Yes, just — 1.0 GB spare
Mistral NeMo 12B at Q4_K_M fits your GeForce RTX 2080 Ti entirely on the GPU at 8K context, at an estimated 54 tokens per second. Past 14K the KV cache pushes it over — quantise the cache to q8_0, or step down a quantisation, to go longer.
Fully on GPU
8K context
Q4_K_M · 6.9 GB
Apache 2.0
Released Jul 2024
Multilingual 12B with a 128K window, built with NVIDIA. A roleplay and fiction favourite that refuses to die.
The VRAM budget
weights 6.9 GB
Weights 6.9 GB
KV cache @ 8K 1.25 GB
Runtime overhead 0.6 GB
Free 1.0 GB of 9.7 GB
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| F16 | 22.7 GB | 24.6 GB | — | ~2.3 | Reference | 14.9 GB over |
| Q8_0 | 12.1 GB | 13.9 GB | — | ~7.3 | −0.1% ppl | 4.2 GB over |
| Q6_K | 9.3 GB | 11.2 GB | — | ~16 | −0.4% ppl | 1.5 GB over |
| Q5_K_M | 8.1 GB | 9.9 GB | 6K | ~38 | −0.8% ppl | 0.2 GB over |
| Q4_K_M | 6.9 GB | 8.7 GB | 14K | 54 | −1.9% ppl | Recommended |
| Q3_K_M | 5.6 GB | 7.4 GB | 22K | 67 | −5.4% ppl | Long context |
| Q2_K | 4.8 GB | 6.6 GB | 27K | 78 | −15% ppl | Long context |
Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it.
How to run it
$ llama-server \
-hf mistralai/Mistral-Nemo-Instruct-2407:Q4_K_M \
-c 8192 -ngl 99
The engine underneath most of the others. Every knob is exposed. More on llama.cpp.
01Download is 6.9 GB. Keep it on an SSD — a first load off a spinning disk takes minutes.
02Close anything else holding VRAM. A browser with hardware acceleration can sit on 1–2 GB.
03Only 1.0 GB is spare, so a long system prompt can still push it over. Its real ceiling here is 14K context.