Check my GPU

on

Can the NVIDIA GeForce RTX 4070 SUPER run Phi-3 Medium (14B)?

73 ptsGreat (A) · 73 of 100 points: Model quality 19.7/45, Memory fit 24.0/25, Speed 15.0/15, Memory use 15.0/15

Yes, with little memory to spare

Tight fit. At Q4_K_M it needs about 10.6 GB of 12.0 GB usable, with an estimated speed of 44.3 tok/s.

Best quant
Q4_K_M
Generation speed
44.3 tok/s
Prompt processing
1,831 tok/s
First token (1K prompt)
0.6 s
Largest safe context
4K

Download Phi-3 Medium (14B)

Recommended for the NVIDIA GeForce RTX 4070 SUPER: Q4_K_M. Every other quantization has its own download link in the table below.

Quickest · Ollama

One command downloads a ready-to-run version and starts a chat.

ollama run phi3:14b-medium-4k-instruct-q4_K_M

Get Ollama at ollama.com/download (opens in a new tab) · model page (opens in a new tab)

Quantized files · GGUF

Works in LM Studio, llama.cpp, Jan and Ollama. No sign-in needed.

Download the Q4_K_M file (opens in a new tab)

huggingface.co/bartowski/Phi-3-medium-4k-instruct-GGUF (opens in a new tab)

ollama run hf.co/bartowski/Phi-3-medium-4k-instruct-GGUF:Q4_K_M

Original weights · Hugging Face

Full-precision files from Microsoft, for fine-tuning, vLLM, Transformers.

huggingface.co/microsoft/Phi-3-medium-4k-instruct (opens in a new tab)

No sign-in needed.

Official repo · GitHub

Reference code, inference scripts and docs from Microsoft. The weights themselves download from Hugging Face.

github.com/microsoft/PhiCookBook (opens in a new tab)

git clone https://github.com/microsoft/PhiCookBook

Microsoft’s official Phi cookbook: setup, inference and fine-tuning guides.

Links checked 7 days ago against Hugging Face, Ollama and GitHub.

How the A grade is scored

Every result gets up to 100 points from four parts. The points decide the letter grade.

Model qualityEditorial quality tier of the model, scaled by the quantization or precision used.19.7 / 45
Memory fitFull marks when everything sits in GPU or unified memory; fewer when part spills to system RAM.24.0 / 25
SpeedGeneration speed against a 40 tokens/s target for LLMs, or render time against a target for image and video.15.0 / 15
Memory useRewards using the memory you have, so a bigger, better variant beats a tiny one that leaves most memory idle.15.0 / 15
Total73.7 / 10073 ptsGreat (A) · 73 of 100 points: Model quality 19.7/45, Memory fit 24.0/25, Speed 15.0/15, Memory use 15.0/15

Grade scale: S 85+ · A 72+ · B 58+ · C 42+ · D 25+ · F under 25 points.

Every quantization on the NVIDIA GeForce RTX 4070 SUPER

Quant Memory needed Available Speed Fit Grade · points Download
Q2_K 7.6 GB 12.0 GB 72.8 tok/s Runs well 69 ptsGood (B) · 69 of 100 points: Model quality 14.5/45, Memory fit 25.0/25, Speed 15.0/15, Memory use 14.7/15 GGUF (opens in a new tab)
Q3_K_M 8.9 GB 12.0 GB 57.3 tok/s Runs well 72 ptsGreat (A) · 72 of 100 points: Model quality 17.4/45, Memory fit 25.0/25, Speed 15.0/15, Memory use 15.0/15 GGUF (opens in a new tab)
Q4_K_M 10.6 GB 12.0 GB 44.3 tok/s Tight fit 73 ptsGreat (A) · 73 of 100 points: Model quality 19.7/45, Memory fit 24.0/25, Speed 15.0/15, Memory use 15.0/15 GGUF (opens in a new tab)
Q5_K_M 12.0 GB 12.0 GB 36.6 tok/s Tight fit 73 ptsGreat (A) · 73 of 100 points: Model quality 20.1/45, Memory fit 24.0/25, Speed 14.4/15, Memory use 15.0/15 GGUF (opens in a new tab)
Q6_K 13.5 GB 12.0 GB 16.6 tok/s Runs with offload 55 ptsUsable (C) · 55 of 100 points: Model quality 20.5/45, Memory fit 10.0/25, Speed 9.7/15, Memory use 15.0/15 GGUF (opens in a new tab)
Q8_0 16.8 GB 12.0 GB 7.2 tok/s Heavy offload 46 ptsUsable (C) · 46 of 100 points: Model quality 20.7/45, Memory fit 4.0/25, Speed 6.4/15, Memory use 15.0/15 GGUF (opens in a new tab)
F16 30.0 GB 12.0 GB 2.0 tok/s Heavy offload 43 ptsUsable (C) · 43 of 100 points: Model quality 20.7/45, Memory fit 4.0/25, Speed 3.3/15, Memory use 15.0/15 Original weights (opens in a new tab)

Memory figures assume 4K of context. Longer conversations or documents grow the KV cache.

Other hardware that runs it

Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.