Yes, just — 0.7 GB spare
Qwen3 8B at Q4_K_M fits your GeForce RTX 5060 entirely on the GPU at 8K context, at an estimated 59 tokens per second. Past 12K 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 · 4.6 GB
Apache 2.0
Released Apr 2025
Apache-2.0 alternative to Llama 3.1 8B, with a switchable thinking mode.
The VRAM budget
weights 4.6 GB
Weights 4.6 GB
KV cache @ 8K 1.13 GB
Runtime overhead 0.6 GB
Free 0.7 GB of 7.0 GB
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| F16 | 15.3 GB | 17.0 GB | — | ~3.4 | Reference | 10.0 GB over |
| Q8_0 | 8.1 GB | 9.8 GB | — | ~10 | −0.1% ppl | 2.8 GB over |
| Q6_K | 6.3 GB | 8.0 GB | 1K | ~22 | −0.4% ppl | 1.0 GB over |
| Q5_K_M | 5.4 GB | 7.1 GB | 7K | ~43 | −0.8% ppl | 0.1 GB over |
| Q4_K_M | 4.6 GB | 6.3 GB | 12K | 59 | −1.9% ppl | Recommended |
| Q3_K_M | 3.7 GB | 5.5 GB | 19K | 73 | −5.4% ppl | Long context |
| Q2_K | 3.2 GB | 4.9 GB | 22K | 85 | −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 Qwen/Qwen3-8B:Q4_K_M \
-c 8192 -ngl 99
The engine underneath most of the others. Every knob is exposed. More on llama.cpp.
01Download is 4.6 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 0.7 GB is spare, so a long system prompt can still push it over. Its real ceiling here is 12K context.