16 GB RAM (CPU) — what LLMs can it run?
59 of 110 indexed models fit comfortably in 16GB at 4K context, each at the highest-quality quant that still leaves headroom.
The short answer for 16 GB RAM (CPU)
16 GB, bandwidth not listed. 59 of 110 models in this index fit comfortably at 4K context; 67 load at all.
- Biggest that fits
- ERNIE 4.5 21B-A3B — 14.1 GB at Q4_K_M — 1.9 GB spare at 4K, so longer context comes out of a thin margin.
- Room to grow
- Stable LM 2 12B Chat — 8.3 GB at Q4_K_M — under 60% of the card, which leaves 7.7 GB for a long context window or a second process.
- The wall you will hit
- Speed, not capacity. Every model on this page "fits" in 16 GB of system memory, but CPU inference reads the weights over DDR channels rather than a graphics bus — typically well under a tenth of a discrete card's bandwidth, and dependent on your DIMM configuration rather than on the RAM total. That is why no bandwidth figure is listed here. Treat the smaller entries as the usable range and the large ones as "will load".
32B · 2
| Model | Quant | Est. VRAM | Headroom | tok/s |
|---|---|---|---|---|
| ERNIE 4.5 21B-A3B21B-A3B | Q4_K_M | 14.05 GB | +1.9 GB | — |
| GPT-OSS 20B21B MoE | MXFP4 | 12.76 GB | +3.2 GB | 1954090 ref |
14B · 18
| Model | Quant | Est. VRAM | Headroom | tok/s |
|---|---|---|---|---|
| InternLM2 20B Chat20B | Q4_K_M | 13.43 GB | +2.6 GB | 784090 ref |
| DeepSeek-Coder-V2-Lite Instruct16B-A2.4B | Q4_K_M | 10.08 GB | +5.9 GB | 1454090 ref |
| DeepSeek-V2-Lite Chat16B-A2.4B | Q4_K_M | 10.08 GB | +5.9 GB | 1424090 ref |
| StarCoder2 15B15B | Q4_K_M | 10.19 GB | +5.8 GB | 924090 ref |
| Qwen3 14B Instruct14B | Q5_K_M | 11.65 GB | +4.4 GB | 784090 ref |
| Qwen2.5 14B Instruct14B | Q5_K_M | 11.73 GB | +4.3 GB | 864090 ref |
| DeepSeek-R1-Distill-Qwen-14B14B | Q4_K_M | 10.14 GB | +5.9 GB | 954090 ref |
| Phi-4 14B14B | Q5_K_M | 11.77 GB | +4.2 GB | 784090 ref |
| Phi-3 Medium 14B Instruct14B | Q6_K | 12.86 GB | +3.1 GB | 884090 ref |
| Mistral Nemo 12B Instruct12B | Q6_K | 11.14 GB | +4.9 GB | 954090 ref |
| Gemma 3 12B IT12B | Q5_K_M | 9.84 GB | +6.2 GB | 924090 ref |
| Mellum2.1 12B-A2.5B Thinking12B-A2.5B | Q8_0 | 13.6 GB | +2.4 GB | — |
| Stable LM 2 12B Chat12B | Q4_K_M | 8.35 GB | +7.7 GB | 1084090 ref |
| Llama 3.2 11B Vision Instruct11B | Q8_0 | 12.9 GB | +3.1 GB | 724090 ref |
| Solar 10.7B Instruct11B | Q4_K_M | 7.6 GB | +8.4 GB | 1254090 ref |
| Falcon 3 10B Instruct10B | Q4_K_M | 7.28 GB | +8.7 GB | 1184090 ref |
| Gemma 2 9B Instruct9B | Q8_0 | 11.7 GB | +4.3 GB | 1084090 ref |
| Qwen3.5 9B9B | Q8_0 | 10.37 GB | +5.6 GB | — |
7B · 25
| Model | Quant | Est. VRAM | Headroom | tok/s |
|---|---|---|---|---|
| GLM-4-9B-Chat9B | Q8_0 | 10.16 GB | +5.8 GB | 1054090 ref |
| Qwen3-VL 8B Instruct8B | Q8_0 | 10.39 GB | +5.6 GB | 1084090 ref |
| Granite 4.2 8B8B | Q8_0 | 10.45 GB | +5.6 GB | — |
| Ministral 3 8B Instruct8B | Q8_0 | 10.02 GB | +6 GB | — |
| Qwen2-VL 7B Instruct7B | Q4_K_M | 5.49 GB | +10.5 GB | 724090 ref |
| Granite 3.1 8B Instruct8B | Q4_K_M | 5.88 GB | +10.1 GB | 1424090 ref |
| Qwen3 8B Instruct8B | Q6_K | 7.64 GB | +8.4 GB | 1224090 ref |
| Llama 3.1 8B Instruct8B | Q8_0 | 9.47 GB | +6.5 GB | 1184090 ref |
| Nous Hermes 3 Llama 3.1 8B8B | Q4_K_M | 5.64 GB | +10.4 GB | 1484090 ref |
| Aya 23 8B8B | Q4_K_M | 5.64 GB | +10.4 GB | 1454090 ref |
| OpenChat 3.6 8B8B | Q4_K_M | 5.64 GB | +10.4 GB | 1464090 ref |
| DeepSeek-R1-Distill-Llama-8B8B | Q5_K_M | 6.51 GB | +9.5 GB | 1284090 ref |
| InternLM2 7B Chat7B | Q4_K_M | 5.45 GB | +10.6 GB | 1484090 ref |
| Qwen2.5 7B Instruct7B | Q6_K | 6.77 GB | +9.2 GB | 1324090 ref |
| Qwen2.5-Coder 7B Instruct7B | Q4_K_M | 5.07 GB | +10.9 GB | 1584090 ref |
| DeepSeek-R1-Distill-Qwen-7B7B | Q4_K_M | 5.07 GB | +10.9 GB | 1524090 ref |
| Gemma 4 E4B ITE4B | Q8_0 | 8.46 GB | +7.5 GB | — |
| OLMo 2 7B Instruct7B | Q8_0 | 10.3 GB | +5.7 GB | 1254090 ref |
| OLMo 3 7B Instruct7B | Q8_0 | 10.3 GB | +5.7 GB | — |
| WizardLM-2 7B7B | Q4_K_M | 5.14 GB | +10.9 GB | 1524090 ref |
| Mistral 7B Instruct v0.37B | Q6_K | 6.75 GB | +9.3 GB | 1354090 ref |
| Zephyr 7B Beta7B | Q6_K | 6.75 GB | +9.3 GB | 1324090 ref |
| Gemma 4 E2B ITE2B | Q8_0 | 5.2 GB | +10.8 GB | — |
| Qwen3.5 4B4B | Q8_0 | 4.97 GB | +11 GB | — |
| Gemma 3 4B IT4B | Q8_0 | 5.05 GB | +11 GB | 1454090 ref |
≤3B · 14
| Model | Quant | Est. VRAM | Headroom | tok/s |
|---|---|---|---|---|
| Qwen3 4B Instruct4B | Q6_K | 4.05 GB | +12 GB | 1454090 ref |
| Phi-4 Mini Instruct3.8B | Q8_0 | 4.81 GB | +11.2 GB | 2624090 ref |
| Phi-3.5 Mini Instruct3.8B | Q8_0 | 5.89 GB | +10.1 GB | 2554090 ref |
| Granite 4.2 3B3B | Q8_0 | 4.41 GB | +11.6 GB | — |
| Llama 3.2 3B Instruct3B | Q8_0 | 4.05 GB | +12 GB | 2854090 ref |
| Qwen2.5 3B Instruct3B | Q8_0 | 3.59 GB | +12.4 GB | 2904090 ref |
| SmolLM3 3B3B | Q8_0 | 3.73 GB | +12.3 GB | — |
| Gemma 2 2B Instruct2B | Q8_0 | 3.34 GB | +12.7 GB | 3204090 ref |
| Qwen3.5 2B2B | Q8_0 | 2.23 GB | +13.8 GB | — |
| Qwen3 1.7B Instruct1.7B | Q8_0 | 2.39 GB | +13.6 GB | 2404090 ref |
| Qwen2.5 1.5B Instruct1.5B | Q8_0 | 1.83 GB | +14.2 GB | 4104090 ref |
| Llama 3.2 1B Instruct1B | Q8_0 | 1.51 GB | +14.5 GB | 4504090 ref |
| Qwen3.5 0.8B0.8B | Q8_0 | 0.94 GB | +15.1 GB | — |
| Qwen2.5 0.5B Instruct0.5B | Q8_0 | 0.6 GB | +15.4 GB | 5404090 ref |
How this list is built
Each row is the lowest-perplexity-loss quant of that model whose estimated total — weights plus KV cache at 4K context plus activation buffer — uses at most 88% of the card. That is the calculator's "green" threshold, so every row here has real headroom rather than only just fitting. Raise the context length and the list shortens; the calculator lets you check any combination directly. The speed column is this card: a measured run, a range scaled from this card's own Llama 2 7B benchmark, or a bandwidth ceiling. A figure that only exists for the RTX 4090 is labelled as that card.
A further 8 models load but with no headroom to spare (up to 105% of VRAM) — the Quant Hub’s GPU chips count those too, which is why its number is higher.
Measured on this card
No benchmark runs in this index were recorded on a 16 GB RAM (CPU). Every figure on this page is calculated from the model architecture and the quant level — treat them as estimates, not measurements.
Setup guides for this card
Step-by-step guides written for the 16 GB RAM (CPU)'s runtime or memory size, with the commands and the checks that prove the model is running on the card.
Stepping up
A 32 GB RAM (CPU) (32GB) fits 27 more of the indexed models than this card. 32 GB RAM (CPU) →
Common questions
What is the best local LLM for a 16 GB RAM (CPU)?
For everyday use, Stable LM 2 12B Chat at Q4_K_M — about 8.3 GB of the card's 16 GB at 4K context, leaving 7.7 GB for a longer window. If you want the largest thing that will load, that is ERNIE 4.5 21B-A3B at Q4_K_M (14.1 GB, 1.9 GB spare). "Best" here means best fit for the memory budget — this index does not run task benchmarks, so it cannot tell you which model is smarter.
Can a 16 GB RAM (CPU) run Codestral 22B?
Only just. Its smallest build here, Q4_K_M, needs about 15.0 GB against 16.0 GB usable — that loads with nothing else running, with no margin for a longer context window. It is not on the list above, which requires a model to stay inside 88% of usable capacity.
16 GB RAM (CPU) or 32 GB RAM (CPU) for local LLMs?
They do not hold the same set: 16 GB RAM (CPU) fits 59 and 32 GB RAM (CPU) 86 of 110 at 4K. Bandwidth is not recorded for both cards here, so this page will not rank them on speed.
59 of 110 indexed models fit comfortably in 16GB at 4K context, each at the highest-quality quant that still leaves headroom.
This page's figures change when the model or the runtime does.Last updated 2026-10-18 RSS → /feed.xml