Current version by Nick Antonaccio
This guy got Qwen 3.6 27b running on 2 ancient 1080ti GPUs, using turboquant:
https://www.youtube.com/watch?v=7V8G7cQBrus
I certainly wouldn't recommend going that low - it's more just an interesting experiment to push the limit of what's possible - but the fact is, you can actually run a useful LLM, with enough context to get real work accomplished, for somewhere around $500.
This guy got Qwen 3.6 27b running on 2 ancient 1080ti GPUs, using turboquant:
https://www.youtube.com/watch?v=7V8G7cQBrus
I certainly wouldn't recommend going that low - it's an interesting experiment to push the limit of what's possible, but the fact is, you can actually run a useful LLM, with enough context to get real work accomplished, for somewhere around $500.
This guy got Qwen 3.6 27b run on 2 ancient 1080ti GPUs, using turboquant:
https://www.youtube.com/watch?v=7V8G7cQBrus
I wouldn't recommend going that low - it's an interesting experiment to push the limit of what's possible, but the fact is, you can run an actually useful LLM with enough context to get real work accomplished, for somewhere around $500.