Is 16GB RAM enough to run an LLM locally?
Yes, for 7B to 14B class models at Q4. On a 16GB Mac the GPU can use about 10.7 GB, which fits Phi 4 reasoning vision 15B at IQ4_XS (est.). On a PC running on the CPU, 16 GB leaves about 10 GB after the OS, enough for Gemma 4 12B at Q4_K_M, but CPU speed is much lower.
Checked 2026-10-22 · memory figures are estimates
16GB Mac (unified memory, ~10.7 GB GPU-usable, 8k ctx, est.)
| Model | Best quant | Needs |
|---|---|---|
| Phi 4 reasoning vision 15B | IQ4_XS | 10.7 GB |
| Ministral 3 14B (2512) | IQ4_XS | 9.7 GB |
| Gemma 4 12B | Q5_K_M | 10.4 GB |
| GLM 4.6V Flash | Q6_K | 9.8 GB |
| Qwen3.5 9B | Q6_K | 9.2 GB |
| Ministral 3 8B (2512) | Q6_K | 9.4 GB |
| granite 4.2 8b | Q6_K | 9.5 GB |
| Qwen3 VL 8B | Q6_K | 9.4 GB |
Full 16GB Mac list · all Mac sizes
16 GB PC, CPU only (~10 GB free after the OS, Q4_K_M, 4k ctx, est.)
| Model | Needs |
|---|---|
| Gemma 4 12B | 8.8 GB |
| GLM 4.6V Flash | 7.3 GB |
| Qwen3.5 9B | 6.9 GB |
| Ministral 3 8B (2512) | 6.8 GB |
| granite 4.2 8b | 6.9 GB |
| Qwen3 VL 8B | 6.8 GB |
| LFM2.5 8B A1B (MoE) MoE | 6.0 GB |
| Gemma 4 E4B | 5.8 GB |
We have no measured CPU-only speeds yet. See Can you run an LLM on CPU only?
How we got this number
Memory needed = parameters × bits per weight ÷ 8 (Q4_K_M is 4.85 bits, Q8_0 is 8.5), plus the KV cache for your context and runtime buffers. Worked example: Phi 4 reasoning vision 15B has 15.12B parameters, so Q4_K_M weights are about 9.2 GB; with an 8k f16 KV cache (1.7 GB) and buffers the total is about 11.9 GB. All estimates. Full method.
FAQ
Is 16GB unified memory on a Mac the same as 16 GB VRAM?
No. macOS lets the GPU use only part of it by default, about 67% below 64 GB, so a 16GB Mac acts like roughly 10.7 GB of VRAM. You can raise the limit with iogpu.wired_limit_mb at your own risk.
Is 8GB RAM enough for an LLM?
Only for small models. An 8GB Mac gives the GPU about 5.4 GB, enough for roughly 3B to 4B models at Q4 (est.). 16 GB is a much more useful floor.
Should I get 16GB or 24GB?
If local AI matters, more memory. A 24GB Mac fits 14B models with more context and some 20B to 30B MoE models at lower quants. Compare the 16GB and 24GB Mac lists.
Open the checker · What fits my GPU?
More short answers
- What LLM can I run on my GPU?
- How much VRAM do I need for a 70B model?
- MLX vs GGUF on a Mac: which is faster?
- How much VRAM do I need for a 7B model?
- How much VRAM do I need for a 32B model?
- Is 12GB VRAM enough for local LLMs?
- How much VRAM does context length use?
- Can you run an LLM on CPU only?
- Q4 vs Q8: which quantization should I use?
Will It Run? Pro NOT BUILT YET
A personal setup for your exact hardware (Ollama / llama.cpp configs ready for your coding agent), an alert when a new model fits, and a one-click cloud fallback when it doesn't.
Not built yet. Join the waitlist, $10/mo planned, no charge now. What's planned
Email only. No payment info.