The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →When Fedora says updates are available but DNF or the Software app shows a different list, the tools are usually reading different snapshots of the same repositories: a stale local metadata cache, a mirror that has not caught up, or a graphical view that has not refreshed. Identify which update system your Fedora installation uses, refresh metadata before changing anything, and only then look at repository configuration. AI model failures are a separate problem. Without the runtime, its version, the model, the GPU and driver stack, and the exact error text, no single cause can be named.
Step one: identify which update system you are running
Fedora does not use one update mechanism across every edition. Conventional installs, such as Fedora Workstation, manage packages with DNF. Image-based variants, such as Fedora Silverblue and Fedora Kinoite, use rpm-ostree, which deploys whole system images. Running DNF repair commands on an rpm-ostree system is the most common way to make a update problem worse, so confirm the system type first.
As an Amazon Associate I earn from qualifying purchases.
| Check | Conventional DNF system | Image-based rpm-ostree system |
|---|---|---|
| Typical editions | Fedora Workstation and other traditional installs | Fedora Silverblue, Fedora Kinoite, and other atomic desktops |
| Confirm the variant | grep VARIANT_ID /etc/os-release |
rpm-ostree status returns deployment details instead of an error |
| See what is pending | sudo dnf check-update |
rpm-ostree status shows the booted and staged deployments |
| Refresh and apply | sudo dnf upgrade --refresh |
rpm-ostree upgrade stages a new deployment; it takes effect after a reboot |
The rest of this article covers DNF-based systems. Image-based systems need their own troubleshooting path, and the Fedora documentation for the atomic desktop you run is the reference to follow.
Free tools Windows power users keep installed
One-click scans. No signup required.
Why the update list differs between tools
Three layers can disagree, and each keeps its own state. A “ghost update” is an update that appears in one interface but not another, or that appears and then fails to install. It is usually one of the layers below.
Stale metadata cache
DNF5 caches data per repository, including repository metadata and the mirror information used to find packages. Its documentation describes --refresh as forcing a metadata update. A cache that is out of date can report packages that have already been superseded, or miss new ones. A successful refresh is strong evidence that the local view was stale. It does not prove that every configured mirror is healthy, so treat it as the first test rather than a verdict.
Mirror selection and slow downloads
Repository definitions point to a mirror list or a metalink, and the resolved mirror list is cached. A cached list can direct downloads to a mirror that is slow, behind on updates, or temporarily unreachable. The telltale pattern is a metadata refresh that completes while a particular package download stalls, times out, or fails its checksum. Re-running the refresh later often resolves it, because the mirror list is refreshed along with the metadata. Editing repository files to force a different mirror is a last step, and you should be able to say which mirror you chose and why.
Graphical software centers
The Software app keeps its own view of available updates and refreshes on its own schedule. It can therefore lag behind a terminal check, or show a stale count after DNF has already installed the packages. Run the refresh in the terminal first, then open the update view in Software and use its refresh or check-for-updates control. If the two still disagree after a successful terminal refresh, record what each one shows before drawing conclusions.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsRefresh sequence for DNF-based systems
- Save the current state. Run
sudo dnf check-updateand copy the complete output, including any warning lines, into a text file. - Run
sudo dnf upgrade --refresh. Expect metadata to download for each enabled repository, followed by a transaction summary. If the command completes and the package list is consistent with what Software now shows, the earlier mismatch came from a stale cache. - If a repository fails to fetch metadata, run
dnf repolistto see which repositories are enabled, and note the repository ID and the exact error. - Only if the refresh still fails, clear the cached metadata with
sudo dnf clean metadataand run the refresh again. Clearing the cache discards only downloaded metadata, so the step is safe to repeat, but it does not fix a bad repository definition. - Reopen Software and compare its list with the output from step 2.
Repository configuration and dependency conflicts
Once metadata is current, a remaining mismatch or failure usually comes from the repository set. Fedora’s upgrade guidance notes that third-party repositories can lag behind Fedora release path changes, particularly before or shortly after a release. A third-party package whose dependencies have not been rebuilt for the current release will fail to resolve even when Fedora’s own repositories are healthy.
- Repository definitions live in
/etc/yum.repos.d/. Compare the enabled repositories against the ones you actually need. - A dependency message that begins with a package name and says it requires something unavailable points to a repository or version gap, not a cache problem.
- Do not reach for
--allowerasingto force a transaction through. It allows DNF to remove packages to resolve a conflict. Read the removal list completely, and if it includes packages you depend on, stop and fix the repository instead.
Release upgrades are a separate operation
A refreshed routine update and a release upgrade are different jobs, and mixing them is a common source of stuck systems. The Fedora upgrade guide the sources point to is versioned for releases 38 through 40 and was last reviewed on April 25, 2024, so use the current Fedora release documentation for the target release’s commands and numbers. The guide describes upgrades spanning at most two releases as officially supported and tested.
- Run
sudo dnf upgrade --refreshand reboot if a new kernel was installed. - Review third-party repositories and disable any that have not published packages for the target release.
- Download the upgrade payload with the system-upgrade plugin, using the target release number from current documentation:
sudo dnf system-upgrade download --releasever=<target>. - Reboot into the offline upgrade with
sudo dnf system-upgrade reboot.
Plugin behavior and release numbers change between Fedora versions, so confirm each command against the current documentation before running it. Image-based systems use their own release-switch procedure and are not covered by these steps.
Rank #4
When an AI model will not load or ignores the GPU
This section is conditional. A model that will not load, or that runs only on the CPU, can have many causes, including memory limits, an unsupported quantization, a driver or backend mismatch, or a runtime that cannot discover the GPU. Without the details below, no particular fix is justified.
- Runtime name and exact version, for example the output of
ollama --versionfor Ollama. - Exact model identifier, including the tag and quantization.
- Fedora release, from
cat /etc/fedora-release. - GPU model and driver stack. On a PCI system,
lspci | grep -iE 'vga|3d'identifies the device. - The exact error text and the runtime’s own log output.
Ollama: confirm where the model is running
For Ollama, ollama ps reports whether a loaded model is running on the GPU, the CPU, or split between them, which tells you whether the GPU was used at all. If Ollama runs as a systemd service, the service log is available through journalctl -u ollama. The upstream Ollama troubleshooting guide covers GPU discovery and runtime diagnostics, but it tracks the main branch, so match its advice to the version you have installed.
Community guidance on runtime and memory settings
The Fedora AI/ML special interest group wiki includes runtime and memory configuration material. It is maintained by the community rather than Fedora’s release process, so check any flags against the current upstream documentation for your runtime before applying them.
Match the symptom to the evidence
| Symptom | Collect first |
|---|---|
| Load fails with a memory error | Model size and quantization, system RAM, and GPU memory reported at load time |
| Model runs, but only on the CPU | GPU model, installed driver, and the runtime’s GPU discovery lines in the log |
| Model worked before a Fedora upgrade and fails now | Runtime version before and after, driver version before and after, and whether the driver was rebuilt |
Use the table to decide what to record, not as a list of fixes. A setting that worked for someone else on different hardware is not evidence about your system until your own logs show the same failure.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
