Skip to content

Qwen2.5 72B Instruct on M3 Ultra 96 GB: VRAM, speed and fit

Qwen2.5 72B Instruct at 6-bit runs on the M3 Ultra 96 GB with 32K of context at about 8.6 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on M3 Ultra 96 GB · 6-bit (Q6_K) · Ollama / LM Studio · Chat

GPU memory
90.3 GB
of 96.0 GB
Context
32K
auto
Speed
8.6 tok/s
Reading speed
First reply
14 s
2K prompt

Where the memory goes

Model
55.5 GB
Context
10.0 GB
Overhead
809 MB
OS
24.0 GB
Free
5.7 GB
Memory
90.3 GB of 96.0 GB
Context
32K tokens · auto
Speed
8.6 tok/s · reading speed
First reply
14 s for a 2K prompt
Running cost
$3.55/mo at home · $122/mo on a cloud 4× RTX A4500

Validation: Not validated

What Qwen2.5 72B Instruct on M3 Ultra 96 GB was checked against
CheckTargetResultStatus
Fits in GPU memory72.0 GB usable66.3 GB neededPass
Headroom5% free5.7 GB free (8%)Pass
Context for chat16K or more32KPass
Writing speed15 tok/s or more8.6 tok/sCheck
First replyunder 3 s, 2K prompt14 sFails

Targets: a reply you can read along with, a first reply the use case tolerates, and 5% of memory left free. Calculated from published specs, not benchmarked.

The setup

Model
Qwen2.5 72B Instruct
Parameters
72.7B
Precision
6-bit (Q6_K)
Context
32K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
M3 Ultra 96 GB
GPU can use
72.0 GB of 96.0 GB

Running cost

At home
$3.55/mo
162 W at the wall
Renting instead
$122/mo
4× RTX A4500 at $1.00/h

4 h a day, power at $0.18 per kWh, cloud at RunPod on-demand prices as of Nov 22, 2026.

How to run it

llama.cpp
llama-server -hf bartowski/Qwen2.5-72B-Instruct-GGUF
vLLM
vllm serve Qwen/Qwen2.5-72B-Instruct
MLX (Mac)
mlx_lm.chat --model mlx-community/Qwen2.5-72B-Instruct-4bit

Qwen2.5 72B Instruct on similar machines

All machines: Qwen2.5 72B Instruct VRAM requirements. Everything that runs here: what can the M3 Ultra 96 GB run?

Other models that run on M3 Ultra 96 GB

Questions

Can Qwen2.5 72B Instruct run on M3 Ultra 96 GB?

Qwen2.5 72B Instruct at 6-bit runs on the M3 Ultra 96 GB with 32K of context at about 8.6 tokens a second.

Report VR-0DAI3LT, calculated Nov 22, 2026 from published specs, not benchmarked.