Redditor runs gpt-oss-120b across a phone, three Macs and two Windows PCs
ANR2ME · reddit · 2026-09-29
Per wccftech, Reddit user "MedicineBlogscanner" managed to run the 4-bit quantized gpt-oss-120b — a model that officially requires at least 60GB of memory — on a wild distributed setup:
- A Galaxy S24+ phone
- An RTX 3060 mini PC with 12GB VRAM
- A Windows laptop with just 12GB of RAM
- An Intel MacBook Pro, a Mac mini, and an M3 MacBook Pro
By pooling memory and VRAM across devices into a distributed inference setup, the user bypassed single-machine memory limits, showing consumer hardware can run 120B-scale open models through unconventional means.
More from Infra
- Nebius cuts agent training batch collection time by 66.9%, from ~10 min to just over 3 — demian_ai · 2026-09-29
- Qwen 27B Q4 with 100K context at ~30 t/s on a 16GB AMD RX 7800 XT: full guide — Haunting-Stretch8069 · 2026-09-29
- Celesto: open-source persistent microVM computers for AI agents, boots in 500ms — aniketmaurya · 2026-09-29
- Meta Muse to cost ~$50 per user per year even under aggressive optimization, back-of-envelope says — bookwormengr · 2026-09-29
- BioNeMo team boosts Mixtral-8x7B training throughput 2.21x vs HF BF16 baseline — AllThingsApx · 2026-09-29
- DeepSeek's elastic compute team is hiring heavily, shares sandbox infra for large-scale agent training — teortaxesTex · 2026-09-29