Apple silicon
Apple Mac mini (M5 Pro)
- Memory
- 64 GB
- Bandwidth
- 307 GB/s
Largest popular model at 4-bit: Qwen3.6 35B A3B
Up to 64 GB at 307 GB/s in a quiet mini PC: 27–35B models at a comfortable pace.
from $1,699 (launch price)Recommended
NVIDIA · Edge AI · 2023
NVIDIA Jetson AGX Orin 64GB Developer Kit
Older hardware
64 GB at 60 W with CUDA. Now outclassed for LLMs by Strix Halo and the Mac mini.
- Memory
- 64 GB
- Bandwidth
- 205 GB/s
- Power
- 60 W
- Largest popular model at 4-bit
- Llama 3.3 70B
from $1,999 (launch price)
Researched from manufacturer data and public sources; not tested by us.· Last reviewed Oct 11, 2026
What it can run
EstimatePopular models at Q4_K_M with a 32K-token context. Memory needed from real model files; speed is the theoretical ceiling.
Change model size or context
Model weightsContext cache (KV)RuntimeUsable limit (colour = verdict)
| Model | Verdict | Memory | Speed ceiling |
|---|---|---|---|
| Qwen3 1.7B | Runs | ≤ 185 tok/s | |
| Qwen3.5 9B | Runs | ≤ 36 tok/s | |
| Gemma 4 12B | Runs | ≤ 29 tok/s | |
| gpt-oss-20b | Runs | ≤ 102 tok/s | |
| Qwen3.6 27B | Runs | ≤ 12 tok/s | |
| Qwen3.6 35B A3B | Runs | ≤ 111 tok/s | |
| Llama 3.3 70B | Runs, tight | ≤ 5 tok/s | |
| gpt-oss-120b | Not enough memory | — | |
| Qwen3.5 122B A10B | Not enough memory | — | |
| Qwen3.5 397B A17B | Not enough memory | — |
Theoretical maximum for generating text: memory bandwidth divided by the bytes read per token. Real runs usually reach about half to three quarters of it.
Where to buy
Prices are editorial references (launch price and a typical street price on the date shown), not live prices. In 2026 the memory shortage has pushed many prices well above launch.
- Jetson AGX Orin 64GB Developer Kit64 GBLaunch price: $1,999
- Check price Amazon.com (opens in a new tab)Amazon.comCheck availability
Other regions (3)
- Check price Amazon.co.uk (opens in a new tab)Amazon.co.ukUnited Kingdom · Check availability · Search results
- Check price Amazon.es (opens in a new tab)Amazon.esSpain · Check availability
- Check price Amazon.de (opens in a new tab)Amazon.deGermany · Check availability · Search results
Prices are shown only if updated in the last 24 hours and may change; the retailer’s price at checkout applies. We may earn a commission from these links at no cost to you. Affiliate disclosure
Overview
Editorial opinionBest for: Edge and robotics projects that also need a mid-size language model.
Editorial opinionWould we buy it today?
Only for edge or robotics work; for a home LLM server look elsewhere.
Pros and cons
Editorial opinionPros
- 64 GB with CUDA at 60 W
- Industrial-grade platform
Cons
- 205 GB/s
- Expensive for what it does in 2026
Specifications
Factual data- Manufacturer
- NVIDIA
- Chip
- Orin
- Architecture
- Ampere GPU + Arm Cortex-A78AE
- Type
- Edge board / accelerator
- Form factor
- Developer kit / board
- Released
- 2023
- Availability
- On sale
- Memory options
- 64 GB
- Memory type
- LPDDR5 unified
- Memory bus
- 256-bit
- Memory bandwidth
- 205 GB/s
- Usable for models (default)
- 58 GB / 64 GB
- INT8 AI performance
- 275 TOPS
- Board / max power
- 60 W
- Compute stack
- CUDA (JetPack)
- Models
- Any model that fits
- Several units for one model
- No
Source: Manufacturer specifications
Software support
Editorial opinionHow well the popular local-AI tools run on it today (editorial assessment).
- llama.cppGood
- OllamaGood
- LM StudioNot supported
- vLLMWorks with effort
- MLXNot supported
Measured speeds
Factual dataMeasured results with their source. We never invent speeds.
No measured results yet. The table above shows the theoretical ceiling instead.
Alternatives
Apple silicon
Apple Mac Studio (M5 Max)
- Memory
- 128 GB
- Bandwidth
- 614 GB/s
Largest popular model at 4-bit: Qwen3.5 122B A10B
128 GB at 614 GB/s: gpt-oss-120b and 122B MoE models at a good pace, almost silently.
from $2,499 (launch price)Highly recommended
NVIDIA GPUs
NVIDIA RTX PRO 6000 Blackwell
- Memory
- 96 GB
- Bandwidth
- 1.79 TB/s
Largest popular model at 4-bit: Qwen3.5 122B A10B
96 GB on one card at RTX 5090 speed: gpt-oss-120b and 122B MoE models entirely in GPU memory.
from $8,565 (typical price)For advanced users