Phi-4 Mini VRAM Requirements

Phi-4 Mini needs about 4.1 GB of VRAM at Q4_K_M with an 8K context: 2.2 GB for weights, 1.0 GB for the KV cache and 0.9 GB of runtime overhead. At Q8_0 it needs 5.7 GB and at full FP16 precision 9.1 GB. The smallest GPU that runs it fully is the RTX 3050 8GB.

By Microsoft · Updated October 14, 2026 · Values from config.json on Hugging Face

4.1 GB VRAM

Phi-4 Mini at Q4_K_M with 8K context and the KV cache in FP16.

  • Weights 2.2 GB
  • KV cache 1.0 GB
  • Runtime overhead 0.9 GB

Which GPUs can run it

Phi-4 Mini VRAM by quantization

Total memory with an FP16 KV cache. "Smallest tier" is the smallest standard VRAM size with a 6% safety margin at 8K.

QuantizationBits/weightWeightsTotal at 8KTotal at 32KSmallest tier
Q4_K_M4.852.2 GB4.1 GB7.1 GB6 GB
Q5_K_M5.692.5 GB4.5 GB7.5 GB6 GB
Q6_K6.592.9 GB4.9 GB7.9 GB6 GB
Q8_08.503.8 GB5.7 GB8.7 GB8 GB
FP16 / BF1616.007.2 GB9.1 GB12.1 GB10 GB
Q3_K_M3.911.7 GB3.7 GB6.7 GB4 GB
Q2_K3.001.3 GB3.3 GB6.3 GB4 GB

VRAM by context length

At Q4_K_M. The KV cache grows linearly with context: 0.13 GB per 1,000 tokens in FP16.

ContextKV cache (FP16)KV cache (Q8_0)Total, FP16 cacheTotal, Q8_0 cache
4K tokens0.5 GB0.3 GB3.6 GB3.4 GB
8K tokens1.0 GB0.5 GB4.1 GB3.6 GB
32K tokens4.0 GB2.1 GB7.1 GB5.2 GB
128K tokens16.0 GB8.5 GB19.1 GB11.6 GB

GPUs that can run Phi-4 Mini

At Q4_K_M with 8K context. "Best quant" is the highest-quality format that fits fully; "max context" is at Q4_K_M.

GPUUsable memoryStatusBest quantMax contextSpeed
RTX 3050 8GB8.0 GBFitsQ8_035K~50 tok/s
RTX 4060 8GB8.0 GBFitsQ8_035K~61 tok/s
RTX 4060 Ti 8GB8.0 GBFitsQ8_035K~64 tok/s
Radeon RX 7600 8GB8.0 GBFitsQ8_035K~54 tok/s
RTX 3060 Ti 8GB8.0 GBFitsQ8_035K~98 tok/s
RTX 3070 8GB8.0 GBFitsQ8_035K~98 tok/s
RTX 5060 8GB8.0 GBFitsQ8_035K~98 tok/s
RTX 5060 Ti 8GB8.0 GBFitsQ8_035K~98 tok/s
RTX 3070 Ti 8GB8.0 GBFitsQ8_035K~132 tok/s
Intel Arc B570 10GB10.0 GBFitsFP16 / BF1650K~65 tok/s
RTX 3080 10GB10.0 GBFitsFP16 / BF1650K~163 tok/s
Apple M4 16GB10.7 GBFitsFP16 / BF1655K~27 tok/s
RTX 2080 Ti 11GB11.0 GBFitsFP16 / BF1657K~134 tok/s
RTX 2060 12GB12.0 GBFitsFP16 / BF1665K~74 tok/s

About Phi-4 Mini

A 3.8B model trained heavily on synthetic reasoning data, with a 200K vocabulary for better multilingual coverage. MIT-licensed, so it suits commercial edge use.

SpecValue
Parameters3.84B
Layers32
Hidden size3072
Attention heads / KV heads24 / 8
Head dimension128
Vocabulary200,064
Max context (used here)128K tokens
KV cache per token (FP16)128 KB
Released2025-02
LicenseMIT

Phi-4 Mini VRAM: common questions

What is the smallest GPU for Phi-4 Mini?

About 4.1 GB at Q4_K_M with an 8K context, so 6 GB cards are comfortable. A 4 GB card works with about 5K of context.

How much VRAM does Phi-4 Mini need?

About 4.1 GB at Q4_K_M with an 8K context, 5.7 GB at Q8_0 and 9.1 GB at FP16. Longer contexts add 0.13 GB per 1,000 tokens with an FP16 cache.

What is the cheapest GPU that can run Phi-4 Mini?

The smallest listed device that fits it fully at Q4_K_M with 8K context is the RTX 3050 8GB (8.0 GB usable).

Related