Memory at each precision
| Version | Weights | With context | Smallest Mac |
|---|---|---|---|
| 8-bit | 78.6 GB | 84.9 GB | 128 GB |
| 6-bit | 60.1 GB | 64.9 GB | 96 GB |
| 4-bit | 41.6 GB | 44.9 GB | 64 GB |
| 3-bit lower quality | 32.4 GB | 35 GB | 48 GB |
| 2-bit lower quality | 23.1 GB | 24.9 GB | 36 GB |
On every Apple chip
The smallest memory size of each chip that runs Nemotron 3.5 Bolt 74B A9B, the best version that fits it, and the writing speed you can expect.
| Chip | Needs | Version | Tokens/s |
|---|---|---|---|
| M6 | 32 GB | 2-bit Only just | 27–49 |
| M5 Ultra | 96 GB | 6-bit Runs well | 74–132 |
| M5 Max | 64 GB | 4-bit Runs well | 54–96 |
| M5 Pro | 64 GB | 4-bit Runs well | 27–48 |
| M5 | 32 GB | 2-bit Only just | 25–44 |
| M4 Max | 64 GB | 4-bit Runs well | 48–86 |
| M4 Pro | 64 GB | 4-bit Runs well | 24–43 |
| M4 | 32 GB | 2-bit Only just | 19–34 |
| M3 Ultra | 96 GB | 6-bit Runs well | 50–90 |
| M3 Max | 64 GB | 4-bit Runs well | 35–63 |
| M3 Pro | 36 GB | 2-bit Runs well | 24–43 |
| M3 | no size fits | Too big | – |
| M2 Ultra | 64 GB | 4-bit Runs well | 71–125 |
| M2 Max | 64 GB | 4-bit Runs well | 35–63 |
| M2 Pro | 32 GB | 2-bit Only just | 32–57 |
| M2 | no size fits | Too big | – |
| M1 Ultra | 64 GB | 4-bit Runs well | 71–125 |
| M1 Max | 64 GB | 4-bit Runs well | 35–63 |
| M1 Pro | 32 GB | 2-bit Only just | 32–57 |
| M1 | no size fits | Too big | – |
How to run it
- The easy way: download MLX Harness, free and open source. Search for Nemotron 3.5 Bolt 74B A9B, download the version that fits your Mac, and chat. It also serves the model to your other apps over an OpenAI-compatible API.Download MLX HarnessFree · Apple Silicon · macOS 14.5 or later
- In Terminal with Apple’s MLX: There is no community MLX build yet, so make one from the official weights:
pip install mlx-lmmlx_lm.convert --hf-path nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16 -q
Step by step: download and run LLMs on a Mac · how to use MLX Harness · MLX vs Ollama vs llama.cpp
About Nemotron 3.5 Bolt 74B A9B
| Made by | NVIDIA |
|---|---|
| Parameters | 74B total · 9B active (mixture of experts) |
| Context | 256K tokens |
| Licence | OpenMDW-1.1 |
| Released | 2026-10 |
| Input | Text and images |
| Intelligence | 21.4 on the Artificial Analysis Intelligence Index |
| Weights | nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16 |
Questions
- How much memory does Nemotron 3.5 Bolt 74B A9B need?
- About 41.6 GB of weights at 4-bit, plus room for the context: 44.9 GB in all. On a Mac that means 64 GB of unified memory or more, since macOS lets the GPU use about 75% of it by default.
- How fast is Nemotron 3.5 Bolt 74B A9B on a Mac?
- Writing speed depends on memory bandwidth. The fastest Mac for it here is an M5 Ultra with 96 GB at about 74–132 tokens a second (6-bit). The smallest Mac that runs it, with 64 GB, manages about 71–125 on an M2 Ultra.
- Why does Nemotron 3.5 Bolt 74B A9B run faster than its size suggests?
- It is a mixture-of-experts model: it holds 74B parameters in memory but reads only 9B for each token, so it writes at the speed of a much smaller model.
Similar models
For your model card or README
Shows the smallest Mac that runs Nemotron 3.5 Bolt 74B A9B well and links here.
[](https://mlx.app/run/nemotron-3.5-bolt-74b-a9b)How it’s worked out
Memory: the model’s parameters at 8, 6, 4, 3 or 2 bits per weight plus MLX’s group scales (half a bit per weight), or its native format, with room for the context on top. By default macOS lets the GPU use about 75% of unified memory. Speed: memory bandwidth divided by the bytes read per token is the ceiling for one conversation; in llama.cpp’s public Apple Silicon benchmarks real runs reach about 45% to 80% of it, which is the range shown. These are estimates, not measurements: measure your own Mac on MLX Bench. Model figures from each official model card, chip figures from Apple’s Tech Specs. Checked 2026-10-19. Same data as mlx.app/compare.