Phi 4 — Hardware Requirements & GPU Compatibility
ChatMathCode
Microsoft Phi 4 is a 14-billion parameter language model from Microsoft Research's Phi series, designed to deliver strong reasoning, mathematical, and coding performance at an efficient size. Phi 4 continues the Phi family's focus on maximizing capability per parameter through high-quality training data curation, achieving benchmark scores that rival much larger models on reasoning and STEM tasks. The model runs well on consumer GPUs with 12-16GB of VRAM in quantized formats. It excels at mathematical problem solving, code generation, and structured reasoning. Released under the MIT license.
995.1K downloads
2.2K likesSpecifications
- Publisher
- Microsoft
- Family
- Phi 4
- Parameters
- 14B
- Architecture
- Phi3ForCausalLM
- Context Length
- 16,384 tokens
- Vocabulary Size
- 100,352
- Release Date
- 2025-11-24
- License
- MIT
Get Started
HuggingFace
How Much VRAM Does Phi 4 Need?
Select a quantization to see compatible GPUs below.
| Quantization | Bits | VRAM | + Context | File Size | Quality |
|---|---|---|---|---|---|
| IQ2_XS | 2.40 | 4.9 GB | 7.9 GB | 4.20 GB | Importance-weighted 2-bit, extra small |
| IQ2_S | 2.50 | 5.1 GB | 8.0 GB | 4.38 GB | Importance-weighted 2-bit, small |
| IQ2_M | 2.70 | 5.4 GB | 8.4 GB | 4.72 GB | Importance-weighted 2-bit, medium |
| IQ3_XS | 3.30 | 6.5 GB | 9.4 GB | 5.78 GB | Importance-weighted 3-bit, extra small |
| Q2_K | 3.40 | 6.7 GB | 9.6 GB | 5.95 GB | 2-bit quantization with K-quant improvements |
| Q3_K_S | 3.50 | 6.8 GB | 9.8 GB | 6.13 GB | 3-bit small quantization |
| IQ3_M | 3.60 | 7.0 GB | 10.0 GB | 6.30 GB | Importance-weighted 3-bit, medium |
| Q3_K_M | 3.90 | 7.5 GB | 10.5 GB | 6.83 GB | 3-bit medium quantization |
| Q4_0 | 4.00 | 7.7 GB | 10.7 GB | 7.00 GB | 4-bit legacy quantization |
| Q3_K_L | 4.10 | 7.9 GB | 10.8 GB | 7.17 GB | 3-bit large quantization |
| IQ4_XS | 4.30 | 8.2 GB | 11.2 GB | 7.53 GB | Importance-weighted 4-bit, compact |
| Q4_1 | 4.50 | 8.6 GB | 11.5 GB | 7.88 GB | 4-bit legacy quantization with offset |
| Q4_K_S | 4.50 | 8.6 GB | 11.5 GB | 7.88 GB | 4-bit small quantization |
| IQ4_NL | 4.50 | 8.6 GB | 11.5 GB | 7.88 GB | Importance-weighted 4-bit, non-linear |
| Q4_K_M | 4.80 | 9.1 GB | 12.1 GB | 8.40 GB | 4-bit medium quantization — most popular sweet spot |
| Q4_K_L | 4.90 | 9.3 GB | 12.2 GB | 8.58 GB | 4-bit large quantization |
| Q5_K_S | 5.50 | 10.3 GB | 13.3 GB | 9.63 GB | 5-bit small quantization |
| Q5_K_M | 5.70 | 10.7 GB | 13.6 GB | 9.97 GB | 5-bit medium quantization — good quality/size tradeoff |
| Q5_K_L | 5.80 | 10.9 GB | 13.8 GB | 10.15 GB | 5-bit large quantization |
| Q6_K | 6.60 | 12.3 GB | 15.2 GB | 11.55 GB | 6-bit quantization, very good quality |
| Q8_0 | 8.00 | 14.7 GB | 17.7 GB | 14.00 GB | 8-bit quantization, near-lossless |
Which GPUs Can Run Phi 4?
Q4_K_M · 9.1 GBPhi 4 (Q4_K_M) requires 9.1 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 12+ GB is recommended. Using the full 16K context window can add up to 2.9 GB, bringing total usage to 12.1 GB. 28 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 3080 Ti.
Runs great
— Plenty of headroom
Decent
— Enough VRAM, may be tightWhich Devices Can Run Phi 4?
Q4_K_M · 9.1 GB27 devices with unified memory can run Phi 4, including NVIDIA DGX H100, NVIDIA DGX A100 640GB.
Runs great
— Plenty of headroomRelated Models
Derivatives (2)
Frequently Asked Questions
- How much VRAM does Phi 4 need?
Phi 4 requires 9.1 GB of VRAM at Q4_K_M, or 14.7 GB at Q8_0. Full 16K context adds up to 2.9 GB (12.1 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 14B × 4.8 bits ÷ 8 = 8.4 GB
KV Cache + Overhead ≈ 0.7 GB (at 2K context + ~0.3 GB framework)
KV Cache + Overhead ≈ 3.7 GB (at full 16K context)
VRAM usage by quantization
Q4_K_M9.1 GBQ4_K_M + full context12.1 GB- What's the best quantization for Phi 4?
For Phi 4, Q4_K_M (9.1 GB) offers the best balance of quality and VRAM usage. Q4_K_L (9.3 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XS at 4.9 GB.
VRAM requirement by quantization
IQ2_XS4.9 GB~57%Q3_K_S6.8 GB~77%IQ4_XS8.2 GB~87%Q4_K_M ★9.1 GB~89%Q4_K_L9.3 GB~90%Q8_014.7 GB~99%★ Recommended — best balance of quality and VRAM usage.
- Can I run Phi 4 on a Mac?
Phi 4 requires at least 4.9 GB at IQ2_XS, which exceeds the unified memory of most consumer Macs. You would need a Mac Studio or Mac Pro with a high-memory configuration.
- Can I run Phi 4 locally?
Yes — Phi 4 can run locally on consumer hardware. At Q4_K_M quantization it needs 9.1 GB of VRAM. Popular tools include Ollama, LM Studio, and llama.cpp.
- How fast is Phi 4?
At Q4_K_M, Phi 4 can reach ~320 tok/s on AMD Instinct MI300X. On NVIDIA GeForce RTX 4090: ~72 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: AMD Instinct MI300X → 5300 ÷ 9.1 × 0.55 = ~320 tok/s
Estimated speed at Q4_K_M (9.1 GB)
AMD Instinct MI300X~320 tok/sNVIDIA GeForce RTX 4090~72 tok/sNVIDIA H100 SXM~239 tok/sAMD Instinct MI250X~198 tok/sReal-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
- What's the download size of Phi 4?
At Q4_K_M, the download is about 8.40 GB. The full-precision Q8_0 version is 14.00 GB. The smallest option (IQ2_XS) is 4.20 GB.