CUDA out of memory during Inductor CUDA graph execution allocator private memory pool fragmentation

TorchInductor exhausted a private CUDA Graph memory pool even though total free memory may appear sufficient. Captured graphs retain stable addresses, so inactive blocks in the private pool cannot always satisfy a differently shaped allocation. This entry explains how to confirm the cause, apply the fix, and separate it from adjacent torch-compile failures.

The literal signature is "CUDA out of memory during Inductor CUDA graph execution allocator private memory pool fragmentation". It is a environment failure associated with torch.compile, TorchInductor, and Triton JIT. The line identifies the failing operation or subsystem, while the surrounding evidence decides whether it is the initiating fault or a downstream symptom.

TorchInductor exhausted a private CUDA Graph memory pool even though total free memory may appear sufficient. Captured graphs retain stable addresses, so inactive blocks in the private pool cannot always satisfy a differently shaped allocation. 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 out of memory during Inductor CUDA graph execution allocator private memory pool fragmentation".
• 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.

• torch.compile, TorchInductor, and Triton JIT
• production-shaped multi-accelerator workloads
• containerized and bare-metal deployments of the same stack

Apply the following checklist to a small reproduction: each box below is a positive signal that you are looking at this exact failure rather than a sibling in the same taxonomy.

• ✓Find the first occurrence of "CUDA out of memory during Inductor CUDA graph execution allocator private memory pool fragmentation" and preserve at least 100 lines before it.
• ✓Identify which rank, node, device, or process emitted the first related warning.
• ✓record reserved, allocated, and inactive split bytes at failure. Check whether new input shapes cause additional graph captures and private pools.
• ✓Repeat the same input after the targeted change and require the signature to disappear.
• ✓Resume from job start after graph capture and shape policy are corrected only after the control passes.

CUDA out of memory during Inductor CUDA graph execution allocator private memory pool fragmentation

Timestamps and exact values vary across runs, but the pattern. An info-level start, an early WARN, an ERROR carrying the symptom. Is the actual fingerprint you should alert on. The Denpex platform flags this combination automatically.

reproduce with CUDA Graphs disabled for the compiled region. If the same batch succeeds, reduce shape variation or recapture scope instead of only lowering batch size. This changes the condition in the causal diagnosis instead of hiding the outer exception. The repeated control proves ownership before recovery from job start after graph capture and shape policy are corrected.

# 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 out of memory during Inductor CUDA graph execution allocator private memory pool fragmentation" <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.

• TorchInductor exhausted a private CUDA Graph memory pool even though total free memory may appear sufficient. Captured graphs retain stable addresses, so inactive blocks in the private pool cannot always satisfy a differently shaped allocation.
• The decisive evidence is the first log line that precedes "CUDA out of memory during Inductor CUDA graph execution allocator private memory pool fragmentation" and differs from a healthy run.
• A nearby failure remains a competing hypothesis until a control separates configuration, capacity, transport, and hardware causes.