Install the Vulkan ICD when an app silently refuses to launch
Checked on this Omarchy 4.0.2-1 workstation against both cited sources and the upstream tree at tag v4.0.3. What held: the Arch wiki's Vulkan page supports every generic claim in the record, the vulkan-icd-loader plus vendor driver split, vulkan-tools being where vulkaninfo and vkcube live, ls /usr/share/vulkan/icd.d/ as the check, and MESA_VK_DEVICE_SELECT requiring vulkan-mesa-implicit-layers with a trailing ! to enforce. Every package the record names still exists: I queried the Arch package JSON API for each one and all resolved, vulkan-mesa-implicit-layers at 26.2.2-1 in extra included. I confirmed on this machine that vulkan-tools is not installed (no vulkaninfo or vkcube on PATH), so that step is genuinely needed on Omarchy.
The cause is wrong for Omarchy 4 and that is the substance of this verdict. The record says the missing vendor driver is "a very common gap on Intel iGPU laptops". Issue 1441 is CLOSED, and reading it in full shows why: it is the thread where contributors proposed an auto-detecting installer script, and Omarchy 4 now ships it. I read /usr/share/omarchy/install/hardware/vulkan.sh here and fetched it at tag v4.0.3, and the two are identical. It installs vulkan-intel, vulkan-radeon or vulkan-asahi per detected vendor, install/hardware/nvidia.sh installs nvidia-utils for the NVIDIA ICD, and install/hardware/all.sh calls both. So this is sound generic Arch advice mis-specialised to Omarchy, exactly the shape the brief warns about. I rewrote the cause to branch Omarchy 4 against plain Arch and to name the three cases that still reach it on Omarchy 4: the hardware scripts run at install time only and no migration re-runs them, which I checked by grepping the migrations directory, the (VGA|Display).*vendor match can miss a device, and 32-bit Vulkan is untouched by it.
I rewrote danger, which was the weakest field. It warned only that lib32-* needs multilib, and then described the failure as harmless, which makes it a note rather than a danger. On Omarchy that warning does not even apply: multilib is uncommented at line 25 of both /etc/pacman.conf here and Omarchy's shipped default/pacman/pacman-stable.conf. The real hazard is the record's own fix block, which lists Intel, AMD and NVIDIA packages in a way a reader can install wholesale. The wiki documents that a stray nvidia_icd.json makes Chromium and Electron applications hang for about 30 seconds before starting, which is indistinguishable from the symptom this record treats.
I also added what the record omitted and Omarchy provides: omarchy-hw-vulkan as the one-line ICD check, omarchy-install-gaming-gpu-lib32 for the 32-bit set, vulkan-nouveau and vulkan-swrast, and the fact that pre-GSP NVIDIA cards need the AUR 580xx branch. No cited URL needed removing. Both resolve and both support what the record claims, issue 1441 for the Omarchy 3 era and the wiki for plain Arch, so sources_remove is empty. On the ALPM guard: I read /usr/bin/omarchy-update-pacman-guard and it aborts only when the parsed pacman arguments carry both a sync and a sysupgrade flag, so none of the record's pacman -S --needed commands is blocked. The thread's own sudo pacman -Syu vulkan-radeon would be, and I warn about it.
I dropped frequency from common to occasional for the same reason the cause changed. Not exercised: this workstation is NVIDIA-only and carries just nvidia_icd.json in /usr/share/vulkan/icd.d/, so I could not test the Intel or AMD paths, could not run vulkaninfo or vkcube (vulkan-tools absent, no sudo to install it), and did not reproduce the 30 second Chromium delay, which is taken from the wiki rather than observed here.
Install only the ICD that matches your GPU. Two vendor ICDs registered at once is its own bug rather than a safety net: the loader reads every file in /usr/share/vulkan/icd.d/, and the Arch wiki records that Vulkan tries the NVIDIA driver before the Radeon one when nvidia_icd.json is present, which can leave Chromium and Electron applications hanging until a 30 second timeout before they start. That looks exactly like the symptom this record is meant to cure. If you already installed the wrong one, install the correct ICD first and remove the wrong one second, because Steam depends on the lib32-vulkan-driver provision the wrong package is currently satisfying.
lib32-* packages come from the multilib repository. It is enabled by default in Omarchy's shipped pacman.conf, so the "target not found" failure only happens on plain Arch, where it is a clean refusal rather than anything dangerous.
Symptom
An application that uses Vulkan (Zed, some Electron builds, games, vkcube) fails to start on a fresh install with no useful message, or exits immediately. vulkaninfo errors out or lists no physical devices.
Cause
The Vulkan ICD loader is installed but no vendor Vulkan driver is, so the loader finds no physical device and the application exits.
**On Omarchy 4 this gap is closed at install time.** /usr/share/omarchy/install/hardware/vulkan.sh reads lspci and installs vulkan-intel, vulkan-radeon or vulkan-asahi for each detected vendor, and install/hardware/nvidia.sh installs nvidia-utils, which carries the NVIDIA ICD. Issue 1441, which reported Zed failing to launch on a fresh Intel install, is closed and that script is what closed it.
Three cases still reach this record on Omarchy 4. The hardware scripts run only during installation and omarchy update does not re-run them, so a GPU added after install is not covered. The vendor match is lspci | grep -iE "(VGA|Display).*(Intel|AMD|Apple)", so a device that reports neither a VGA compatible controller nor a Display controller class, or that names its vendor differently, is skipped. And 32-bit Vulkan for Steam, Proton and Wine is a separate set of packages that the install-time script does not touch.
On **plain Arch, EndeavourOS and Manjaro** the original cause stands as written: installing vulkan-icd-loader alone gives you no driver.
Fix
**On Omarchy 4, check what is already there first.**
ls /usr/share/vulkan/icd.d/
pacman -Q vulkan-intel vulkan-radeon vulkan-asahi nvidia-utils 2>&1
An empty icd.d is the real failure. Omarchy ships omarchy-hw-vulkan, which tests exactly that and is the quickest check:
omarchy-hw-vulkan && echo "an ICD is registered" || echo "no ICD"
**Install the tools, which Omarchy does not ship.** vulkaninfo and vkcube live in vulkan-tools:
sudo pacman -S --needed vulkan-tools
vulkaninfo | head -30
vkcube
**Install the driver for the GPU you actually have.** Install one vendor's driver, not all of them:
# Intel iGPU or Arc:
sudo pacman -S --needed vulkan-intel
# AMD:
sudo pacman -S --needed vulkan-radeon
# Apple silicon:
sudo pacman -S --needed vulkan-asahi
# NVIDIA proprietary, Turing and newer: the ICD ships inside nvidia-utils
sudo pacman -S --needed nvidia-utils
# NVIDIA Maxwell, Pascal and Volta are on the frozen 580 branch:
# nvidia-580xx-utils, from the AUR
# NVIDIA open-source NVK instead of the proprietary driver:
sudo pacman -S --needed vulkan-nouveau
On a machine with no usable GPU, such as a VM, install the software rasteriser:
sudo pacman -S --needed vulkan-swrast
None of these is blocked by Omarchy's ALPM guard, which aborts only a transaction carrying both -S and -u. Do not follow the sudo pacman -Syu vulkan-radeon form that appears in the upstream issue thread. On Omarchy the guard refuses it, and on any Arch system pacman -Sy <pkg> is a partial upgrade.
**32-bit Vulkan for Steam, Proton and Wine.** Omarchy has a helper that detects your GPUs and installs the matching set, which is safer than picking by hand:
omarchy-install-gaming-gpu-lib32
The manual equivalent, matching your vendor:
sudo pacman -S --needed lib32-vulkan-icd-loader
sudo pacman -S --needed lib32-vulkan-intel # Intel
sudo pacman -S --needed lib32-vulkan-radeon # AMD
sudo pacman -S --needed lib32-nvidia-utils # NVIDIA, must match nvidia-utils exactly
multilib is already enabled in Omarchy's shipped /etc/pacman.conf. On plain Arch, enable it first:
grep -A1 '^\[multilib\]' /etc/pacman.conf
sudo sed -i '/^#\[multilib\]/,+1 s/^#//' /etc/pacman.conf
sudo pacman -Syu
**On a multi-GPU box you can force a specific Mesa device.** This needs vulkan-mesa-implicit-layers and works only for Mesa drivers, not for the NVIDIA proprietary ICD:
sudo pacman -S --needed vulkan-mesa-implicit-layers
MESA_VK_DEVICE_SELECT=list vulkaninfo
MESA_VK_DEVICE_SELECT=<vendorID>:<deviceID>! <command>
The trailing ! enforces the selection.
Verify
ls /usr/share/vulkan/icd.d/ lists a JSON file for your vendor. vulkaninfo | grep deviceName names your GPU rather than reporting no physical devices. vkcube renders a spinning cube. The application that would not start now launches, and starts promptly rather than after a pause of about 30 seconds.
Sources
- https://github.com/basecamp/omarchy/issues/1441
- https://wiki.archlinux.org/title/Vulkan
- https://github.com/omacom/omarchy/blob/v4.0.3/install/hardware/vulkan.sh
- https://github.com/omacom/omarchy/blob/v4.0.3/install/hardware/nvidia.sh
- https://github.com/omacom/omarchy/blob/v4.0.3/install/hardware/all.sh
- https://github.com/omacom/omarchy/blob/v4.0.3/bin/omarchy-install-gaming-gpu-lib32
- https://github.com/omacom/omarchy/blob/v4.0.3/bin/omarchy-hw-vulkan
- https://archlinux.org/packages/extra/x86_64/vulkan-mesa-implicit-layers/
- https://archlinux.org/packages/extra/x86_64/vulkan-tools/