On a host that meters Git LFS, why does storage keep growing after files are deleted, and what drives bandwidth use?
answer
- every pushed version is whole
- history is not the meter
- downloads land on the owner
- forks and CI count too
basics
~20 sEach pushed LFS version is stored whole and stays after files are deleted; on GitHub only deleting the repository or a support purge frees it. Every download, CI and forks included, counts against the owner's bandwidth.
solid answer
~50 sLFS stores no deltas: every version you push is a complete object, so a 500 MB scene changed by one byte and pushed twice uses 1 GB. Removing the file, untracking it or rewriting history leaves the objects on the server; GitHub says they keep counting until the repository is deleted and recreated or support purges them. Bandwidth is the other meter: on GitHub every LFS download counts against the repository owner, including CI jobs, pulls through forks and source archives that include LFS objects, while uploads count only toward storage. Over the included quota, a zero budget blocks LFS for the rest of the month; without a payment method, exceeding storage leaves clones with pointers only and blocks new pushes, and exceeding bandwidth disables LFS until next month. Other hosts set their own quotas.
go deeper
Remember the two meters, storage and bandwidth, and that every pushed version of an LFS file is stored whole.
Explain why deleting or untracking a file does not reduce storage, and which activities consume bandwidth on a host such as GitHub.
Estimate a pipeline's LFS bandwidth from clone rate and checkout size, cut it, and know what the host does at the quota with and without a budget.
Own the cost model: compare host quotas, self-hosted LFS storage and changes to how binaries are produced, and set guardrails before a bill surprises the team.
## Two meters: storage and bandwidth Hosts that offer Git LFS usually meter two things: how much LFS **storage** a repository's objects occupy, and how much LFS **bandwidth** downloads consume. Git LFS itself defines no quotas; the Batch API only reserves optional status codes for a server that runs out (507 for storage capacity, 509 for a bandwidth limit). Every host sets its own numbers and rules. GitHub's documentation, read on 2026-10-06, is the worked example here. | Event (GitHub) | Storage | Bandwidth | |---|---|---| | push a new LFS file or version | whole file added to the owner's storage | not measured | | download an LFS file (clone, pull, checkout) | none | counted against the owner | | a CI job downloads LFS files | none | counted against the owner | | someone pulls through a fork | none | counted against the parent repository | | a source archive including LFS objects is downloaded | none | counted against the repository | ## Why storage only grows - **No deltas.** Each pushed version of an LFS file is a complete object. GitHub's example: push a 500 MB file, change one byte, push again, and you use 1 GB of storage. - **History is not the meter.** Storage counts all LFS objects associated with the repository, regardless of when they were uploaded. Deleting a file, untracking it, or rewriting history to remove its pointers leaves the objects on the server. - **Reclaiming is a host operation.** GitHub's documented way to remove LFS objects is to delete and recreate the repository (which also deletes issues, stars and forks), or to ask support to purge them. - **Mid-month deletion does not reduce that month.** GitHub does not recalculate storage within the month; storage usage resets at the start of the next one. ## What drives bandwidth Bandwidth follows **downloads**, and automation is usually the biggest downloader. A pipeline cloning a repository forty times an hour, with 2 GB of LFS content in the checked-out tree, transfers about 80 GB an hour, nearly 2 TB a day, all on the owner's meter. Developers re-downloading after pruning, new hires cloning and forks pulling add to it. Uploads do not count as bandwidth on GitHub, and anyone with write access can push LFS files without touching their own account's quota. ## What happens at the quota on GitHub 1. Each account includes free storage and bandwidth; GitHub's billing page lists 10 GiB of each for Free, Pro and Free for organizations, and 250 GiB for Team and Enterprise Cloud. 2. Bandwidth resets each billing cycle; storage is metered hourly and converted to GiB-months. 3. With a **budget of zero**, overage is not charged but LFS is blocked for the rest of the month. 4. With **no payment method**, exceeding storage means clones still work but return pointer files only, and new files cannot be pushed; exceeding bandwidth disables LFS on the account until next month. 5. With a payment method and no budget limit, overage is billed per GiB downloaded and per GiB-month stored. 6. Email alerts arrive at 90% and 100% of the included usage. ## Turning hours into a charge GitHub meters storage above the included amount hourly and converts it to **GiB-months**. Its own worked example: 1. 1 GiB over the quota for the first 15 days of April: 1 × 15 × 24 = 360 GiB-hours. 2. 2 GiB over for the remaining 15 days: 2 × 15 × 24 = 720 GiB-hours. 3. Total 1,080 GiB-hours, divided by 720 hours in the month, is 1.5 GiB-months billed. Bandwidth is simpler: each GiB downloaded above the included amount is billed. The billing page is the place to read current rates; they change, so a plan should be expressed in GiB stored and GiB downloaded, not in a remembered price. ## Planning the bill - **Stop CI downloading what it never opens**: skip the smudge step, then pull only needed paths. - **Keep a cache of the local LFS store** on build agents so repeated jobs reuse objects. - **Avoid churn on huge files**: re-exporting a multi-gigabyte asset on every small change multiplies storage. - **Watch forks and archives**: both draw on the parent repository's bandwidth. - **Read the host's own rules**: other hosts and self-hosted LFS servers set their own quotas and rules.
- How do you actually free LFS storage on GitHub after removing large files from history?Rewriting history removes the pointers, not the objects. GitHub's documented route is to delete and recreate the repository, which also deletes its issues, stars and forks, then push back what you want to keep; if deleting is not possible, support can purge specific objects.
- Why can the LFS bandwidth bill jump when a fork's contributors become active?On GitHub, pulling LFS content through a fork counts against the parent repository's bandwidth, and pushing LFS objects to a fork counts against the parent's quotas, not the fork owner's.
- How does GitHub turn storage over the quota into a charge?It meters hourly usage and divides by the hours in the month. Being 1 GiB over for 15 days and 2 GiB over for the next 15 gives 360 plus 720 GiB-hours, 1,080 GiB-hours, or 1.5 GiB-months for that month.
- Deleting a large file and pushing frees its LFS storage.
- Git LFS stores only the changed bytes of each new version.
- Uploading to LFS uses the pusher's own bandwidth quota.
- CI downloads from a hosted LFS store do not count as bandwidth.
- Pulls through a fork are billed to the fork owner's account.