News:

Welcome to Qday.forum  :: Be kind, courteous and help other people.

Main Menu

Rate my quantum computing hot desk setup, also what is everyone running at home for AI work - what would you do

Started by BretHart_Mike, May 19, 2026, 05:58 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

Topic: Rate my quantum computing hot desk setup, also what is everyone running at home for AI work - what would you do   Views(Read 47 times)

BretHart_Mike

Working from home three days a week and I have finally sorted the setup properly. Two monitors, one portrait for reading papers and one landscape for everything else. Mechanical keyboard with red switches which my partner hates. A whiteboard that takes up most of one wall which I thought was overkill and turned out to be the most important thing in the room.

For AI tools I am running local models via Ollama for anything sensitive on an M4 Mac Mini with 64gb unified memory, which handles Qwen and Llama well enough for most local tasks. Cloud models for everything that does not have to stay on device. The unified memory architecture is genuinely good for local inference at this scale and the price point is reasonable.

Curious what other people are running, particularly anyone doing serious numerical simulation or ML training at home rather than on cloud. The home lab energy costs are becoming a real thing to manage

StringTheory51

Portrait monitor for papers changed my reading life, cannot believe I spent so long on landscape for PDFs

HeartbreakKid

M4 Mac Mini 64gb is basically the recommended home AI workstation right now at that price point, good call

RustyHawk

Running a small cluster of Raspberry Pi 5s for hobby ML which is completely impractical and I love it

Brett42

The whiteboard call is so correct. I have filled mine completely twice and both times the process of photographing and clearing it was somehow clarifying

Coder22

64gb unified memory handles 32b parameter models reasonably, have you tried Qwen 2.5 72b at 4 bit quantisation
Normal is overrated

WhatUQuant

Yes, it works but the generation speed is slow enough to be annoying for anything interactive, fine for batch tasks
git commit -m "fixed everything"

GlassyCandle

Home energy costs for serious GPU work are genuinely significant now, I track mine and a 3090 running inference adds about 30 pounds a month to my bill
Cashback on everything or it didn't happen

Gareth_11

That is the argument for cloud inference on anything you run infrequently, the economics only flip if you are running it constantly

TheRock96

Running an older Threadripper workstation for numerical simulation, power hungry but the memory bandwidth is better than Apple silicon for certain workloads
Normal is overrated

Zach91

The standing desk versus sitting desk conversation feels relevant here, what are people doing

JayJ

Sit stand desk, mostly sitting, the standing is for when I am stuck and need to think rather than when I am actually working

Related Topics (6)