@liambuilds.ai: Building a 4-GPU AI rig from scratch. No API costs. No rate limits. No asking permission. Local LLMs. Local inference. Full control. Installing drivers. Pulling models. Testing all 4 GPUs. Everything runs on my desk now. Full build breakdown coming soon. Follow Liambuilds.ai to see what this machine can do. #ai #localai #gpu #pcbuild #llm

Myple :

dude spent 15k to run a 50-70b param model off a basic rag with ollama as if that beats my 20 dollar opus 4.6 😹

Mr. Wesley :

$20k set up no rate limits 😂 just a pay as you go electric bill API🤣

Brent Harrington Prime :

Any Claude model you download right now like opus 4.7 or 4,8 are only the perceived model. The exact architecture, transformer design and training is something that currently only lives on Anthropics servers, so you get what you pay for.

Name :

GSD v2 was just released. Holy hell. In some ways it's better than Claude Code. Can't wait for it to get strong support for git workspaces. It's immature and it still cooks.

wallopingwizard :

lol you need 8x H100s MINIMUM to run anything bro

user2618301715675 :

Can’t do anything serious with local models sorry

hoperegained :

how can you get local LLM models?

❌️ mrplacetx ❌️ :

Why are you running multiple GPUs? Are you planning a service? I run a 24gb 3090 and a 32gb 5090. 32b model on one and. 70b on the other.

cutelittle :

buy a Blackwell

mrkrabs.ai :

Basically Claude is building it. 💀 You’re just prompting.

Johnny :

and a solar system for electricity and UPS for power outage

eldermael :

I have this with solar panels and life is good man

Bebbi :

4 GPUs = 4 * 850 Watt + CPU 200 ~ 4000 Watt = 0.72 $ 1 hour = 518 $ a month 🤣🙈 (if it runs 24/30 in high load)

Random Chaos (>-.-)> :

what's your fav coder model and what's your tps on it.

Doodle-Shop :

piyaheart 🇺🇸 :

Nice. would love to see the setup.

FreeSpace :

What models tho? Qwen?

dijeiu :

Senza nessun costo 🤣🤣 solo di gpu e corrente ti pagavi 100 anni di claude

ky-no official :

nicr!

George Liontos |AI & Marketing :

Electrical bill api

johnjohn :

Seriously you compare the cost of a subsription on claude code lets say 180. Per month with the cost of 4 gpu which is like 20.000 dollar?

Mark G.Z. 💬 :

Opus 4.6 still better in some use cases

Tree Luvva :

The rate limit is you get 1/50th the token throughout

itsNate :

mousewithoutborders bro

Daniel Krueger :

amd and nvidia got AI boxes with 128gb and they're like 2-4k. going the gpu route seems like a waste of money because you need a $8000 Blackwell card for the same memory of a $3000 spark or amd Ai workstation(cant remember what they named it).