willitrun·ai

Can Phi-4-reasoning-plus 14B run on RTX 3060 12GB?

BARELY — Tight on Memory

A72Great
Estimated from fit model

Phi-4-reasoning-plus 14B needs ~14.1 GB VRAM. RTX 3060 12GB has 12.0 GB. With Q4_K_M quantization, expect ~15 tok/s.

Runtime: llama.cppCapacity: OffloadBandwidth: LowStack: StandardBottleneck: Host offload
Share:

Operating mode

Choose the run profile you care about

Interactive favors responsiveness, while light API and scale-out lean harder on serving readiness. The fit stays the same, but the recommendation lens changes.

Current mode

Balanced

Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.

Capabilities:

Select quantization to explore

Q4_K_M (Medium quality) — 14.1 GB, 14.6 tok/s, Very compromised (needs ~1.3 GB host RAM)
14.1 GB required12.0 GB available
118% VRAM needed

2.1 GB over capacity — needs offload or smaller quantization

Fit status

Very compromised (needs ~1.3 GB host RAM)

Decode

14.6 tok/s

TTFT

13263 ms

Safe context

5K

Memory

14.1 GB / 12.0 GB

Offload

20%

Memory breakdown

Weights9.0 GB
KV Cache3.1 GB
Runtime0.9 GB
Headroom1.2 GB

See how fast it feels

See how fast it feelsPhi-4-reasoning-plus 14B on RTX 3060 12GB
1st promptCold start — includes initialization
>Solve: If a train travels at 60 mph for 2.5 hours, how far does it go?
Let me work through this step by step. Given information: - Speed = 60 mph (miles per hour) - Time = 2.5 hours Formula: Distance = Speed x Time Calculation: Distance = 60 mph x 2.5 h = 150 miles The train travels 150 miles. Double-check: at 60 mph the train covers 60 miles each hour. In half an hour it covers 30 miles. So in 2.5 hours: 60 + 60 + 30 = 150 miles. Confirmed.
2nd promptWarm — model ready, no init overhead
>What if the train stops for 20 minutes halfway?
The stop doesn't change the total distance covered while moving, but it affects total trip time. Breakdown: - First half: 1.25 hours at 60 mph = 75 miles - Stop: 20 minutes (no distance) - Second half: 1.25 hours at 60 mph = 75 miles Total distance: 150 miles (same) Total time: 2.5 h + 0.33 h = 2.83 hours Average speed: 150 / 2.83 = 53 mph The distance stays the same but average speed drops to 53 mph because of the stop.
Estimated: 14.6 tok/s decode · 13.3s TTFT (warm) · 37 tok/s prefill

What limits this setup

It fits through host-memory offload, and offload is the main reason performance drops.

CPU or host-memory offload is active

About 20% of the working set spills out of accelerator memory, which usually hurts latency and sustained decode throughput.

Very little memory headroom

You can run the model, but there is not much room left for longer context, bigger batches, extra apps, or future model updates.

Best improvement path

Remove offload with more accelerator memory

Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.

Buy headroom, not only minimum fit

A slightly larger memory tier gives you safer context growth and makes the recommendation more future-proof.

Increase host RAM if you keep offloading

This setup may need roughly 1.3 GB of extra host RAM just for the offloaded portion, before OS and other tools.

Performance by workload

WorkloadGradeFitDecodeTTFTContext
ChatSRuns with offload (needs ~0.4 GB host RAM)18.6 tok/s5686 ms5K
CodingAVery compromised (needs ~1.3 GB host RAM)14.6 tok/s13263 ms5K
Agentic CodingFToo heavy9.7 tok/s29124 ms5K
ReasoningAVery compromised (needs ~1.3 GB host RAM)14.6 tok/s15674 ms5K
RAGFToo heavy9.7 tok/s36405 ms5K

Inference speed

Phi-4-reasoning-plus 14B inference speed — tokens per second by GPU & Mac

Estimated decode speed (tokens/sec) for Phi-4-reasoning-plus 14B at Q4_K_M across popular GPUs and Apple Silicon, using the fastest local runtime per device. Fastest is RTX 5090 32GB at ~144 tok/s. Speed is memory-bandwidth bound, so cards that fit the whole model in VRAM run far faster than ones that offload to system RAM.

GPU / MacMemoryQuantSpeed (tok/s)Fits?
NVIDIARTX 5090 32GB
32 GBQ4_K_M143.9Fits
NVIDIARTX 4090 24GB
24 GBQ4_K_M91.8Fits
RX 7900 XTX 24GB
24 GBQ4_K_M82.9Fits
NVIDIARTX 3090 24GB
24 GBQ4_K_M78.5Fits
NVIDIARTX 4080 Super 16GB
16 GBQ4_K_M75.5Tight
Mac Studio M3 Ultra 256GB
256 GBQ4_K_M66.8Fits
Mac Studio M2 Ultra 128GB
128 GBQ4_K_M55.6Fits
Mac Studio M1 Ultra 128GB
128 GBQ4_K_M52.7Fits
MacBook Pro M4 Max 128GB
128 GBQ4_K_M37.6Fits
MacBook Pro M4 Max 64GB
64 GBQ4_K_M37.6Fits
MacBook Pro M3 Max 64GB
64 GBQ4_K_M28.8Fits
MacBook Pro M1 Max 64GB
64 GBQ4_K_M26.4Fits
NVIDIARTX 4070 12GB
12 GBQ4_K_M24.9Heavy offload
MacBook Pro M4 Pro 48GB
48 GBQ4_K_M23.0Fits
NVIDIARTX 3060 12GB
12 GBQ4_K_M14.6Heavy offload
NVIDIARTX 4060 8GB
8 GBQ4_K_M5.5Too big

Estimates for single-stream decoding at Q4_K_M; real tokens/sec varies with prompt length, context, batch size, and runtime build. Prompt processing (prefill) is faster than the decode figures shown here.

Quantization options

How Phi-4-reasoning-plus 14B (14.699999809265137B params) fits at each quantization level on RTX 3060 12GB (12.0 GB usable).

QuantBitsVRAMQualityFit
Q2_K
2
5.7 GB
LowS92
Q3_K_S
3
7.2 GB
LowS92
NVFP4Best for your GPU
4
8.2 GB
MediumS91
Q4_K_M
4
9.0 GB
MediumF0
Q5_K_M
5
10.6 GB
HighF0
Q6_K
6
12.1 GB
HighF0
Q8_0
8
15.7 GB
Very HighF0
F16
16
30.1 GB
MaximumF0

Get started

Copy-paste commands to run Phi-4-reasoning-plus 14B on your machine.

Run

ollama run phi4-reasoning

Frequently asked questions

Can RTX 3060 12GB run Phi-4-reasoning-plus 14B?

Yes, RTX 3060 12GB can run Phi-4-reasoning-plus 14B with a A grade (Very compromised (needs ~1.3 GB host RAM)). Expected decode speed: 14.6 tok/s.

How much VRAM does Phi-4-reasoning-plus 14B need?

Phi-4-reasoning-plus 14B (14.699999809265137B parameters) requires approximately 14.1 GB of memory with Q4_K_M quantization.

What is the best quantization for Phi-4-reasoning-plus 14B?

The recommended quantization for Phi-4-reasoning-plus 14B is Q4_K_M, which balances quality and memory efficiency.

What speed will Phi-4-reasoning-plus 14B run at on RTX 3060 12GB?

On RTX 3060 12GB, Phi-4-reasoning-plus 14B achieves approximately 14.6 tokens per second decode speed with a time-to-first-token of 13263ms using Q4_K_M quantization.

Can RTX 3060 12GB run Phi-4-reasoning-plus 14B for coding?

For coding workloads, Phi-4-reasoning-plus 14B on RTX 3060 12GB receives a A grade with 14.6 tok/s and 5K context.

What context window can Phi-4-reasoning-plus 14B use on RTX 3060 12GB?

On RTX 3060 12GB, Phi-4-reasoning-plus 14B can safely use up to 5K tokens of context. The model's official context limit is 33K, but available memory constrains the safe maximum.

What should I upgrade first if Phi-4-reasoning-plus 14B feels slow on RTX 3060 12GB?

Remove offload with more accelerator memory. Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.

See all results for RTX 3060 12GBSee all hardware for Phi-4-reasoning-plus 14B
Embed this result▼

Paste this snippet into any page to show a live fit card.

<iframe src="https://willitrunai.com/embed/phi-4-reasoning-plus-14b-on-rtx-3060-12gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>

Preview: