTH
- 200 GB/s
- FP16 COMPUTE
- 9 TFLOPS
- TDP
- 30 W
- ARCHITECTURE
- M5 Pro
With 48 GB of its 64 GB reaching a model and 200 GB/s bandwidth, this machine handles models up to 65.2B parameters.
Speed ≈ bandwidth / model_size × efficiency. A 7B model at Q4 runs at ~23 tok/s.
| MODEL | SIZE | VRAM Q4 | TOK/S | AVG |
|---|---|---|---|---|
| LLaMA 1 65B | 65.2B | 40.3 GB | 3 | 42.6 |
| Jamba 2 Mini | 52B | 32.3 GB | 15 | 9.2 |
| Jamba 1.5 Mini 52B | 51.6B | 32.0 GB | 15 | 24.2 |
| Kimi-Linear-48B-A3B | 48B | 29.8 GB | 59 | 26.6 |
| Nemotron-H 47B | 47B | 29.2 GB | 4 | 84.6 |
| Mixtral-8x7B | 46.7B | 29.0 GB | 14 | 18.8 |
| Nous-Hermes-2-Mixtral-8x7B-DPO | 46.7B | 29.0 GB | 14 | 27.4 |
| Dolphin 2.6 Mixtral 8x7B | 46.7B | 29.0 GB | 14 | 23.8 |
| Phi-3.5 MoE 42B | 41.9B | 26.1 GB | 27 | 56.7 |
| Falcon 40B | 40B | 24.9 GB | 4 | 20.9 |
| InternVL3 38B | 38B | 23.7 GB | 5 | 78.9 |
| Seed-OSS 36B Instruct | 36B | 22.5 GB | 5 | 54.4 |
| c4ai-command-r-v01 35B | 35B | 21.9 GB | 5 | 27.5 |
| Qwen 3.5 35B A3B | 35B | 21.9 GB | 59 | 53.3 |
| Qwen 3.6 35B A3B | 35B | 21.9 GB | 59 | 53.9 |
| Nous Capybara 34B | 34.4B | 21.5 GB | 5 | 42.0 |
| Yi-1.5 34B | 34.4B | 21.5 GB | 5 | 45.3 |
| Falcon-H1 34B | 34B | 21.3 GB | 5 | 66.1 |
| CodeLlama 34B | 34B | 21.3 GB | 5 | 25.4 |
| Nous Hermes 2 34B | 34B | 21.3 GB | 5 | 47.0 |