Hardware calculator
What hardware do you need to run LocalAIBox?
LocalAIBox runs on hardware you own — from a compact NVIDIA Spark appliance you can lock in a closet to a full multi-GPU workstation. Tell us your team size and how capable you want the AI to be, and we'll suggest memory and storage configurations. You own the hardware; Mako Logics provides, sets up, and supports the software.
Recommended appliance
NVIDIA Spark (GB10)
Runs large models up to ~70B on a compact, silent box.
- GB10 Grace Blackwell · ~1 petaFLOP AI
- 128 GB unified memory
- Up to 4 TB self-encrypting NVMe
- 20-core Arm CPU · Blackwell GPU
The new sweet spot: workstation-class memory in an appliance that sits on a desk and sips power. Available as a Windows RTX Spark or a Linux DGX Spark — self-encrypting storage makes it a natural fit for HIPAA and GLBA.
These hardware suggestions are for planning purposes. Performance depends on your specific configuration, models, and workload. Mako Logics confirms the exact specification for your needs during your demo. See our Terms of Service.
A compact appliance, a workstation, or a Mac — your choice.
LocalAIBox runs the same software and keeps every prompt and document inside your building on all three. They just trade off in different ways.
NVIDIA Spark
The compact GB10 appliance.
- 128 GB unified memory runs large models up to ~70B
- Sits on a desk, nearly silent, sips power — no tower
- Self-encrypting storage — a natural fit for HIPAA & GLBA
- Ships as a Windows RTX Spark or a Linux DGX Spark
- Best value for most small-to-mid regulated teams
NVIDIA workstation
Maximum speed and concurrency.
- Best for the fastest responses and many simultaneous users
- Discrete pro GPUs (RTX 5090, RTX PRO 6000) for heavy load
- Add or upgrade GPUs later as your team grows
- A tower — needs more space, power, and cooling
- For 30+ heavy users or the very largest models
Mac mini / Studio
Silent Apple Silicon.
- Best for teams already standardized on macOS
- Runs models up to ~32B on a mini; 70B+ on a Mac Studio
- Palm-sized and nearly silent — sits on a desk, sips power
- Premium, low-maintenance, rock-solid appliance
- Unified memory scales to 256 GB on a Mac Studio
For most regulated teams the NVIDIA Spark is the sweet spot — workstation-class capacity in a silent box you can lock in a closet.
Not sure? We'll recommend one at your demo