$10-12K build for local agentic coding: 1x 5090 vs 4x 3090 vs Mac Studio M5 Ultra
Shadmelor · reddit · 2026-10-26
A developer is planning a $10K (up to $12K) machine for local agentic coding with Qwen3-27B and asks for sanity checks from people who run coding agents locally.
Use case: agentic coding, sometimes 2–4 parallel subagents, reverse-engineering; optional Hunyuan3D-2.1 texturing (CUDA, 21GB VRAM).
Options under consideration:
- 1x RTX 5090
- 4x used RTX 3090
- Hybrid: 5090 + 2x 3090 as separate vLLM instances (x8/x8/x4 on AM5)
- Mac Studio M5 Ultra 256GB
Open questions: does a single 5090 feel better day-to-day than 4x 3090 for long-context agentic work? Is the 3090's long-context prefill really as bad as reported? Is the hybrid worth the headache? Any better options like waiting for the 50 Super or a modded 48GB 4090?
More from coding & agent
- Someone sent the first customer fax via ChatGPT over Onetime Fax's new MCP server — tdhopper · 2026-10-26
- China Telecom's XingChen lab open-sources Xing4.0-29B-Intent, a 29B MoE with 4B activated params — FellMentKE · 2026-10-26
- Codebases sized to fit one context window dramatically improve agent outcomes — julianharris · 2026-10-26
- 42-Minute Video Breaks Down Why AI Agents Fail: Mostly Not the Model's Fault — JohnSavill · 2026-10-26
- Giraffe: Open-Source Baseline Testing for OpenAI-Compatible LLM Endpoints — ImmediateBig8479 · 2026-10-26
- Running a bed-side AI agent: Obsidian split-screen plus Hyo plugin as a mobile workstation — evielync · 2026-10-26