---
title: "Dual-GPU bind one card for LLM: questions to ask before you start"
url: https://memory.wiki/PvtJiG6W
updated: 2026-11-16T21:55:23.296Z
source: "memory.wiki"
---
# Dual-GPU bind one card for LLM: questions to ask before you start

## Why "Dual-GPU bind one card for LLM" deserves a written plan

Pin processes; leave the second card for embeddings or idle.  Local AI hardware decisions tend to be made in chat threads and forgotten a month later. When something breaks, nobody remembers why a particular PSU or case was chosen. A one-page plan avoids that.

## Inputs to collect before spending

- Target models and quantisation levels you have already tested on someone else's machine or in a rental.
- The longest realistic context your team uses, measured from real documents, not guessed.
- Number of concurrent users at the busiest hour.
- Room constraints: noise tolerance, ambient temperature, power circuit, and where the box will physically live.
- Who maintains it: driver updates, model pulls, and backups need an owner.

## Translating inputs into hardware

Memory is the first constraint. Estimate weights at your chosen quantisation, add room for the KV cache at your real context length, and keep a margin so the runtime never has to offload layers in the middle of a busy afternoon. Reference tables such as [hardware reviews](https://microaione.com/hardware/) help you sanity-check the estimate. Compute speed matters less than people think for small teams; capacity and stability matter more.

Next comes power and cooling. Measure wall power with an inexpensive plug meter during idle and under a sustained load. Finance teams appreciate real numbers, and you will spot a struggling PSU early. Keep dust filters in the maintenance calendar; clogged intakes are a quiet cause of throttling.

## Common mistakes

- Buying the biggest card and pairing it with an unknown-brand PSU.
- Forgetting that a second GPU needs slot spacing, lanes and a bigger power budget.
- Assuming a laptop GPU behaves like its desktop namesake under long loads.
- Skipping a burn-in, then discovering instability during a demo.

## A worked example without fake numbers

Imagine a five-person office that wants a private assistant for drafting and document search. They write down the model family they tested, the quantisation that still answered Chinese and English questions well, and the longest documents they paste. From that they estimate memory with a comfortable margin and decide that one card with enough VRAM is safer than two smaller cards that complicate the setup. They choose a quality PSU with headroom, a case with clear front intake, and enough system RAM to run an embedding model alongside the main one. After assembly they run a burn-in, check temperatures and logs, and only then run their evaluation prompts. Every figure they share afterwards comes from their own measurements, with the method written next to it.

See also [device-to-model matching](https://microaione.com/match/).

## Bottom line

Local AI rewards patience and record keeping more than raw spending. Decide what the system must do, measure it on your own hardware and documents, write down every change with a date and an owner, and review the setup after each significant upgrade. Small, documented steps are easier to debug, easier to hand over and easier to defend when someone asks why a particular choice was made. When in doubt, test on a smaller scale first and expand only when the evidence supports it.

## Quick reminders

- Publish what the assistant will refuse to do.
- Photograph cable routing before closing the case.
- Cap concurrent sessions before lowering quantisation.
- Leave memory headroom for the OS and the embedding model.

*Published by [MicroAI One (本地AI 1号网)](https://microaione.com/), a Shenzhen team working on local LLM hardware, software and deployment.*

---

## Summary
Building a reliable dual GPU system for local LLMs requires careful planning based on measured memory requirements and power constraints. Documenting these technical choices and performing thorough burn in tests ensures system stability and easier maintenance.

## Themes
- Local AI hardware planning
- VRAM and power management
- System documentation and maintenance

## Key takeaways
- A written plan prevents the loss of technical context regarding hardware choices over time.
- Memory requirements must include model weights, KV cache for real context lengths, and a safety margin to prevent layer offloading.
- Power supply units should be chosen for quality and headroom rather than just raw wattage.
- Burn in testing is necessary to identify instability before the system is used for production tasks.
- System maintenance requires a designated owner for driver updates, model management, and physical cleaning.

## Insights
- Compute speed is often prioritized over capacity, even though VRAM and stability are more critical for small team workloads.
- Physical constraints like slot spacing and power circuit limitations are frequently overlooked in favor of raw GPU performance.
- Documenting the reasoning behind hardware choices is essential for long term maintenance and troubleshooting.

## Open questions / gaps
- What specific metrics or software tools are recommended for performing the burn in test mentioned in the document?

