Two RTX 3060s for 24GB local LLM VRAM: does the second slot's reduced bandwidth matter?
Loose_Doubt367 · reddit · 2026-11-13
A Reddit user with an AM4 B550M-A board and one RTX 3060 plans to add a second card for 24GB total VRAM to run local LLMs, but worries the second slot's reduced PCIe bandwidth will hurt performance.
The post asks for guidance on running both GPUs while minimizing bandwidth loss — a discussion that touches on B550 lane-splitting options and their real-world impact on inference throughput.
More from Infra
- Zen 6 details: four 256-bit vector loads per core per cycle, faster VNNI throughput — lemire · 2026-11-13
- Bolt Graphics unveils Zeus GPU with expandable memory up to 384GB, mass production in 2027 — HankYeomans · 2026-11-13
- Open OCR costs fell 5x in 18 months: $0.51 to $0.095 per 1,000 pages, benchmarked on 850 pages — vanstriendaniel · 2026-11-13
- Microsoft extends mail timeout to 48 hours as outage hits organizations worldwide — dhruv2038 · 2026-11-13
- Linus Ekenstam: China's energy dominance will 'blow everything else out of the water' — LinusEkenstam · 2026-11-13
- Nscale admits no binding commitments for $32B Anthropic megaproject as losses hit $4 per $1 revenue — GaryMarcus · 2026-11-13