Memory at each precision
| Version | Weights | With context | Smallest Mac |
|---|---|---|---|
| 8-bit | 4.2 GB | 5.7 GB | 8 GB |
| 6-bit | 3.2 GB | 4.7 GB | 8 GB |
| 4-bit | 2.2 GB | 3.7 GB | 8 GB |
| 3-bit lower quality | 1.7 GB | 3.2 GB | 8 GB |
| 2-bit lower quality | 1.2 GB | 2.7 GB | 8 GB |
On every Apple chip
The smallest memory size of each chip that runs Nemotron-Labs Dory 1.0 4B, the best version that fits it, and the writing speed you can expect.
| Chip | Needs | Version | Tokens/s |
|---|---|---|---|
| M6 | 16 GB | 8-bit Runs well | 11–20 |
| M5 Ultra | 96 GB | 8-bit Runs well | 87–155 |
| M5 Max | 36 GB | 8-bit Runs well | 33–59 |
| M5 Pro | 24 GB | 8-bit Runs well | 22–40 |
| M5 | 16 GB | 8-bit Runs well | 11–20 |
| M4 Max | 36 GB | 8-bit Runs well | 30–53 |
| M4 Pro | 24 GB | 8-bit Runs well | 20–35 |
| M4 | 16 GB | 6-bit Runs well | 11–20 |
| M3 Ultra | 96 GB | 8-bit Runs well | 59–106 |
| M3 Max | 36 GB | 8-bit Runs well | 22–39 |
| M3 Pro | 18 GB | 8-bit Runs well | 11–19 |
| M3 | 8 GB | 6-bit Runs well | 10–17 |
| M2 Ultra | 64 GB | 8-bit Runs well | 58–103 |
| M2 Max | 32 GB | 8-bit Runs well | 29–52 |
| M2 Pro | 16 GB | 8-bit Runs well | 15–26 |
| M2 | 8 GB | 6-bit Runs well | 10–17 |
| M1 Ultra | 64 GB | 8-bit Runs well | 58–103 |
| M1 Max | 32 GB | 8-bit Runs well | 29–52 |
| M1 Pro | 16 GB | 8-bit Runs well | 15–26 |
| M1 | 8 GB | 8-bit Slow | 5–9 |
Measured by others
Decode speeds for Nemotron-Labs Dory 1.0 4B that other people measured on their own Macs and posted publicly. These are third-party measurements, not ours, and not the estimates above.
| Mac | Runtime | Precision | Decode tok/s | Source |
|---|---|---|---|---|
| MacBook Pro M5 Max 128 GB | dory-mlx | 8-bit (MLX) | 692 loops (default), short prompt, median of repeated runs | NetProjects @netprojects |
| MacBook Pro M5 Max 128 GB | dory-mlx | 8-bit (MLX) | 54at 32K context2 loops (default), after a 32K-token prompt, median of repeated runs | NetProjects @netprojects |
Results vary by runtime and settings: its version, the precision, the context length, the prompt and reasoning on or off all change the speed.
How to run it
Step by step: download and run LLMs on a Mac · MLX vs Ollama vs llama.cpp
Prefer an app to the Terminal? Compare the Mac apps that run local models in the AI App Store: price, licence and where your data goes.
About Nemotron-Labs Dory 1.0 4B
| Made by | NVIDIA |
|---|---|
| Parameters | 3.98B (dense) |
| Context | 256K tokens |
| Licence | OpenMDW-1.1 |
| Released | 2026-10 |
| Input | Text |
| Weights | nvidia/NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16 |
Questions
- How much memory does Nemotron-Labs Dory 1.0 4B need?
- About 2.2 GB of weights at 4-bit, plus room for the context: 3.7 GB in all. On a Mac that means 8 GB of unified memory or more, since macOS lets the GPU use about 75% of it by default.
- How fast is Nemotron-Labs Dory 1.0 4B on a Mac?
- Writing speed depends on memory bandwidth. The fastest Mac for it here is an M5 Ultra with 96 GB at about 87–155 tokens a second (8-bit). The smallest Mac that runs it, with 8 GB, manages about 10–17 on an M3.
- Does quantization hurt Nemotron-Labs Dory 1.0 4B?
- At 4 bits most models stay close to the original quality; below 4 bits quality drops faster. Pick the highest precision that fits your Mac with room to spare.
Similar models
For your model card or README
Shows the smallest Mac that runs Nemotron-Labs Dory 1.0 4B well and links here.
[](https://mlx.app/run/nemotron-labs-dory-1.0-4b)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.
Open data: mlx.app/data (JSON, CSV)