PCIe Link Running at Gen1 x1 Instead of Expected Speed – DGX-Spark

Accelerated Computing DGX Spark / GB10 User Forum DGX Spark / GB10
1

Dear NVIDIA Support,

We are running an Ollama inference workload on our DGX-Spark server. During monitoring, we found that the PCIe link to the GPU (device 000f:01:00.0) is operating at Gen1 x1​ instead of the expected higher speed.

**System information:**​

  • Server model: DGX-Spark

  • GPU: NVIDIA GB10 (device ID 10de:2e12)

  • PCIe slot: 000f:01:00.0 (connected via 000f:00:00.0 PCIe Root Port)

  • Driver version: 580.126.09

  • Kernel: 6.17.0-1008-nvidia

**Observed behavior:**​

  • lspci -vv -s 000f:01:00.0shows:

    • LnkCap: Speed 2.5GT/s, Width x16

    • LnkSta: Speed 2.5GT/s, Width x1 (downgraded)

  • dmesgreports:

    • “0.000 Gb/s available PCIe bandwidth, limited by Unknown x0 link … capable of 32.000 Gb/s with 2.5 GT/s PCIe x16 link”

    • Some PCIe devices report “insufficient power on the PCIe slot”.

**Questions:**​

  1. Is this PCIe slot designed to support only Gen1 speeds, or should it reach Gen2/Gen3/Gen4?

  2. Could the reduced link width (x1) be caused by power or thermal throttling?

  3. Are there any firmware updates or configuration changes that could allow higher PCIe speed and width?

Best regards

3

@ruofei it’s not an error! The GPU is connected to the CPU via NVLink-C2C (chip-to-chip) bus and not PCIe bus. Search the forum for other posts discussing this issue. It’s safe to ignore those messages.

1 Like