CUDA error: operation not permitted when stream is capturing

Attempting to perform CPU synchronization or stream switching (e.g..item() or dynamic tensor operations) while a CUDA graph is actively capturing operations. This entry explains how to confirm the cause, apply the fix, and separate it from adjacent cuda-graphs failures.

Symptom

CUDA error: operation not permitted when stream is capturing

Root cause

Attempting to perform CPU synchronization or stream switching (e.g..item() or dynamic tensor operations) while a CUDA graph is actively capturing operations.

Recommended fix

Remove dynamic control flow or CPU-syncing operations from the captured code block, or use eager execution.

How Denpex helps

Denpex investigates CUDA error: operation not permitted when stream is capturing using the evidence you provide or your connected workload collects. Earlier rank, host or application evidence is needed to distinguish an initiating failure from a downstream report.

The literal signature is "CUDA error: operation not permitted when stream is capturing". It is a environment failure associated with CUDA Graphs, custom kernels, and vision pipelines. The line identifies the failing operation or subsystem, while the surrounding evidence decides whether it is the initiating fault or a downstream symptom.

Attempting to perform CPU synchronization or stream switching (e.g..item() or dynamic tensor operations) while a CUDA graph is actively capturing operations. The failure becomes visible at this call site because the operation first requires the missing resource, valid state, healthy peer, or correct result. Earlier log lines and a known-good control carry more causal value than the final wrapper exception.

• The workload stops or loses forward progress after emitting "CUDA error: operation not permitted when stream is capturing".
• A retry on the same configuration reproduces the failure because the causal state has not changed.
• The outer framework exception can hide the rank, node, allocation, or dependency that failed first.
• Increasing timeouts or reducing workload size can suppress the symptom without correcting the cause.

• CUDA Graphs, custom kernels, and vision pipelines
• production-shaped multi-accelerator workloads
• containerized and bare-metal deployments of the same stack

Use this checklist to test the hypothesis against a small reproduction. No single line proves the root cause, so preserve the preceding events and compare one variable at a time.

• ✓Find the first occurrence of "CUDA error: operation not permitted when stream is capturing" and preserve at least 100 lines before it.
• ✓Identify which rank, node, device, or process emitted the first related warning.
• ✓Compare the failing rank, node, input, or configuration with one known-good control.
• ✓Repeat the same input after the targeted change and require the signature to disappear.
• ✓Resume from restart script without CUDA graph capture only after the control passes.

• Attempting to perform CPU synchronization or stream switching (e.g..item() or dynamic tensor operations) while a CUDA graph is actively capturing operations.
• The decisive evidence is the first log line that precedes "CUDA error: operation not permitted when stream is capturing" and differs from a healthy run.
• A nearby failure remains a competing hypothesis until a control separates configuration, capacity, transport, and hardware causes.

CUDA error: operation not permitted when stream is capturing

Use this text as a lookup key in logs and upstream issue trackers. It is not presented as a captured customer log. Confirm the cause from your own preceding events, versions, configuration and the cited references.

Remove dynamic control flow or CPU-syncing operations from the captured code block, or use eager execution. This changes the condition in the causal diagnosis instead of hiding the outer exception. The repeated control proves ownership before recovery from restart script without CUDA graph capture.

# Preserve evidence before restarting
rg -n -i 'error|exception|timeout|failed' <log-file>
nvidia-smi
python -m torch.utils.collect_env

# Find the exact signature in the complete log
rg -n -F -- "CUDA error: operation not permitted when stream is capturing" <log-file>

Adapt the snippet to your framework. The same pattern holds for PyTorch Lightning, Hugging Face Trainer, DeepSpeed, Megatron-LM, and vLLM training wrappers. Where the wrapper exposes a config flag (for examplelr_scheduler_type in Trainer), prefer the flag over the imperative API to keep the schedule declarative and reproducible.

Select the traceback or open the failed terminal, then run Denpex locally to see the initiating rank, collateral failures, exact fix, and verification command without uploading the log.

CUDA reports errors asynchronously, so the traceback usually points at whatever line synchronised next rather than the one at fault. The hub covers every common CUDA error and how to make it report honestly.