Can I run Qwen3 14B on the GeForce RTX 3060 12GB?
Runs wellQ4_K_M fits in VRAM with 0.9 GB to spare: Runs well, est. 24.4 tok/s (21.4–27.3, calibrated estimate ±12%). Q2_K (heavy quality loss) would reach Runs great, est. 35.5 tok/s (31.3–39.8, calibrated estimate ±12%).
12 GB VRAM · 360.0 GB/s · FP16 12.7 TFLOPS. Assumes 32 GB of DDR5-5600 system RAM, Windows with this GPU driving the display, 8k context and an f16 KV cache. The verdict and quant file above stay the same anywhere from 16 to 128 GB of system RAM.
Every quant of Qwen3 14B on the GeForce RTX 3060 12GB
| Quant | File size | Verdict | Speed | Memory | Runs as | Notes |
|---|---|---|---|---|---|---|
| Q2_K | 5.75 GB | Runs great | est. 35.5 tok/s31.3–39.8calibrated estimate ±12% | 8.0 / 12.0 GB | Full GPU | — |
| IQ4_XS | 8.14 GB | Runs well | est. 26.6 tok/s23.4–29.8calibrated estimate ±12% | 10.2 / 12.0 GB | Full GPU | — |
| Q4_K_Mbaseline | 9.00 GB | Runs well | est. 24.4 tok/s21.4–27.3calibrated estimate ±12% | 11.1 / 12.0 GB | Full GPU | — |
| Q8_0 | 15.70 GB | Runs slowly | est. 5.2 tok/s4.2–6.3calibrated estimate ±20% | 12.0 / 12.0 GB + 6.2 GB RAM | Partial offload | — |
Models with a thinking mode spend extra tokens thinking, so their speed thresholds are 1.5× stricter (30 / 12 / 3 tok/s). With thinking turned off, a model may do one grade better.
Where the memory goes at Q4_K_M
- Weights
- 8.6 GB
- KV cache
- 1.3 GB
- Compute buffer
- 0.6 GB
- OS reserve
- 0.6 GB
- Free
- 0.9 GB
0.44 GB of the 9.00 GB file stays off the graphics card: llama.cpp keeps the input embedding table in system RAM.
Context length vs. KV cache
| Context | KV f16 | KV q8_0 | KV q4_0 |
|---|---|---|---|
| 4k | 9.8 GBRuns well est. 26.1 tok/s22.9–29.2calibrated estimate ±12% | 9.5 GBRuns well est. 26.9 tok/s23.7–30.2calibrated estimate ±12% | 9.3 GBRuns well est. 27.4 tok/s24.1–30.7calibrated estimate ±12% |
| 8k | 10.5 GBRuns well est. 24.4 tok/s21.4–27.3calibrated estimate ±12% | 9.9 GBRuns well est. 25.9 tok/s22.8–29.0calibrated estimate ±12% | 9.5 GBRuns well est. 26.9 tok/s23.6–30.1calibrated estimate ±12% |
| 16k | 11.9 GBRuns slowly est. 14.6 tok/s11.7–17.5calibrated estimate ±20% | 10.6 GBRuns well est. 24.1 tok/s21.2–27.0calibrated estimate ±12% | 10.0 GBRuns well est. 25.8 tok/s22.7–28.9calibrated estimate ±12% |
| 32k | 14.7 GBRuns slowly est. 6.0 tok/s4.8–7.2calibrated estimate ±20% | 12.2 GBRuns slowly est. 12.8 tok/s10.3–15.4calibrated estimate ±20% | 10.9 GBRuns well est. 23.9 tok/s21.1–26.8calibrated estimate ±12% |
S = Runs great · A = Runs well · B = Runs slowly · F = Won't run
At Q4_K_M the verdict stays Runs well up to 8k context with an f16 KV cache, and up to 16k with q8_0.
Estimated speed
- Generation
- est. 24.4 tok/s (21.4–27.3, calibrated estimate ±12%)
- Prompt processing
- about 916 tok/s (549–1,282, rough estimate ±40%)
Measured on this exact combination
No public measurement for this exact combination yet; the numbers above are estimates.
If this is not enough
A larger model that also runs well on the GeForce RTX 3060 12GB
Runs wellQwen3.5 35B-A3B: Runs well, est. 49.4 tok/s (34.6–64.2, theoretical estimate ±30%)
A lower quant of Qwen3 14B
Runs greatQ2_K (5.75 GB, heavy quality loss): Runs great, est. 35.5 tok/s (31.3–39.8, calibrated estimate ±12%)
Cheapest GPU that runs Qwen3 14B great
Runs greatGeForce RTX 5060 Ti 16GB — $569 street (as of 2026-08-09): est. 30.3 tok/s (26.7–34.0, calibrated estimate ±12%).
How to run it
Commands for the Q4_K_M file. Both tools download from Hugging Face on first run.
File: Qwen3-14B-Q4_K_M.gguf (9.00 GB) from unsloth/Qwen3-14B-GGUF.
llama-server -hf unsloth/Qwen3-14B-GGUF:Q4_K_M -c 8192 -ngl all -fa on- -ngl all loads every layer on the GPU.
- -fa on enables flash attention (needed for KV cache quantization).
- --cache-type-k q8_0 --cache-type-v q8_0 shrinks the KV cache to 0.72 GB at 8k context.
PowerShell (quit the Ollama tray app first):
$env:OLLAMA_CONTEXT_LENGTH="8192"; ollama serve
ollama run hf.co/unsloth/Qwen3-14B-GGUF:Q4_K_M- Ollama defaults to 4k context on GPUs under 24 GB. Set OLLAMA_CONTEXT_LENGTH when starting the server, or /set parameter num_ctx inside the chat.
/set parameter num_ctx 8192
Related pages
Guide: How much VRAM do local LLMs need? GPU requirements for 8–32 GB (2026)
Other models on the GeForce RTX 3060 12GB
Qwen3 14B on other hardware
Frequently asked questions
Can I run Qwen3 14B on the GeForce RTX 3060 12GB?
Q4_K_M fits in VRAM with 0.9 GB to spare: Runs well, est. 24.4 tok/s (21.4–27.3, calibrated estimate ±12%). Q2_K (heavy quality loss) would reach Runs great, est. 35.5 tok/s (31.3–39.8, calibrated estimate ±12%). Q4_K_M needs 10.5 GB at 8k context; this setup has 11.4 GB of usable memory and 28.0 GB of free system RAM.
How much context can Qwen3 14B use on the GeForce RTX 3060 12GB?
At Q4_K_M the verdict stays Runs well up to 8k context with an f16 KV cache (1.3 GB of KV) and up to 16k with a q8_0 KV cache (1.4 GB). The model supports up to 32k.
Which quant should I use, and how fast is it?
Q4_K_M (9.00 GB) is the recommended balance: Runs well, est. 24.4 tok/s (21.4–27.3, calibrated estimate ±12%). It is also the largest tracked file that gets this verdict.
Speeds are estimates from memory bandwidth, calibrated against public benchmarks, and each one comes with an error band and a confidence label. Real results vary with drivers, backend, context length and thermals.