Can an RTX 3060 12GB run Phi 4?

In practice

Phi 4 sits comfortably on RTX 3060 12GB. At Q4_K_M it takes 9.96 GB of the 11.2 GB available, which leaves 1.24 GB spare, enough for the 8k window we assumed and for a window of up to about 8k tokens before memory runs out. Generation runs at about 27.8 tokens per second, well past the speed anyone reads at, so a 500-token answer takes about 20 seconds.

Of the memory in use, 80% is the weights and 16% is the attention cache. With this much room the question is quality, not fit. The next step up, Q5_K_M, needs 11.37 GB and does not fit, so Q4_K_M is where we would stay.

Reading a 4,000-token prompt, roughly a long email or a few pages of text, takes about 6 seconds before the first word appears. That part is limited by compute and not by bandwidth, so it follows a different spec from the writing speed.

The numbers

Phi 4 has 14.66 billion parameters across 40 layers. The GeForce RTX 3060 12GB has 12 GB of VRAM at 360 GB/s, of which about 11.2 GB is left for a model once the operating system has taken its share.

6 of the 9 levels fit entirely. Levels marked degraded are listed for completeness, not as advice. Rows marked partial or experts in RAM run with part of the model in system RAM, at the speed shown.

How long a conversation it holds

Memory rises with the length of the conversation because the attention cache keeps a key and a value for every token. At Q4_K_M, the longest window that still fits on this device is 8k tokens. Setting the window to what you need and no more is the cheapest way to buy back memory.

The settings we would use

llama-server -m <model>.gguf -ngl all -c 8192 -fa on

In Ollama, set OLLAMA_CONTEXT_LENGTH to the window you want, run ollama ps while the model is loaded and check that the processor column reads 100% GPU. If it shows a split on a model that fits, the Ollama guide shows how to fix it. The llama.cpp flags guide explains each option above.

What else an RTX 3060 12GB runs well

For comparison, these are the other popular models that fit this device at a sound quantization, best pairings first.

The same model on nearby hardware

Memory decides whether Phi 4 loads and bandwidth decides how quickly it answers. Here is what a step down, a card with the same memory and a step up do with it.

Wait for the first word775 ms

Then writes at27.8 tok/s

Whole answer7.0 s

YouWhy does my model use more memory when the conversation gets longer?

Model

Simulated from our estimate at a 512-token question, not a recording. Assumes nothing else is competing for the GPU.

Questions people ask

Keep going

Everything the GeForce RTX 3060 12GB runs, the full breakdown for Phi 4, how the numbers are made in the methodology, or a different pairing in the calculator.