Enabling P2P mode on dual RTX 3090s; before/after numbers (Qwen3.6-27B INT4, 256k ctx)

A user tested the impact of enabling Peer-to-Peer (P2P) mode on a dual NVIDIA RTX 3090 setup using PCIe 4.0 8x/8x links. The benchmark involved running five passes with nvbandwidth and a standard decode/soak test script for the Qwen3.6-27B INT4 model with a 256k context window.

The author notes that while driver versions changed between runs, the performance direction aligns with previous reports. The testing required approximately 4.5 hours of configuration adjustments to verify the results.

Enabling P2P mode is recommended for users running daily inference workloads on dual GPU rigs, though purchasing a second RTX 3090 solely for this benefit is not advised.