Git LFS Storage Calculator
Estimate Git LFS storage footprint and monthly bandwidth usage from tracked object count, average size and pull volume.
Inputs
Number of distinct large-file versions tracked via Git LFS.
Average size of one LFS-tracked object (e.g. a design file, model, or media asset).
Number of times LFS objects are downloaded per month (CI runs, clones, checkouts).
LFS Storage Footprint
7.32GB
Monthly LFS Bandwidth
4.39GB
LFS Storage Footprint
7,500MB
Monthly LFS Bandwidth
4,500MB
Step by step
Storage: objects × avg size
500 × 15MB
= 7500MB
Monthly bandwidth: pulls × avg size
300 × 15MB
= 4500MB
How it works
Git LFS bills storage and bandwidth separately from repository size. Formula: storage = lfs_objects × avg_size; monthly_bandwidth = monthly_pulls × avg_size. Every clone, fetch or checkout that touches an LFS-tracked file re-downloads it (unless cached), so bandwidth usage scales with team size and CI activity, not just repository growth.
Formulas
LFS storage
storage_GB = lfs_objects × avg_size_MB / 1024
- lfs_objects
- Number of LFS-tracked objects
- avg_size_MB
- Average object size (MB)
Monthly bandwidth
bandwidth_GB = monthly_pulls × avg_size_MB / 1024
- monthly_pulls
- LFS object downloads per month
- avg_size_MB
- Average object size (MB)
Frequently Asked Questions
Why is my LFS bandwidth bill higher than expected?
Every CI job that clones the repo re-downloads all LFS objects referenced by the checked-out commit unless you cache the LFS store between runs — this is the most common cause of runaway bandwidth usage.
How do I reduce LFS bandwidth costs?
Cache `.git/lfs` between CI runs, use `git lfs fetch --recent` to skip old history, and consider `GIT_LFS_SKIP_SMUDGE=1` for CI jobs that don't need the actual file content.
Does deleting an LFS-tracked file reduce storage?
No — like regular Git objects, LFS objects remain in history until you rewrite it (e.g. `git lfs prune` for local caches, or provider-specific history rewriting for the remote store).
What files should go in Git LFS?
Binary or large files that change often and don't diff well as text — images, videos, datasets, compiled models, design files — so history doesn't balloon with full copies of every version.