Building a sub-AUD 5k local inference rig: used 3090s vs AMD vs Intel Arc
Smooth-Television-48 · reddit · 2026-09-27
A Reddit user lays out options for a local inference rig under AUD 5,000:
- Entry-level local inference starts around AUD 2000-2500 (used 3090, dual 3060s, B60 48GB, r9700 32GB)
- Candidates: two used 3090s, four 16GB cards, Intel Arc B60 dual-GPU (48GB on one slot), AMD r9700 or waiting for the 10x series
- Unified memory systems judged not worth the price/performance today
Workloads include DL/ML training, running a coding agent, local imagegen and finetuning/distillation. The author's take: NVIDIA still dominates speed and software ecosystem, but price is the primary concern for other vendors' lag.
More from Infra
- AI agent cuts Orpheus 3B TTS RTF from 1.03 to 0.87 via TensorRT tuning — TheMoonMidas · 2026-09-27
- TensorFold doubles Qwen 27B decode to 158 tok/s on M3 Ultra with parallel-lane speculative decoding — TheMoonMidas · 2026-09-27
- Skeptical math: DeepSeek unlikely to profit on Huawei hardware vs Liang's 10-month payback target — teortaxesTex · 2026-09-27
- Oracle invokes force majeure as blocked data centers hit $200 billion — generativist · 2026-09-27
- I measured an LLM semantic cache: unsafe queries embed NEARER than safe ones — most systems shouldn't run one — bfeeny · 2026-09-27
- Sugon-8000 showcased with immersion phase-change liquid cooling using 50C refrigerant — teortaxesTex · 2026-09-27