DGX OS Software Stack

This appendix describes the NVIDIA DGX OS and BaseOS software stack.

This software stack is a validated set of packages with compatible versions of the GPU driver, DOCA-OFED, and related DGX platform packages for each release. To see the current versions of the listed components, refer to Current Software Versions.

BaseOS uses metapackages to install and maintain groups of related packages as one supported unit. A metapackage does not usually provide software files itself but declares dependencies on packages that include those files.

Common and DGX-specific Packages

The following sections describe common and DGX-specific metapackages in BaseOS 9 and list their direct dependencies where practical. For metapackages with large dependency sets, use the provided apt commands to view the dependencies for the installed release. Package dependencies can change between releases.

nvidia-system-core

Packages	Description
baseos-cli	Provides BaseOS release management commands.
baseos-release	Provides BaseOS release metadata.
ipmitool	Provides IPMI management utilities.
nvidia-nvme-options	Automatically enables NVMe Interrupt Coalescing at boot on all Samsung, Kioxia, and Micron drives.
nvme-cli	Provides the NVMe management command-line interface.
nvidia-tuned-profiles	Supplies the platform performance and crash-dump profiles. It applies several settings that earlier releases configured through individual packages. For profile management information, see Managing and Customizing TuneD Profiles in the NVIDIA BaseOS User Guide.
nv-grubmenu	Makes the GRUB menu visible during boot.
tpm2-tools	Provides tools for managing TPM 2.0 devices.

nvidia-system-mlnx-drivers

Package	Description
mlnx-pxe-setup	Configures Mellanox PXE boot support.
nvidia-mlnx-config	Configures applicable Mellanox adapters during system configuration. It discovers adapters and applies PCIe MRRS and MAX_ACC_OUT_READ settings when the platform configuration requires them. The implementation uses the adapter BDF and does not use a fixed setpci vendor-ID command. The package provides /usr/bin/nvidia-mlnx-config.sh and its systemd services.
mlnx-nvme-dkms	Builds the Mellanox NVMe peer-memory kernel module.
mlnx-nfsrdma-dkms	Builds the Mellanox NFS-over-RDMA kernel module.
mlnx-fw-updater	Updates Mellanox adapter firmware.

For applicable DGX platforms, the DGX OS 8.0.0 ISO installs the nvidia-system-mlnx-drivers metapackage with the release-qualified doca-ofed metapackage. nvidia-system-mlnx-drivers installs the DGX-specific MLNX driver support packages in the preceding table, while doca-ofed installs DOCA-OFED and its dependencies. Together, they provide the DOCA-OFED stack used by the release.

nvidia-system-utils

Package	Description
chrpath	Provides build-path inspection.
cifs-utils	Provides CIFS file-system support.
fping	Provides network reachability testing.
iperf	Provides network performance testing.
lsscsi	Lists SCSI devices.
net-tools	Provides legacy network utilities.
nfs-common	Provides NFS client support.
nvidia-container-toolkit	Enables NVIDIA GPUs in container runtimes.
openssh-server	Provides the OpenSSH server.
perftest	Provides RDMA performance-testing utilities.
pm-utils	Provides power-management utilities.
powercap-utils	Provides power-cap management utilities.
quota	Provides file-system quota utilities.
rasdaemon	Provides RAS event monitoring.
samba-common	Provides common Samba files for file sharing.
scsitools	Provides SCSI utilities.
shim-signed	Provides the signed Secure Boot shim.
smartmontools	Provides storage-health monitoring utilities.
vlan	Provides VLAN configuration support.

nvidia-system-extra

Package	Description
automake	Provides build-system generation tools.
bison	Provides a parser generator.
build-essential	Provides the essential compiler and build-tool set.
cmake	Provides a cross-platform build system.
flex	Provides a lexical-analyzer generator.
gdb	Provides a source-level debugger.
git	Provides source-control tools.
libelf-dev	Provides ELF development libraries.
libltdl-dev	Provides libtool dynamic-loader development files.
m4	Provides a macro processor.
python3-venv	Provides Python virtual-environment tools.
swig	Provides interface code-generation tools.

nvidia-system-extra-localization

Packages	Description
witalian, wngerman, wogerman, wportuguese, wspanish, and wswiss	Provide Italian, German, Portuguese, Spanish, and Swiss spell-checking dictionaries.

nvidia-system-diag

Package	Description
busybox	Provides compact system utilities.
fio	Provides storage I/O testing.
i2c-tools	Provides I2C diagnostic utilities.
libmuparser2v5	Provides an expression-parser library used by diagnostic tools.
libnbd0	Provides the Network Block Device client library.
minicom	Provides serial-console communication.
python3-pandas	Provides Python data-analysis libraries.
python3-tabulate	Provides Python table-formatting libraries.
sshpass	Provides scripted SSH password entry.
stressapptest	Provides memory stress testing.
systemd-container	Provides systemd container utilities.
systemd-coredump	Provides system crash-dump collection.

dgx-system-utils

Package	Description
dgx-release	Updates DGX OS release information.
nv-common-apis	Installs commonly used NVIDIA system scripts.
nv-grubserial	Displays the GRUB menu over a serial console.
nv-limits	Increases file limits.
nv-update-disable	Disables OS update prompts.
nvgpu-services-list	Lists GPU-related services.
nvidia-dgx-setup	Performs DGX system setup.
nvidia-esm-hook-epilogue	Clarifies ESM policy.
nvidia-ipmisol	Enables IPMI Serial-over-LAN.
nvidia-logrotate	Provides NVIDIA log-rotation policy.
nvidia-motd	Provides the DGX message of the day.
nvidia-raid-config	Configures DGX RAID.
nvidia-redfish-config	Configures the Redfish host interface.

Platform-Specific Packages

The following packages are installed according to the DGX platform:

Package Name	Platforms	Description
nvidia-mig-manager	DGX systems that support MIG	NVIDIA MIG Partition Editor and systemd service.
open-nvdebug	DGX servers	NVIDIA GPU debugging tools.
dgx-server-grub	DGX servers	Customizes GRUB for DGX servers.
nvidia-system-station-core	DGX workstations	Installs the DGX Station core desktop packages listed below.

nvidia-system-station

The following DGX Station tables identify the package source when a metapackage combines packages from the DGX/BaseOS repository, NVIDIA driver repository, or Ubuntu archive.

Package	Description
nvidia-system-station-core	Provides the core DGX Station desktop environment.
nvidia-system-station-common	Provides common DGX Station desktop components.
nvidia-system-station-utils	Provides DGX Station desktop utilities.

APT installs the following recommended packages by default unless configured not to: nvidia-system-station-localization, nvidia-system-station-games, and nvidia-system-station-apps.

nvidia-system-station-core

Package	Description	Package source
apport-gtk	Provides graphical crash reporting.	Ubuntu archive
apt-config-icons-hidpi and apturl	Provide high-resolution APT icons and a graphical package installer.	Ubuntu archive
avahi-autoipd and avahi-utils	Provide automatic local-network addressing and service discovery.	Ubuntu archive
bluez and bluez-cups	Provide Bluetooth and Bluetooth printing support.	Ubuntu archive
branding-ubuntu, yaru-theme-gtk, yaru-theme-icon, and yaru-theme-sound	Provide Ubuntu desktop branding and themes.	Ubuntu archive
nv-yaru-theme-all	Provides NVIDIA Yaru desktop themes.	DGX/BaseOS repository
cups-browsed, cups-bsd, and cups-pk-helper	Provide printer discovery, printing commands, and printer authorization.	Ubuntu archive
dgxstation-desktop and dgxstation-grub	Provide DGX Station desktop and bootloader configuration.	DGX/BaseOS repository
gnome-initial-setup	Provides the GNOME first-run setup application.	Ubuntu archive
gnome-keyring-pkcs11 and libpam-gnome-keyring	Provide GNOME Keyring PKCS#11 and PAM integration.	Ubuntu archive
gnome-power-manager	Provides GNOME power-management functionality.	Ubuntu archive
gnome-remote-desktop	Provides GNOME remote-desktop functionality.	Ubuntu archive
gnome-shell-extension-prefs	Provides GNOME Shell extension preferences.	Ubuntu archive
gnome-themes-extra-data	Provides additional GNOME theme data.	Ubuntu archive
gsettings-ubuntu-schemas	Provides Ubuntu desktop settings schemas.	Ubuntu archive
gtk-im-libthai	Provides Thai input-method support.	Ubuntu archive
grub-common, plymouth-label, and plymouth-themes	Provide bootloader and graphical boot-splash support.	Ubuntu archive
nvidia-plymouth	Provides DGX graphical boot-splash support.	DGX/BaseOS repository
gvfs-fuse	Provides a user-space virtual file system.	Ubuntu archive
ibus, ibus-gtk, ibus-gtk3, ibus-gtk4, and im-config	Provide input-method framework support.	Ubuntu archive
libpam-pwquality	Provides PAM password-quality checking.	Ubuntu archive
network-manager-gnome	Provides graphical network management.	Ubuntu archive
nvidia-update-manager	Provides NVIDIA package-update management.	DGX/BaseOS repository
policykit-desktop-privileges	Provides desktop authorization configuration.	Ubuntu archive
ubuntu-desktop	Provides the Ubuntu desktop metapackage.	Ubuntu archive
ubuntu-report	Provides Ubuntu system-reporting functionality.	Ubuntu archive
xdg-utils	Provides desktop-integration utilities.	Ubuntu archive

Through its Recommends dependencies, nvidia-system-station-core installs the following packages under APT’s default configuration. APT installs them unless the system is configured to disable installation of recommended packages.

Recommended packages

Package	Description	Package source
nvidia-settings	Provides the NVIDIA X Server Settings application.	NVIDIA driver repository
nvidia-conf-xconfig	Provides NVIDIA X configuration support.	DGX/BaseOS repository
nv-docker-gpus	Limits Docker to use 3D-controller-class NVIDIA GPUs (compute GPUs).	DGX/BaseOS repository
libnvidia-compute or nvidia-compute-utils	Provides NVIDIA GPU compute libraries or utilities, as appropriate for the driver packaging.	NVIDIA driver repository
gnome-core	Provides the core GNOME desktop components.	Ubuntu archive

Both nvidia-system-station-common and nvidia-system-station-utils depend on nvidia-system-station-core. The core metapackage is provided by the DGX/BaseOS repository; their remaining direct dependencies are provided by the Ubuntu archive. To view the direct dependencies configured for your release, run:

apt-cache depends nvidia-system-station-common
apt-cache depends nvidia-system-station-utils

nvidia-system-station-common

Installs nvidia-system-station-core if it is not already installed, plus common Ubuntu desktop components, English localization, accessibility, multimedia, and supporting libraries.

nvidia-system-station-utils

Installs nvidia-system-station-core if it is not already installed, plus Ubuntu utilities for backup, storage, remote desktop, printing, scanning, networking, and USB devices.

nvidia-gpu-desktop-utils

Package	Description	Package source
nvidia-settings	Provides the NVIDIA X Server Settings application.	NVIDIA driver repository
nvidia-conf-xconfig	Provides NVIDIA X configuration support.	DGX/BaseOS repository
libnvidia-compute or nvidia-compute-utils	Provides NVIDIA GPU compute libraries or utilities, as appropriate for the driver packaging.	NVIDIA driver repository

nvidia-system-station-localization

Installs nvidia-system-station-core and the non-English language packs, fonts, dictionaries, input methods, and internationalization support for the DGX Station desktop environment.

nvidia-system-station-apps

Installs nvidia-system-station-core, nvidia-desktop-default-snaps, and desktop applications, including LibreOffice, media applications, OpenVPN, and document and image utilities.

nvidia-system-station-games

Package	Description
nvidia-system-station-core	Provides the core DGX Station desktop environment.
aisleriot, gnome-mahjongg, gnome-mines, and gnome-sudoku	Provide GNOME desktop games.
gamemode	Provides the GameMode performance-optimization service.
libgamemode0 and libgamemodeauto0	Provide GameMode runtime libraries.

Packages Replaced by TuneD Profiles

In BaseOS 9, the following former configuration packages are empty transitional packages that depend on nvidia-tuned-profiles. They can be removed after an upgrade. The active TuneD profile, or a static systemd drop-in installed by nvidia-tuned-profiles, provides the applicable configuration for the platform.

Former package	Configuration now provided by nvidia-tuned-profiles
nv-cpu-governor	Sets the CPU governor to performance through the nvidia-base profile.
nvidia-acs-disable	Configures the disable-acs-redir kernel parameter through the nvidia-acs-disable profile when a platform profile includes it.
nvidia-arm-cspmu-loader	Loads the Arm CSPMU module through the applicable Arm64 performance profile.
nvidia-crashdump	Configures crash dumps through the platform crash-dump profiles.
nvidia-disable-init-on-alloc	Sets the init_on_alloc=0 kernel parameter through the nvidia-base profile.
nvidia-disable-numa-balancing and nvidia-earlycon	Configure the NUMA-balancing and early-console kernel parameters through the applicable Arm64 performance profile.
nvidia-enable-power-meter-cap	Configures the ACPI power-meter module parameter through the applicable Arm64 performance profile.
nvidia-fs-loader	Loads the nvidia_fs module, with the required module parameters, through applicable DGX performance profiles.
nvidia-ib-umad-loader	Loads the ib_umad module through the nvidia-base profile.
nvidia-pci-bridge-power	Configures PCIe bridge runtime power through the nvidia-pci-bridge-power profile on DGX A100 and A800 systems.
nvidia-pci-no-realloc and nvidia-pci-realloc	Configure PCI resource reallocation through applicable platform performance profiles.
nvidia-relaxed-ordering-gpu	Configures GPU PCIe relaxed ordering through applicable platform performance profiles.
nvidia-conf-cachefilesd	Configures cachefilesd by using a static systemd vendor drop-in, rather than a TuneD profile.
nvidia-cppc-cpufreq-options	Configures the CPPC CPU-frequency driver through the DGX Spark performance profile.

nvidia-relaxed-ordering-nvme is no longer provided. nvidia-nvme-options replaces it and provides the /usr/bin/nvidia-relaxed-ordering-nvme.sh script.

DGX Platform JSON Configuration

The DGX Platform JSON configuration file is installed at /etc/nvidia-platform.d/nvidia-platform-configs.json by nv-common-apis. It identifies the platform and supplies metadata to DGX configuration utilities. Its contents are release-specific; consult the file on the installed system for the current platform entries and values.

DGX Platform JSON Configuration Definitions

The following attributes continue to provide metadata to DGX utilities. An attribute can be absent for a platform when its default behavior applies.

Name	Definition
PlatformType, PlatformName, PlatformPrettyName, and PlatformString	Human-readable platform identifiers used by DGX utilities.
PlatformSerialSource	Specifies whether the platform serial number is obtained from the system or chassis.
GrepStr and UsesDmiSystemFamilyForDetection	Specify how the platform is identified from DMI information.
ConfigureDGXA100Raid and ConfigureDGXStationA100Raid	Select the RAID layout used by nvidia-raid-config.
NVMERelaxedOrdering	Indicates whether nvidia-nvme-options configures NVMe relaxed ordering.
BMCPasswordMinLength, BMCPasswordMaxLength, BMCPasswordSupportsZerofill, and BMCPasswordComplexityReq	Define BMC password requirements for the OEM installation screens.
NVSMAlertsSupported	Indicates whether NVSM alert reporting is supported for the platform.
NeedsMRRSConfig and NeedsAccBytesTuning	Select Mellanox adapter PCIe configuration performed by nvidia-mlnx-config.
IPMIDefaultSerialTTY	Selects the default IPMI Serial-over-LAN TTY.
NeedsOEMXconfigOverride, NeedsInitialNvidiaXconfig, and NeedsAdaptiveNvidiaXconfig	Control applicable NVIDIA X configuration behavior for DGX workstations.
NeedsContainerdOverride	Selects an applicable container-runtime override for supported workstation platforms.
NeedsOemConfigPostActNetplanApply	Indicates whether OEM installation must run netplan apply after post-install configuration.
UsesFabricManager	Indicates whether the platform uses NVIDIA Fabric Manager.
IsDgxServerType and IsDgxDesktopType	Identify whether a platform is a DGX server or workstation for release-upgrade processing.
BaseOSDefaultStack	Identifies the default BaseOS stack for the platform.

Note

In BaseOS 9, TuneD profiles set the following platform-specific settings rather than using the corresponding values in this file:

• NeedsDisableInitOnAlloc (init_on_alloc)

• UsePciDisableAcsRedir (PCIe ACS redirection)

• CrashdumpMem (crash-kernel memory)

• IommuPt (IOMMU passthrough)

• NeedsDisableNumaBalance (NUMA balancing)

• NeedsEarlycon (early console)

• ConsoleSetting and SerialBaudRate (console settings)

• PciRealloc (PCI resource reallocation)

• GpuRelaxedOrdering (GPU relaxed ordering)

• EnablePowerMeterCap (ACPI power-meter capping)