B200 errors

B200 and GB200 architectures introduce massive NVLink domains across the NVL72 rack, meaning a single link failure or NVSwitch hang can stall the entire multi-node cluster.

Hardware diagnostics on Blackwell require distinguishing between a localized GPU fault, a tray-level power delivery issue, or a cross-rack fabric drop. Always start by pulling the system management logs.

Diagnose your log with evidence Full failure encyclopedia

Working in VS Code or Remote-SSH? Diagnose your log locally with the free VS Code extension. No account or upload required for local diagnosis.

Every common b200 error

The literal string is what you paste into a search bar, so it is the heading. “Class” is a triage category. It is not proof of ownership in your workload.

NetIR link down

Hardware or fabric

Xid 149: NetIR link down

The NVLink network interface encountered a link drop. On B200, this often points to a physical cable, tray connector, or NVSwitch port failure interrupting the NVL72 fabric.

First action: Run nvidia-smi -q -d NVLINK and check dmesg for the exact port identifier before reseating cables or trays.

Full entry: root cause, fix and prevention

NVLink fabric hang

Hardware or fabric

Xid 150: NVLink fabric hang detected

A transaction on the NVLink fabric hung and timed out. This stalls all GPUs participating in the stalled collective across the GB200 NVL72 rack.

First action: Check dmesg across all nodes for Xid 150 to 154 to find the initiating device, rather than trusting the first node to timeout.

Full entry: root cause, fix and prevention

NIXL page OOM

Environment / runtime

NIXL firmware page OOM

The network interface processor ran out of memory for firmware pages. This drops network packets and disrupts distributed training runs.

First action: Inspect dmesg for firmware crash signatures and update the host OS drivers to a release with fixed NIXL page allocations.

Full entry: root cause, fix and prevention

Errors not listed here exist. Rather than guess at their meaning, check NVIDIA Blackwell Architecture Guide, which is the authority for the strings above.

Frequently asked questions

Why does a single B200 link failure hang the whole rack?

Because the entire NVL72 rack acts as a single giant GPU domain connected by NVSwitches. A broken link stalls the collective operation, causing all other healthy GPUs to wait indefinitely.

How do I tell if a B200 error is a GPU fault or a switch fault?

Check the error counters on both the GPU NVLink interface and the NVSwitch. If only one end shows physical layer errors, the problem is likely localized to that specific hardware interface.

What should I do when I see an Xid 149 on a Blackwell node?

Capture the full dmesg output, the system management logs, and the nvidia-smi topology dump. You must identify the exact lane and port before physical intervention is attempted.

Is it safe to reset a B200 tray after a firmware hang?

Only after capturing the crash dump and system logs. Resetting the tray clears the volatile state needed to diagnose whether the hang was caused by thermal pressure or a firmware bug.

Knowing the error is not knowing the cause

Which rank failed first, whether the checkpoint is safe to resume from, and whether this is your code or the hardware. Paste the log and Denpex answers all three, free and without an account.

Diagnose your logs free