Local model planning

Size the hardware around the workload.

A practical guide to common parameter classes, approximate memory requirements, and the kinds of local hardware that can support them.

Planning estimates · Not a performance guarantee
01Choose a model class

Start with the smallest parameter range likely to meet the workflow.

02Choose a precision

4-bit runs are more compact; 8-bit runs need more memory but preserve more numerical detail.

03Add headroom

Reserve capacity for context, software, concurrent users, and future expansion.

Reference table

Model size to hardware class

Memory figures are broad planning ranges for inference, not training. Actual requirements depend on architecture, quantization format, context length, and runtime.

Model sizeApprox. 4-bit memoryApprox. 8-bit memorySuggested system RAMPotential hardwareTypical fit
1–3B parameters2–4 GB4–7 GB8 GB minimum · 16 GB comfortableRecent laptop or mini PC; integrated graphics; entry-level GPU with 4–6 GB VRAMLight assistants, classification, extraction, and constrained edge tasks
7–8B parameters6–8 GB10–14 GB16 GB minimum · 32 GB comfortableModern desktop CPU; 8–12 GB GPU; or a system with 16+ GB unified memoryGeneral chat, document help, summarization, and everyday local workflows
13–14B parameters10–14 GB18–24 GB32 GB minimum · 48 GB comfortable12–16 GB GPU for compact runs; 24 GB GPU or 32+ GB unified memory for more headroomStronger writing, reasoning, coding, and private knowledge workflows
20–24B parameters16–20 GB28–38 GB48 GB minimum · 64 GB comfortable24 GB GPU; 48+ GB unified-memory workstation; or CPU inference with 64 GB RAMHigher-quality business assistance and more demanding multi-step work
30–35B parameters22–28 GB42–56 GB64 GB minimum · 96 GB comfortableSingle high-memory GPU above 24 GB; two coordinated GPUs; or 64–96 GB unified memoryAdvanced analysis, coding, and higher-accuracy professional workflows
65–70B parameters42–50 GB80–100 GB96 GB minimum · 128+ GB comfortableMultiple GPUs totaling 48+ GB VRAM; 96–128 GB unified-memory workstation; or large RAM serverHigh-capability local deployments where quality matters more than compact hardware
100–120B parameters65–85 GB130–180 GB192 GB minimum · 256+ GB comfortableMulti-GPU server; high-memory accelerator platform; or enterprise workstation with very large unified memorySpecialized, costly installations with careful performance and power planning

Planning principles

Capacity is more than parameter count.

These estimates create a useful starting boundary. A pilot should still validate the exact model, runtime, document size, response speed, and number of simultaneous users.

01

Model weights

The table estimates memory for quantized model weights plus practical runtime headroom.

02

Context matters

Long prompts and large documents increase the key-value cache and can materially raise memory use.

03

Speed varies

A model that fits in system RAM may still run slowly without enough memory bandwidth or GPU acceleration.

04

Workload first

The best appliance is the smallest configuration that reliably handles the intended private workflow.

From estimate to appliance

Start with the work that must stay local.

The final hardware choice should follow the workload, privacy boundary, response-time target, and expected number of users.

Discuss a pilot