I just received my DGX Spark yesterday and everything was working perfectly. I installed a 65GB LLM, but after rebooting, the system never fully started up again. It gets all the way to the GDM login service, shows everything as “[OK]” in the boot messages, but then just hangs there indefinitely.
I have been troubleshooting this for hours, I tried:
GRUB access - I can get to the boot menu, but when I try recovery mode or any of the advanced options, it still hangs at the same point.
Serial console - I hooked up a serial cable and can see the boot messages, but once it hits GDM it stops responding. Ctrl+Alt+F2 just gives me a blinking cursor.
Secure Boot - I was getting Secure Boot errors initially, but I disabled that in the BIOS. That got me past those errors, but the boot hang remains.
NVIDIA recovery image - I downloaded the official DGX Spark recovery image (version 1.91.51) and created a bootable USB drive. The files look correct (has boot/, efi/, fw/ folders), but my system’s UEFI won’t detect it as a bootable device. I even tried manually adding a boot entry.
I just can’t go beyond this screen !
I was really excited to start working with this system and now I am just staring at a frozen boot screen.
@ashreaperone , I agree, tried even reaching out to their support team as always they suggested to post here and hope atleast here, we will get some response.
@aniculescu yes, I followed this guide and at step-6, they mentioned to select USB drive under “Boot Override” but I don’t see USB drive. I have only two options under boot override, UBUNTU (Samsung) and UEFI - PXE IPV4 realtex PCIe 10 GBE family contributor
@aniculescu Thank you for the USB suggestion. I wanted to clarify my current situation and seek your assistance on the best recovery path.
The boot failure occurs specifically at the GDM service after installing a large AI model. The system shows all services as [OK] until GDM starts, then hangs indefinitely.
My key question: Would you recommend attempting a system reflash, or are there less invasive troubleshooting steps I should try first?
It looks like in the DGX Spark Recovery Image, the CreateUSBKeyMacOS.sh script was moved out of the usbimg.customer/scripts folder, but the path inside the script wasn’t updated. The script still references /usbimg.customer/../usb instead of /usbimg.customer/usb, which causes it to fail on macOS with the error:
Install folder not found:
I was able to fix it by simply moving the script back into the scripts folder. This issue seems to affect only the CreateUSBKeyMacOS.sh script. The others seem to work fine.
You can try accessing the system remotely via SSH. If you can connect would you be able to collect the complete output of journalctl and attach it here, or send it to us? If you can also run sudo nvidia-bug-report.sh and collect the log file from that, that would also be helpful.
The recovery process boots successfully from USB and displays the recovery menu. After selecting “START RECOVERY,” the system attempts to mount the USB device at /dev/sda1 but fails repeatedly after 10 attempts. The screen then freezes at the error message.
The DGX Spark has NVMe internal storage drives that are assigned /dev/sda, /dev/sdb, etc. during boot. This causes the USB recovery drive to be assigned a different device name (likely /dev/sdc1 or /dev/sdd1), but the recovery scripts appear to be hardcoded to look for /dev/sda1.
On my laptop, the same USB drive is correctly detected as /dev/sda1 and mounts properly, confirming the USB was created correctly.
I’ve Tried:
Multiple USB drives (different brands/capacities)
SSD via USB adapter
Different USB ports on the hub
Recreating the recovery media multiple times
Complete power cycles of the DGX Spark
Verifying the USB partition structure (correct FAT32, bootable, proper label)
Kindly help, since 4-days, I am just trying to reflash.
