AppleMac / unified memoryCurrent

Apple M5 Ultra for local LLMs

Name: LLM Configurator — GPU VRAM Checker
Author: LLM Configurator

Written by Jakub Rusinowski · Last updated September 19, 2026

With 512 GB of LPDDR5X at 1,229 GB/s, the M5 Ultra runs 163 catalogued models at Q4_K_M with 8K context. The largest that fits is Step 5 Preview 600B-A27B (~366.4 GB), and the top pick is Nemotron 3 Ultra 550B at 12–26 tok/s.

Apple's quad-die Ultra: up to 36 CPU cores, 80 GPU cores and 512 GB of unified memory at about 1.2 TB/s — the most memory bandwidth in any machine a person can buy off the shelf, and enough capacity to hold models that otherwise need a rented server.

Check my hardware →Full benchmark table

Models that run on the M5 Ultra

Q4_K_M, 8K context, 384 GB usable. Ranked by quality and speed.

Model	VRAM	Speed	Fit
Nemotron 3 Ultra 550B Nemotron 3 Ultra	339.3 GB	12–26 tok/s	Fits
Beam 1.0 501B-A23B Reflection Beam	303.5 GB	21–44 tok/s	Fits
Qwen3-Coder 480B-A35B (MoE) Qwen3-Coder	291.6 GB	18–38 tok/s	Fits
MiniMax M3 428B-A23B MiniMax M3	259.2 GB	22–45 tok/s	Fits
MiniMax M3.1 Flash 428B-A23B MiniMax M3	259.2 GB	22–45 tok/s	Fits
Llama 4 Maverick 17B Llama 4	243.9 GB	32–65 tok/s	Fits
Trinity-Large-Thinking Trinity	242 GB	42–88 tok/s	Fits
Qwen 3.5 397B-A17B Qwen 3.5	240.5 GB	27–55 tok/s	Fits
Nex-N2.5 Pro Nex-N2.5	240.5 GB	27–55 tok/s	Fits
Nex-N2 Pro Nex-N2	240.5 GB	27–55 tok/s	Fits
K2 Horizon 375B-A23B K2 Horizon	231.8 GB	24–51 tok/s	Fits
GLM-5.3-Flash 320B-A18B GLM-5.3-Flash	194 GB	26–54 tok/s	Fits

Showing 12 of 163

Buy it or rent the same memory

Buy the card, or rent a GPU with the same memory by the hour to try models first.

Buy this hardware Apple Mac Studio M3 Ultra — 512 GB VRAM · 60 W board power Deploy in the cloud now RunPod

or compare on Vast.ai

As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.

Speed vs other GPUs

Llama 3.1 8B, Q4_K_M. Estimated ranges. How this is calculated

M5 Ultra59–123 tok/s M3 Ultra43–90 tok/s M1 Ultra43–89 tok/s M2 Ultra43–89 tok/s M5 Max34–71 tok/s M4 Max31–64 tok/s M5 Max (32-core GPU)27–56 tok/s M4 Max (32-core GPU)24–50 tok/s

Specifications

Specs last updated 2026-09-19.

Memory: 512 GB LPDDR5X
Memory bandwidth: 1,229 GB/s
Memory bus: 1024-bit
Architecture: ARM, 3nm TSMC
Series: Apple Silicon
Board power: 120 W
Release year: 2026
Compute backends: METAL
Usable for models: 384 GB (~75% of unified)

Best for512 GB unified memoryfrontier models locallyMac Studio

Frequently asked questions

Can the Apple M5 Ultra run local LLMs?

Yes. With 512 GB (384 GB usable by a model) it runs 163 of the catalogued models at Q4_K_M with 8K context; the largest is Step 5 Preview 600B-A27B, needing about 366.4 GB.

How fast is the Apple M5 Ultra for AI inference?

It is estimated to run Llama 3.1 8B at 59–123 tok/s at Q4_K_M. Llama 3.3 70B is estimated at 9.6–20 tok/s. These are modelled estimates from memory bandwidth, not measurements; the methodology page shows the formula.

What LLMs can I run on 512 GB?

Among the best that fit: Nemotron 3 Ultra 550B, Beam 1.0 501B-A23B, Qwen3-Coder 480B-A35B (MoE), MiniMax M3 428B-A23B, MiniMax M3.1 Flash 428B-A23B.

More for this card

Tools

Guides