Skip to content

Phi 4 Mini Reasoning VRAM requirements

How much VRAM Phi 4 Mini Reasoning needs
4-bit weights: 2.2 GBWith 8K context: 4.1 GBWith 32K context: 7.1 GB8-bit, 8K context: 5.7 GB
Parameters
3.8B
Longest context
128K
Released
Apr 2025
License
MIT
Strongest at
Not rated

How much VRAM Phi 4 Mini Reasoning needs

Memory Phi 4 Mini Reasoning needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit7.1 GB9.1 GB12.1 GB24.1 GB
8-bit3.8 GB5.7 GB8.7 GB20.7 GB
6-bit2.9 GB4.9 GB7.9 GB19.9 GB
5-bit2.5 GB4.5 GB7.5 GB19.5 GB
4-bit2.2 GB4.1 GB7.1 GB19.1 GB
3-bit1.7 GB3.7 GB6.7 GB18.7 GB
2-bit1.3 GB3.3 GB6.3 GB18.3 GB
1.6-bit755 MB2.7 GB5.7 GB17.7 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run Phi 4 Mini Reasoning

The best way Phi 4 Mini Reasoning runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit20K45 tok/s0.5 sPDF: Phi 4 Mini Reasoning on RTX 4060 report
RTX 50608-bit20K72 tok/s0.4 sPDF: Phi 4 Mini Reasoning on RTX 5060 report
RTX 30708-bit20K72 tok/s0.4 sPDF: Phi 4 Mini Reasoning on RTX 3070 report
RTX 3080 10 GB8-bit36K116 tok/s0.2 sPDF: Phi 4 Mini Reasoning on RTX 3080 10 GB report
RTX 3060 12 GB8-bit52K58 tok/s0.6 sPDF: Phi 4 Mini Reasoning on RTX 3060 12 GB report
Arc B5808-bit52K51 tok/s0.4 sPDF: Phi 4 Mini Reasoning on Arc B580 report
RTX 40708-bit52K80 tok/s0.3 sPDF: Phi 4 Mini Reasoning on RTX 4070 report
RTX 4070 Super8-bit52K80 tok/s0.2 sPDF: Phi 4 Mini Reasoning on RTX 4070 Super report
RTX 50708-bit52K104 tok/s0.2 sPDF: Phi 4 Mini Reasoning on RTX 5070 report
RTX 4060 Ti 16 GB8-bit80K47 tok/s0.3 sPDF: Phi 4 Mini Reasoning on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit80K72 tok/s0.3 sPDF: Phi 4 Mini Reasoning on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit80K61 tok/s0.4 sPDF: Phi 4 Mini Reasoning on Arc A770 16 GB report
Radeon RX 9070 XT8-bit80K82 tok/s0.2 sPDF: Phi 4 Mini Reasoning on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit80K104 tok/s0.2 sPDF: Phi 4 Mini Reasoning on RTX 4070 Ti Super report
RTX 4080 Super8-bit80K113 tok/s0.1 sPDF: Phi 4 Mini Reasoning on RTX 4080 Super report
RTX 5070 Ti8-bit80K134 tok/s0.2 sPDF: Phi 4 Mini Reasoning on RTX 5070 Ti report
RTX 50808-bit80K142 tok/s0.1 sPDF: Phi 4 Mini Reasoning on RTX 5080 report
Radeon RX 7900 XTX8-bit128K115 tok/s0.3 sPDF: Phi 4 Mini Reasoning on Radeon RX 7900 XTX report
RTX 30908-bit128K139 tok/s0.2 sPDF: Phi 4 Mini Reasoning on RTX 3090 report
RTX 40908-bit128K148 tok/s0.1 sPDF: Phi 4 Mini Reasoning on RTX 4090 report
Radeon AI PRO R97008-bit128K82 tok/s0.2 sPDF: Phi 4 Mini Reasoning on Radeon AI PRO R9700 report
RTX 50908-bit128K238 tok/s0.1 sPDF: Phi 4 Mini Reasoning on RTX 5090 report
2× RTX 30908-bit128K139 tok/s0.1 sPDF: Phi 4 Mini Reasoning on 2× RTX 3090 report
2× RTX 40908-bit128K148 tok/s0.1 sPDF: Phi 4 Mini Reasoning on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit128K189 tok/s0.1 sPDF: Phi 4 Mini Reasoning on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit128K142 tok/s0.1 sPDF: Phi 4 Mini Reasoning on RTX 6000 Ada report
L40S8-bit128K130 tok/s0.1 sPDF: Phi 4 Mini Reasoning on L40S report
2× RTX 50908-bit128K238 tok/s0.1 sPDF: Phi 4 Mini Reasoning on 2× RTX 5090 report
4× RTX 30908-bit128K139 tok/s0.1 sPDF: Phi 4 Mini Reasoning on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit128K238 tok/s0.1 sPDF: Phi 4 Mini Reasoning on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit44K23 tok/s4.7 sPDF: Phi 4 Mini Reasoning on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit80K23 tok/s4.7 sPDF: Phi 4 Mini Reasoning on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit120K23 tok/s4.7 sPDF: Phi 4 Mini Reasoning on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit80K48 tok/s2.3 sPDF: Phi 4 Mini Reasoning on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit128K48 tok/s2.3 sPDF: Phi 4 Mini Reasoning on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit128K48 tok/s2.3 sPDF: Phi 4 Mini Reasoning on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit128K80 tok/s1.2 sPDF: Phi 4 Mini Reasoning on M4 Max 128 GB report
M5 Max 128 GB8-bit128K88 tok/s0.3 sPDF: Phi 4 Mini Reasoning on M5 Max 128 GB report
M3 Ultra 96 GB8-bit128K98 tok/s0.7 sPDF: Phi 4 Mini Reasoning on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit128K98 tok/s0.7 sPDF: Phi 4 Mini Reasoning on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit128K98 tok/s0.7 sPDF: Phi 4 Mini Reasoning on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit128K128 tok/s0.2 sPDF: Phi 4 Mini Reasoning on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit128K36 tok/s0.7 sPDF: Phi 4 Mini Reasoning on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit128K36 tok/s0.7 sPDF: Phi 4 Mini Reasoning on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit128K49 tok/s0.2 sPDF: Phi 4 Mini Reasoning on RTX Spark report
DGX Spark8-bit128K45 tok/s0.2 sPDF: Phi 4 Mini Reasoning on DGX Spark report
Dual DGX Spark8-bit128K45 tok/s0.1 sPDF: Phi 4 Mini Reasoning on Dual DGX Spark report
DGX Station GB3008-bit128K562 tok/s0.1 sPDF: Phi 4 Mini Reasoning on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit128K238 tok/s0.1 sPDF: Phi 4 Mini Reasoning on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit128K238 tok/s0.1 sPDF: Phi 4 Mini Reasoning on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit128K238 tok/s0.1 sPDF: Phi 4 Mini Reasoning on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit128K262 tok/s0.1 sPDF: Phi 4 Mini Reasoning on A100 80 GB SXM report
H100 SXM8-bit128K371 tok/s0.1 sPDF: Phi 4 Mini Reasoning on H100 SXM report
4× H200 NVL8-bit128K460 tok/s0.1 sPDF: Phi 4 Mini Reasoning on 4× H200 NVL report
8× H200 NVL8-bit128K460 tok/s0.1 sPDF: Phi 4 Mini Reasoning on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf unsloth/Phi-4-mini-reasoning-GGUF
vLLM
vllm serve microsoft/Phi-4-mini-reasoning
MLX (Mac)
mlx_lm.chat --model mlx-community/Phi-4-mini-reasoning-4bit

Questions

How much VRAM does Phi 4 Mini Reasoning need?

Phi 4 Mini Reasoning needs 2.2 GB for its weights at 4-bit, 4.1 GB with an 8K context and 7.1 GB with 32K. At 8-bit, 5.7 GB.

What is the minimum VRAM for Phi 4 Mini Reasoning?

About 3.7 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 4060.

How big is Phi 4 Mini Reasoning?

3.8B parameters: 7.1 GB of weights at 16-bit, 2.2 GB at 4-bit.

Can Phi 4 Mini Reasoning run on an RTX 4090?

Yes. At 8-bit it holds 128K of context and writes about 148 tokens a second.

More models