Memory Tuning

When a training run OOMs or is close to the memory limit:

1. Set PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True first. This fixes fragmentation-induced OOM with zero performance cost. Most Slurm launch templates already include it.
2. Add selective activation recompute (recompute_modules=[core_attn]) if not already enabled. See skills/perf-techniques/activation-recompute/SKILL.md.
3. Avoid increasing TP as a memory fix — doubling TP dramatically increases NVLink all-reduce volume and often kills throughput (-28% on Llama3 70B).
4. Avoid increasing PP at the cost of DP — halving DP doubles gradient accumulation steps and hurts throughput (~6%).
5. Consider mlp recompute if still OOM. Saves ~3 GB but costs ~16% GPU utilization on large dense models (Llama3 70B).
6. CPU offloading is blocked when PP > 1.

Enablement

Expandable segments (recommended first step)

Set in the job's environment before launching:

export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True

In Slurm scripts this is typically placed alongside other env vars:

export CUDA_DEVICE_MAX_CONNECTIONS=1
export NVTE_ALLOW_NONDETERMINISTIC_ALGO=1
export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True

No model config changes needed. Zero throughput cost.

Parallelism resizing

If the model genuinely does not fit (not fragmentation), adjust parallelism:

Activation recompute

See skills/perf-techniques/activation-recompute/SKILL.md for full details.

CPU offloading

cfg.model.cpu_offloading = True

Incompatible with PP > 1. Only usable when pipeline_model_parallel_size = 1.

A Note on VPP

Virtual pipeline parallelism (VPP) is primarily a throughput optimization that reduces pipeline bubble overhead by interleaving smaller model chunks. Its effect on peak memory is minimal — changing VPP does not meaningfully change the total activation, parameter, or optimizer memory on a GPU.

In earlier experiments we incorrectly attributed an OOM fix to VPP tuning (VPP 5→10). The actual fix was PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True which eliminated memory fragmentation. The VPP=10 run actually used slightly more peak memory (60.2 GB vs 58.8 GB) but did not OOM because expandable segments prevented fragmentation.

VPP should be tuned for pipeline bubble reduction (see docs/parallelisms.md), not as a memory fix.

Compatibility and Constraints

• expandable_segments:True is incompatible with --use-nccl-ub (NCCL user-buffer registration). See Megatron-FSDP docs.
• When using CUDA graphs with expandable_segments:True, set NCCL_GRAPH_REGISTER=0 (required on pre-Blackwell GPUs, enforced by MCore CudaGraphManager).
• CPU offloading requires pipeline_model_parallel_size = 1.
• Distributed optimizer requires use_distributed_optimizer = True in the optimizer config.

Measured Results

Llama3 70B SFT on 32x H100 80GB, FP8 (Current Scaling):

• Baseline: TP=4, PP=4, VPP=5, DP=2, MBS=1, GBS=32, seq_len=4096
• Golden GPU utilization: 709.93 TFLOP/s/GPU
• Regression threshold: 5%

Strategy comparison: parallelism changes for memory reduction

Key takeaways:

• expandable_segments:True is the winner. The baseline OOM was caused by memory fragmentation, not insufficient capacity. Setting this env var eliminated the OOM with zero throughput cost and no parallelism changes.
• PP=8 works for memory but loses DP (2→1), meaning 32 gradient accumulation steps per batch, which hurts throughput by ~6%.
• TP=8 is catastrophic (-28%) because doubling TP increases all-reduce communication volume proportionally across NVLink, and DP=1 means no micro-batch overlap.

CPU offloading: blocked

ValueError: Currently there is no support for Pipeline parallelism with CPU offloading. This approach is blocked for any model using PP > 1.

Activation recompute: expensive alternative

Selective activation recompute with mlp saved ~3 GB peak memory but cost ~16% GPU utilization on this workload. See skills/perf-techniques/activation-recompute/SKILL.md for full results.

Code Anchors

CPU offloading PP incompatibility (MCore)

        if self.cpu_offloading and self.pipeline_model_parallel_size > 1:
            raise ValueError(
                "Currently there is no support for Pipeline parallelism with CPU offloading"
            )

VPP config and layer divisibility validation (MCore)

            if pipeline_parallel_size and self.virtual_pipeline_model_parallel_size is not None:
                num_layers_per_middle_pipeline_rank = num_layers // pipeline_parallel_size
                if (
                    not num_layers_per_middle_pipeline_rank
                    % self.virtual_pipeline_model_parallel_size
                    == 0
                ):
                    raise ValueError(
                        f"number of layers on each middle pipeline rank:"
                        f"{num_layers_per_middle_pipeline_rank} must be divisible by virtual"
                        f"pipeline parallel degree {self.virtual_pipeline_model_parallel_size}"
                    )

Parallelism docs on interleaved pipeline schedule

To minimize the pipeline bubble, the computation on each GPU can be divided into multiple subsets of layers (referred to as model chunks), rather than a single contiguous block. Enable this by setting `virtual_pipeline_model_parallel_size`:

model_config = GPTModelProvider(
    pipeline_model_parallel_size=4,
    virtual_pipeline_model_parallel_size=2,  # 2 model chunks per pipeline stage
    # ... other model parameters
)

Failure Diagnosis

Known Limitations

• expandable_segments:True is incompatible with NCCL user-buffer registration
• CPU offloading is blocked when PP > 1
• Parallelism resizing (TP/PP) often has significant throughput costs
• No automatic memory profiling to recommend the optimal strategy

Verification

Quick check that expandable_segments:True is active:

import os
assert "expandable_segments:True" in os.environ.get("PYTORCH_CUDA_ALLOC_CONF", "")

For Slurm jobs, verify the env var is exported before the training command in the launch script.