self-hosted/ai
§01·compatibility · /check

Qwen3 14B on RTX 3060

Yes — Qwen3 14B runs on the RTX 3060 (12 GB). Fastest community-measured result: 972.6 prefill tokens/s.

✓ runsllmactive30 series12GB VRAM
step-by-step recipe for this pair

Qwen3-14B on RTX 3060 12GB: Q4_K_M GGUF via Ollama or llama.cpp

llmbeginner12GB+
model
name
Qwen3 14B
slug
qwen3-14b
vertical
llm
status
active
repo
huggingface.co ↗
open detail ↗
gpu
name
RTX 3060
slug
rtx-3060
vram
12 GB
series
30
open detail ↗
§02·benchmarks
TaskQuantSpeedVRAMWorksConfidenceSourceVerified
llmQ4_K31.2tokens/s✓hardware-corner.net· web2026-06-13
llmQ4_K972.6prefill tokens/s✓hardware-corner.net· web2026-06-13
llmQ4_K_M35.9tok/s✓smeltcore.com· community2026-08-08
§03·how it was measured
  • llmQ4_K31.2 tokens/s

    Token generation @ 4K (Hardware Corner RTX 3060 12GB; Q4_K 31.2/22.7 tok/s @ 4K/16K, 32K does not fit)

    recorded from hardware-corner.net ↗ · verified 2026-06-13

  • llmQ4_K972.6 prefill tokens/s

    Prompt processing @ 4K (Hardware Corner RTX 3060 12GB; 972.6/678.2 @ 4K/16K)

    recorded from hardware-corner.net ↗ · verified 2026-06-13

  • llmQ4_K_M35.9 tok/s

    Dense baseline, all layers resident. llama.cpp b10088, CUDA sm_86, flash-attn on. Rig: i7-7700, 32GB DDR4-2133 dual-channel (~34 GB/s), headless 3060. Cmd: llama-bench -ngl 99 -fa 1 -p 512 -n 128 -d 0 -r 3 Depth: 35.9 / 31.4 / 27.9 tok/s at 0 / 8K / 16K. Dense offload cliff, same card: -ngl 41 (all) 35.9; -ngl 20 (~half in RAM) 5.7 (6.3x); -ngl 0 3.05 (11.8x). Dense re-reads every param per token, so offload is catastrophic. Contrast: Qwen3.6-35B-A3B (~3B active) at max offload holds 28.1 — a 1.4x dip. GOTCHA: llama.cpp counts n_layer+1 offloadable layers (40 blocks + output). -ngl 40 here silently leaves block 0 in RAM (log: "offloaded 40/41") and costs ~20%: 28.5 vs 35.9. Use -ngl 99.

    recorded from smeltcore.com ↗ · verified 2026-08-08

§04·more Qwen3 14B recipes
§05·common questions
Can you run Qwen3 14B on RTX 3060?

Yes — Qwen3 14B runs on the RTX 3060 (12 GB). Fastest community-measured result: 972.6 prefill tokens/s.

Which quantizations have been tested for Qwen3 14B on RTX 3060?

Q4_K, Q4_K_M — measured in community benchmarks.

How fast is Qwen3 14B on RTX 3060?

Up to 972.6 prefill tokens/s (llm), the fastest community-measured result.

Are there step-by-step instructions for Qwen3 14B on RTX 3060?

Yes — a step-by-step recipe documents Qwen3 14B on the RTX 3060, linked at the top of this page.