Qwen3 4B
Q4_K_M · 2.5 GB file
4 GB estimated planning VRAM
Good hardware fit
Smallest download here; leaves more room for context and other GPU use.
Ollama tag and sourceModel compatibility detailsYes. The RTX 3060 12GB has an Ollama NVIDIA GPU acceleration path. Qwen3 8B Q4_K_M is a useful starting configuration; Qwen3 14B Q4_K_M has much less spare VRAM. Check context, drivers, and system RAM before loading.
These configurations use the existing catalog and Ollama fit rules. Planning VRAM is model file size plus the larger of 1.5 GB or 25% of file size, rounded to one decimal. It is an estimate with generic headroom, not a measured allocation at any fixed context. For an initial text-only test, set 4,096 tokens and one request at a time; this is a suggested starting setting, not a guarantee that longer contexts fit.
Q4_K_M · 2.5 GB file
4 GB estimated planning VRAM
Good hardware fit
Smallest download here; leaves more room for context and other GPU use.
Ollama tag and sourceModel compatibility detailsQ4_K_M · 5.2 GB file
6.7 GB estimated planning VRAM
Good hardware fit
A middle-sized starting point with more memory headroom than the 14B configuration.
Ollama tag and sourceModel compatibility detailsQ4_K_M · 9.3 GB file
11.6 GB estimated planning VRAM
Good hardware fit
The estimate is close to the card capacity. Start with short context and check actual GPU allocation.
Ollama tag and sourceModel compatibility detailsFull GPU fit means weights and runtime data can stay on the GPU under the chosen settings. A tight fit leaves limited room for the KV cache and other apps; offload uses CPU and system RAM for part of the model. Even a good planning result can become tight as context grows. The advertised model context below is not a promise of usable context on this card.
Install Ollama and a compatible NVIDIA driver using the official GPU guide. Run ollama run qwen3:8b, then /set parameter num_ctx 4096 before your test prompt. In another terminal, use ollama ps to check the PROCESSOR and CONTEXT columns: 100% GPU indicates full GPU loading, while a CPU/GPU split indicates offload. See Ollama's context guide.
New to Ollama's app prerequisites? Read the Ollama system requirements guide before choosing a model; this page focuses on GPU model fit.
VRAM
12 GB
Architecture
Ampere
Acceleration
CUDA
Windows 11
Supported
Interactive check
Change the model or context assumption and see how the hardware-fit signal changes. This does not predict speed or model quality.
Hardware fit
Good hardware fitGPU VRAM
12 GB
Planning VRAM
6.7 GB
Context
8K supported
What this means
The GPU has enough VRAM for the model weights plus the conservative planning headroom used by this tool.
Model file size and planning VRAM are not official VRAM minimums. Exact memory use changes with context, runtime behavior, and offloading.
All recorded models
These results compare the current model catalog against the GPU's recorded VRAM. A model file is not the same thing as an official VRAM requirement.
qwen3:4b
Model file
2.5 GB
Planning VRAM
4 GB
Recorded context
256K
qwen3:8b
Model file
5.2 GB
Planning VRAM
6.7 GB
Recorded context
40K
qwen3:14b
Model file
9.3 GB
Planning VRAM
11.6 GB
Recorded context
40K
gemma3:4b
Model file
3.3 GB
Planning VRAM
4.8 GB
Recorded context
128K
gemma3:12b
Model file
8.1 GB
Planning VRAM
10.1 GB
Recorded context
128K
gemma3:27b
Model file
17 GB
Planning VRAM
21.3 GB
Recorded context
128K
Your exact PC may behave differently
Add your real GPU, RAM, operating system, context needs, and workflow to get the full Advisor and Upgrade Advisor result.
Check my exact PCEvidence
Hardware-fit estimates are planning guidance, not official manufacturer requirements or performance guarantees.