Zero-Residue Migration¶

Flexium's zero-residue migration ensures that when a training job migrates from one GPU to another, no memory is left behind on the source GPU.

How It Works¶

Traditional approaches (like model.to(device)) leave memory fragments due to PyTorch's caching allocator. Flexium uses driver-level migration (requires driver 580+):

Capture - Complete GPU state is captured at driver level
Release - Source GPU is completely freed (0 MB)
Restore - State is restored on target GPU

Benefits¶

Full GPU Reclamation - The source GPU is immediately available for other workloads
No Memory Fragmentation - Clean memory state on both source and target GPUs
Seamless Continuation - Training resumes exactly where it left off

Usage¶

Zero-residue migration is automatic:

import flexium.auto

with flexium.auto.run():
    # Your training code here
    # Migration happens when triggered via dashboard
    train_model()

Verification¶

You can verify zero-residue behavior by monitoring GPU memory:

# Before migration
nvidia-smi

# After migration - source GPU should show 0 MB used by the process
nvidia-smi