FSDP All-Gather Timeout
FSDP all-gather timeouts stall training when sharded parameters cannot be collected from distributed ranks.
FSDP all-gather timeouts stall training when sharded parameters cannot be collected from distributed ranks.
What this failure is
FSDP All-Gather Timeout is a Distributed Training failure seen during ML training runs. FSDP all-gather timeouts stall training when sharded parameters cannot be collected from distributed ranks. Common tags: Fsdp, All Gather, Timeout, Communication.
Is this what broke your run? Paste your log.
You're reading about FSDP All-Gather Timeout. Paste your own crash log or traceback below and get the real root cause for YOUR run, not this generic entry. No account, no card. Logs are masked at ingress and never saved to account history.
Want 14 days on the Scale plan?
Request an evaluation code. A verified workplace organization activates up to 50 diagnoses a day, alerts, history, and follow-up questions. No credit card or automatic subscription.
Why it happens (the mechanism)
A slow rank (straggler) fails to send its shard parameters within the all-gather timeout. CPU-memory offload latency variability: when FSDP offloads parameters to CPU, a single rank can exceed the all-gather timeout if its CPU memory is busy with swap or other processes. NCCL IB fabric congestion delays the all-gather communication on specific node pairs. Uneven model sharding: if the model has asymmetric layers, ranks with larger shards take longer to gather, causing timeout. Taken together, these mechanisms explain why the failure is reproducible, why it tends to surface on specific workloads or scales, and why generic mitigation attempts often fall short without addressing the underlying cause.
What you'll observe
- Training stalls at the first step after FSDP wrapping
- GPU utilization drops to 0% on some ranks while others are at 100%
- A timeout error appears after a specific number of steps or batches
Common symptoms and what they mean
| Symptom | Why it happens |
|---|---|
| RuntimeError: AllGather timed out waiting for work on rank X | A slow rank (straggler) fails to send its shard parameters within the all-gather timeout |
| FSDP forward pass hangs at param all-gather: wait() never completes on rank Y | CPU-memory offload latency variability: when FSDP offloads parameters to CPU, a single rank can exceed the all-gather timeout if its CPU memory is busy with swap or other processes |
| NCCL timeout error during FSDP parameter shard collection | NCCL IB fabric congestion delays the all-gather communication on specific node pairs |
| Rank X shows CUDA memory 0 bytes allocated while other ranks have full utilization | Uneven model sharding: if the model has asymmetric layers, ranks with larger shards take longer to gather, causing timeout |
Which systems are affected
- Multi-node FSDP training across 2+ nodes
- FSDP with CPU offload for parameters
- FSDP with gradient checkpointing enabled
- Heterogeneous GPU setups (different GPU models or memory sizes across ranks)
How to confirm this is the problem
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.
- ✓Reproduce the failure from a clean checkpoint/seed: the symptom must appear without warm-up state from a previous run.
- ✓Verified signal present: RuntimeError: AllGather timed out waiting for work on rank X
- ✓Verified signal present: FSDP forward pass hangs at param all-gather: wait() never completes on rank Y
- ✓Verified signal present: NCCL timeout error during FSDP parameter shard collection
- ✓Verified signal present: Rank X shows CUDA memory 0 bytes allocated while other ranks have full utilization
- ✓A targeted fix from the "How to fix it" section eliminates or substantially reduces the symptom within one validation pass.
The fix and the prevention pattern
The root cause is on this page and stays free. A free account adds the exact remediation steps, saved history, and the fix on every entry in the encyclopedia.
Sign up free. Unlock the full analysisNo credit card. Daily allowance follows verified trust tier. Instant access.
Diagnose this failure in VS Code
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.
Install the free VS Code extensionRelated failures to investigate next
Root cause
- A slow rank (straggler) fails to send its shard parameters within the all-gather timeout
- CPU-memory offload latency variability: when FSDP offloads parameters to CPU, a single rank can exceed the all-gather timeout if its CPU memory is busy with swap or other processes
- NCCL IB fabric congestion delays the all-gather communication on specific node pairs
- Uneven model sharding: if the model has asymmetric layers, ranks with larger shards take longer to gather, causing timeout
The fix and how to prevent it
Evaluate Denpex on your own logs
Request a Scale evaluation code. A verified workplace organization activates 14 days with up to 50 diagnoses a day. Every account keeps its current diagnosis allowance and gets a verification path. No card or automatic subscription.
Don't just read the fix, diagnose your run
The encyclopedia tells you what went wrong. Denpex tells you what went wrong in YOUR training run. With your logs, your config, and your stack.