Improvements needed for DGX Spark system restore

Accelerated Computing DGX Spark / GB10 User Forum DGX Spark / GB10
1

To improve the system restore image for the DGX Spark, I want NVIDIA to:

  1. Distribute it as a USB disk image that can be flashed using any standard tool like dd, balenaEtcher, Raspberry Pi Imager, Rufus etc. Get rid of these temperamental custom scripts or make their use optional. I’ve still not been able to get them working on the Mac without jumping through hoops.
  2. Provide the option to reinstall the OS without wiping the home folder. It is extremely frustrating that this is not an option.
  3. Get rid of the forced update after reimaging the system. This should not be a requirement to have a running machine. I think it’s fine to have it be an optional step, but there should be an option to skip the update. The machine should be fully functional once you’ve reimaged it without relying on network access to finish the restoration process. It is basically online DRM, and rendered most DGX Sparks completely inoperable during the bootloop fiasco, where a bad update caused the Spark to brick, and even if you attempted to restore it, it was still left completely broken. There is absolutely no reason why network access should be required to reset the system to a functional state.
  4. It would be good to have the option to boot into a live environment so you can access and back up your files on a non-bootable system before a factory reset, this would also allow you to rescue your system using a Timeshift snapshot, which I would highly encourage everyone to have. This live environment should have Timeshift pre-installed, as it is a very commonly used app to get a Linux system back to a running state.
  5. Fix the keyboard compatibility issues on the rescue image.

I love the DGX Spark, but the system restore process is awful, it needs to be improved.

2

The recovery process is intentionally called

 /==================================\
 || _____         _    ___  ____   ||
 |||  ___|_ _ ___| |_ / _ \/ ___|  ||
 ||| |_ / _` / __| __| | | \___ \  ||
 |||  _| (_| \__ \ |_| |_| |___) | ||
 |||_|  \__,_|___/\__|\___/|____/  ||
 \==================================/
 For DGX Spark

because you end up with a fully (sometimes!) functional system instead of going through the installation process if an ISO would be used.

FastOS is fast, ISO time consuming!

3

Thanks for responding, but I’m hoping NVIDIA can read and respond to my suggestions, hopefully with a commitment to improve the state of the recovery tools.

It’s really not great when your system is bricked by an update, you are stuck with a bunch of custom scripts to create a USB that doesn’t work, and when you do get it working it doesn’t support your keyboard, so now you have to go out and buy another keyboard to use the recovery tool, and when you get that far, you lose absolutely everything you’ve been working on since it doesn’t provide any tools for accessing your files or reinstalling the OS without wiping your home directory. And then the cherry on top, the recovery process doesn’t even leave you with a running system, so now you have to wait around with a bricked system until NVIDIA fixes the broken update on their end before you have a running system again.

I don’t think it’s unreasonable to say this is unacceptable and needs to be improved. System recovery is extremely important, and right now it feels more like an afterthought than a well conceived feature.

4

I agree with you @RazielAU! Nvidia’s DGX Spark experience needs improvements. Other vendors selling GB10 based systems, which are 99% identical to Nvidia’s FE, provide an ISO for recovery with options for the Spark.

From the grub.cfg of Dell’s ISO:

    menuentry "Install DGX OS 7.2.3 (for DGX systems)" {
    menuentry "Install DGX OS 7.2.3 Without Reformatting Data RAID (for DGX systems - does not mount /raid)" {
        menuentry "Subiquity Installer" {
        menuentry "Install DGX OS 7.2.3 Without NVIDIA Drivers" {
        menuentry "Install DGX OS 7.2.3 With Encrypted Root" {
        menuentry "Install DGX OS 7.2.3 With Encrypted Root and Without Reformatting Data RAID (does not mount /raid)" {
        menuentry "Install DGX OS 7.2.3 (for non-DGX systems)" {
    menuentry "Boot Into Rescue Shell" {
    menuentry "Check Media for Defects" {
    menuentry "Install DGX OS 7.2.3 for DGX Spark" {
    menuentry "Install DGX OS 7.2.3 Without NVIDIA Drivers for DGX Spark" {
        menuentry "Install DGX OS 7.2.3 for DGX Spark (with 6.14 kernel)" {
        menuentry "Subiquity Installer" {
        menuentry "Install DGX OS 7.2.3 With Encrypted Root for DGX Spark" {
    menuentry "Boot Into Rescue Shell" {
    menuentry "Check Media for Defects" {
menuentry 'Boot from next volume' {
menuentry 'UEFI Firmware Settings' {
5

+1 and an option to order a shippable restore USB for current owners

7

I’m stunned, that is a big improvement over what the NVIDIA restore image offers. It looks like the option to install without reformatting is only for “DGX systems” though, not the “DGX Spark” specifically, but that is much closer to what I would expect from a rescue image. They also have the option to boot into a rescue shell, it’s likely not GUI based, but at least it’s better than what we have on NVIDIA’s rescue image, which basically just wipes the full drive and does nothing else, without even a single option to help with the recovery of data…

When other manufacturers selling the same thing have a more robust recovery solution, it becomes pretty evident that indeed, the rescue image provided by NVIDIA is awful and needs to be improved.