Dual RTX 5060 Ti P2P tests fail with NaNs despite peer access being enabled
Environmental_Hand35 · reddit · 2026-07-25
A user reports trying to run PCIe peer-to-peer on a dual RTX 5060 Ti workstation with patched NVIDIA open kernel modules. nvidia-smi reports peer access as available in both directions, but the CUDA simpleP2P correctness test fails with NaNs and the direct P2P path is slower than the fallback path.
Setup and findings:
- MSI MPG Z890 Carbon WiFi motherboard
- 2× RTX 5060 Ti 16GB
- Core Ultra 7 270K Plus, 48GB DDR5-8800
- VT-d/IOMMU disabled, with several boot-flag combinations tried
- cudaMemcpyPeer / cudaMemcpy between GPUs reaches about 5.09 GB/s
- The kernel test still reports NaNs on verification
The post asks other multi-GPU Blackwell users to share simpleP2P or nvbandwidth results for comparison and troubleshooting.
More from Infra
- SF Compute founder: buying compute is 'an absolutely awful experience' right now — IgorCarron · 2026-09-11
- SmolVM open-sources persistent computer infrastructure for agents that outlive chat sessions — aniketmaurya · 2026-09-11
- PyTorch Day Korea 2026 launches first offline conf, CFP closes Sept 13 — PyTorch · 2026-09-11
- Local LLM server dilemma: 4x CMP-170HX (price up 53% in 20 days) vs Mac Studio M5 Ultra — rumboll · 2026-09-11
- llama.cpp lands Flash Attention tuning for RDNA4, big prefill gains on AMD — pmttyji · 2026-09-11
- Your p99 latency benchmark may be lying: a deep dive into coordinated omission — Franc0Fernand0 · 2026-09-11