Skip to content

Set up installed NVIDIA applications - #12723

Merged
birkskyum merged 3 commits into
omacom:quattrofrom
jdvmi00:spark/nvidia-app-setup-delivery-20260920
Oct 16, 2026
Merged

birkskyum merged 3 commits into
omacom:quattrofrom
jdvmi00:spark/nvidia-app-setup-delivery-20260920

Conversation

@jdvmi00

@jdvmi00 jdvmi00 commented Sep 21, 2026 •

Copy link
Copy Markdown
Contributor

Configure Docker's NVIDIA runtime when AI Workbench is installed, and enable an installed DGX Dashboard service for first boot. Absent applications are left alone; services are not started in the installer. Unlike #3369, this does not install the toolkit on every NVIDIA system.

Isolated checks cover all four Workbench/Dashboard combinations, with and without optional Ollama, installer wiring and failures from runtime configuration and service enablement. Removing the wiring or enabling Ollama fails the tests. The setup has earlier community-image hardware evidence, but this exact revision has not had a fresh installation. Depends on package PRs omacom/omarchy-pkgs#566 and omacom/omarchy-pkgs#567; Docker remains required, alongside the Podman proposal in #11032.

@llstrk

llstrk commented Sep 23, 2026

Copy link
Copy Markdown

Verified: No actionable issue found in the source-reviewed setup at f024da58. The new installer step keeps the two optional applications independent:

Installed package Setup action
nvidia-ai-workbench Request NVIDIA runtime configuration for Docker
dgx-dashboard Enable dgx-dashboard.service for boot
Neither Neither action

Neither branch requests service startup. The linked Workbench package declares Docker and NVIDIA Container Toolkit dependencies; the linked dashboard package installs the named service.


Review information

Test scope: Pinned installer and supporting package-source inspection. Synthetic command-routing checks supplemented the review; native NVIDIA runtime configuration, service enablement, and a fresh Spark installation were not validated.

Community review: Independent automated community review, unaffiliated with the Omarchy team, intended to help prepare PRs for their review.

Automated AI review: Astra Medium coordination, independent Opus 5.5 High and GPT 6 Sol Xhigh technical reviews, followed by evidence reconciliation and an Astra Medium editorial check.

@omarchybot omarchybot added the enhancement New feature or request label Sep 27, 2026
@greptile-apps

greptile-apps Bot commented Oct 4, 2026 •

Copy link
Copy Markdown
Contributor

RetriggerView in GreptileConfidence Score: 5/5

[Medium impact] The PR appears safe to merge; no new actionable issue was established.

Summary

Sets up installed NVIDIA applications without installing them on other machines.

  • AI Workbench configures Docker’s NVIDIA runtime when installed.
  • DGX Dashboard is enabled, then started on deferred first boot.

Reviews (7) · Last reviewed commit: "Start Dashboard during deferred first-bo..." · Reviewed by Greptile

@birkskyum
birkskyum changed the base branch from dragon to n1x October 9, 2026 22:33
@birkskyum
birkskyum force-pushed the spark/nvidia-app-setup-delivery-20260920 branch from 1a88de5 to 9b37f6e Compare October 9, 2026 22:33
@ryanrhughes
ryanrhughes force-pushed the n1x branch 2 times, most recently from dcdea0b to 2f83505 Compare October 10, 2026 04:01
@jdvmi00
jdvmi00 force-pushed the spark/nvidia-app-setup-delivery-20260920 branch from 9b37f6e to bbaa471 Compare October 10, 2026 15:51
@birkskyum

Copy link
Copy Markdown
Contributor

I looked at this against today's n1x, and nvidia-apps-test.sh passes here in an ARM container. Two thoughts.

  1. This runs in hardware setup, so only at install or on an image's first boot. Someone who installs nvidia-ai-workbench or dgx-dashboard later gets neither step. Could both live in the packages themselves (omarchy-pkgs About.desktop fixed font size #566 and Add Discord #567), as an install scriptlet and a preset for the service? Then they run whenever the package arrives, and the runtime needs no change.
  2. nvidia-ctk runtime configure --runtime=docker rewrites /etc/docker/daemon.json, which omarchy-settings ships as a backup file. After that, later changes to Omarchy's Docker defaults land as .pacnew on that machine. That may be fine, but it is worth deciding on purpose.

@jdvmi00

jdvmi00 commented Oct 10, 2026

Copy link
Copy Markdown
Contributor Author

Thanks. Workbench's service starts GPU projects with --runtime=nvidia, and NVIDIA's setup adds that runtime to Docker's daemon.json too, so Workbench on Docker needs the entry, and the .pacnew you point out comes with it. Moving both steps into #566 and #567 as install scriptlets would also cover later installs, at the cost of a package editing omarchy-settings' daemon.json and turning on a service at install. Which would you and Ryan prefer?

@birkskyum

Copy link
Copy Markdown
Contributor

I'd keep both steps in the runtime, as this PR has them, and not move them into the packages. A scriptlet that edits omarchy-settings' daemon.json and turns a service on at install is more than a package should do, and it would run on every machine that installs the package, not only where Omarchy set it up.

For installs after the first one, the Install menu is where Omarchy already does this kind of thing. Install > AI > Ollama picks the CUDA build there on NVIDIA machines. An entry for Workbench and one for the Dashboard, shown on NVIDIA machines, could install the package and run the same step. Then this PR covers an install that brings them in, and the menu covers everything later.

The .pacnew is fine by me as long as it is on purpose. A line in the script saying so would do.

@ryanrhughes may see it differently, so let's hear that too.

@birkskyum
birkskyum force-pushed the spark/nvidia-app-setup-delivery-20260920 branch from bbaa471 to 3a62fcd Compare October 12, 2026 20:53
@jdvmi00
jdvmi00 force-pushed the spark/nvidia-app-setup-delivery-20260920 branch from 3a62fcd to 0b29b17 Compare October 16, 2026 17:13
Comment thread install/hardware/nvidia-apps.sh Resolved
@birkskyum

Copy link
Copy Markdown
Contributor

@jdvmi00 Greptile's first-boot point holds, and install/hardware/bluetooth.sh already handles the same case. On an image's first boot omarchy-provision-hardware sets OMARCHY_IMAGE_DEFERRED_HARDWARE=1, and bluetooth.sh starts its service there with systemctl start --no-block bluetooth.service || true, because a unit enabled after that boot's start job was made waits for the next boot. The same three lines after the Dashboard's systemctl enable cover it.

@birkskyum
birkskyum changed the base branch from n1x to quattro October 16, 2026 20:44
@birkskyum

Copy link
Copy Markdown
Contributor

@jdvmi00 Thanks, the deferred start looks right, and the test passes on Arch Linux ARM. I changed the base to quattro, since the runtime n1x branch went into it with #14820 on 15 October and the edge packages and the ISO are built from quattro. The branch still carries the old n1x history, so the PR shows 54 commits. Your three commits apply cleanly onto today's quattro and the test passes there. Please rebase onto quattro, then it's ready to merge.

@jdvmi00
jdvmi00 force-pushed the spark/nvidia-app-setup-delivery-20260920 branch from 4ec3cb1 to 5db79da Compare October 16, 2026 21:02
@birkskyum
birkskyum merged commit 499c618 into omacom:quattro Oct 16, 2026
2 checks passed
@emirb emirb added this to the 4.5 milestone Oct 17, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

enhancement New feature or request

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants