768GB of VRAM for less than one RTX 6000: a 12x CMP170HX budget inference rig
segmond · reddit · 2026-09-18
Redditor segmond built a 768GB VRAM inference rig from twelve 64GB CMP170HX mining cards for less than the price of a single RTX 6000 Pro, linked via fiber to a second rig for RPC memory expansion. Running GLM, DeepSeek, Qwen, Kimi and MiniMax models through vLLM or llama.cpp, he claims performance no single RTX 6000 or M3 Mac Studio can match, and urges others to pounce on compute deals as demand stays high.
More from Infra
- Fire at Taiwan Panasonic PCB materials plant may tighten CCL supply further — zephyr_z9 · 2026-11-04
- Same Claude model served by Bedrock one day and Anthropic the next, dev finds — georgos_pappas · 2026-11-04
- Self-improving software is inevitable: closing the agent-observability loop — bibryam · 2026-11-04
- Gigabyte's AI TOP motherboard packs 2TB RAM and 9 GPUs for home AI labs — vincent_koc · 2026-11-04
- Building a 7-GPU 132GB VRAM local inference rig: consolidate on EPYC or split across two systems? — infectiousstupidity · 2026-11-04
- imec neural probe integrates photonics directly into 130nm CMOS backend — jwt0625 · 2026-11-04