Local LLM Hardware Upgrade Path: RTX 5080 vs Strix Halo
bsofiato · reddit · 2026-07-13
Currently running local models (like Qwen 35b A3B) on a Ryzen 9 5900X and RTX 5080, the author seeks community advice on hardware upgrades to run larger dense and MoE models.
The discussion evaluates the pros and cons of several upgrade paths:
- Upgrading the GPU to an RTX 5090.
- Adding a second RTX 5060 Ti for a dual-GPU setup (limited by motherboard PCIe bandwidth).
- Mixing a Radeon Pro AI R9700 with the RTX 5080.
- Switching to a Strix Halo machine with 128GB unified memory (with concerns about slow speeds).
- Expanding system RAM directly to 128GB to brute-force large parameter MoE models.
More from Infra
- NeurIPS 2026 workshop will focus on on-device intelligence and local execution — YiMaTweets · 2026-07-21
- How to build a PostgreSQL-backed semantic search pipeline with pgvector and Ollama — KhuyenTran16 · 2026-07-21
- NeurIPS 2026 workshop calls papers on on-device intelligence — YiMaTweets · 2026-07-21
- Milled from Solid Aluminum: AI Rig Multi-GPU Case for Local Compute — dee_hw · 2026-07-21
- FutureCaribbean’s Buildathon offers $50K, H200 compute, and an NYSE pitch — HeyAmit_ · 2026-07-21
- A new series tests which data-science workflows can run on GPUs today — pandeyparul · 2026-07-21