PCIe bottleneck dilemma: Adding RAM vs. GPU for local AI workloads
dsdt · reddit · 2026-08-28
A user seeks advice on upgrading a build (x870e, 9700x, dual 5060 ti, 32GB RAM) for local AI. The choice is between adding another 32GB RAM for dual-channel or a third 5060 ti for more VRAM. The constraint is the motherboard's PCIe slot layout (x16/x4/x1), forcing the new GPU to run at x1 bandwidth. The user is concerned about performance impact from the limited PCIe lanes and currently runs Q6 with 132k context at 70 t/s.
More from Infra
- MCPay: Pay-Per-Call Metering and Runaway Loop Protection for MCP Servers — TooDu0 · 2026-08-28
- GLM-5.3 Flash speedup with DFlash2 speculative decoding — No_Afternoon_4260 · 2026-08-28
- AI Surge Drives Record Profits for Chinese MLCC Makers — pstAsiatech · 2026-08-28
- MiniMax Boosts Alibaba Cloud Budget by 220% to Fuel AI Model Growth — pstAsiatech · 2026-08-28
- Metal port of SP1 prover shows up to 5.75x speedup on M4 Max — StefanoGogioso · 2026-08-28
- Open source Open edX MCP server enables natural language course management — Ornery_Landscape6715 · 2026-08-28