DGX Spark GB10: GPU fails to initialize, GSP_INIT_DONE timeout (Xid 119) and SEC2 secure-boot timeout, RmInitAdapter failed (0x62:0x65:2028)

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

Hardware: DGX Spark (GB10)
OS / DGX Spark software version: 7.5.0
Kernel: 6.17.0-1021-nvidia
NVIDIA driver/module: 580.159.03 (nvidia-580-open)

Symptom: GPU fails to initialize. nvidia-smi returns “No devices were found.” The GPU was working previously; the failure developed at some point after routine updates, including an OTA firmware update applied via fwupdmgr. It now reproduces on every boot.

Confirmed working:

  • GPU enumerates on PCIe: 000f:01:00.0 VGA compatible controller: NVIDIA Corporation Device 2e12 (rev a1)
  • NVIDIA kernel modules load: nvidia, nvidia_drm, nvidia_modeset, nvidia_uvm
  • nvidia_drm modeset = Y, no NVIDIA blacklist active

The failure (from dmesg): the GSP boots and establishes RPC (GSP_SET_SYSTEM_INFO succeeds, GSP_LOCKDOWN_NOTICE received), then hangs. The driver waits for GSP_INIT_DONE (function 4097) but the RPC buffer holds UCODE_LIBOS_PRINT (function 4108):

NVRM: Xid (PCI:000f:01:00): 119, Timeout after 7s of waiting for RPC response from GPU0 GSP! Expected function 4097 (GSP_INIT_DONE)
NVRM: RmInitAdapter: Cannot initialize GSP firmware RM
NVRM: nvCheckOkFailedNoLog: Check failed: Call timed out [NV_ERR_TIMEOUT] (0x00000065) returned from ksec2PrepareBootCommands_HAL … @ kernel_gsp_gh100.c:806
NVRM: GPU 000f:01:00.0: RmInitAdapter failed! (0x62:0x65:2028)

The SEC2 PrepareBootCommands timeout and RmInitAdapter failure repeat on every retry. Earlier boots also showed bad register reads (0xbadf5720). The OTA checker additionally reports a persistent failed “kernel” firmware component.

This presents as a GSP/SEC2 secure-boot / firmware-init failure on the GPU rather than a driver or OS configuration problem. Recovery steps already taken (set multi-user.target, enabled GRUB menu, repaired fwupd, removed a stale nvidia-drm modeset=0 override and rebuilt initramfs, applied the offered USBPD firmware update to 0x00000516) did not change the GPU failure.

Full nvidia-bug-report.log.gz attached. NVIDIA hardware support has asked me to obtain confirmation from the developer team that this is a hardware fault before proceeding with an RMA. Can the team confirm?

2

Looks like the nvidia-bug-report didn’t attach properly, can you try again or send me a DM?

3

nvidia-bug-report.log.gz (121.6 KB)

fieldiag-results.zip (18.4 KB)

4

Hello,

Thank you for the update.

As an additional troubleshooting step, we would recommend performing a System Recovery on the DGX Spark using the official recovery image. This process restores the system software and firmware to a known-good state and can help rule out software or firmware corruption before proceeding further.

Steps : System Recovery — DGX Spark User Guide

Please use the official DGX Spark System Recovery Image download page and follow the recovery steps from the DGX Spark User Guide.

DOWNLOAD DGX Spark System Recovery Image

If the issue persists even after System Recovery, then based on the current logs and field diagnostic behavior, the problem would appear more consistent with a hardware-level issue.

Please feel free to reach out to us if you have any questions or concerns. We are here to help.

Best regards,
Zia
NVIDIA Customer Care

Update: System Recovery is complete. I verified the official 1.135.34 recovery image against its checksums, built the recovery USB, restored UEFI defaults, confirmed Secure Boot enabled with factory keys restored, and ran the recovery to completion (OS and firmware reflashed, SSD erased).

After recovery, the unit powers on and is on the network (it acquired a new DHCP address and responds reliably to ping). However, it produces no video past the NVIDIA firmware splash, the screen goes dark, so I cannot complete the local first-boot setup. As a result, SSH is not enabled: both an SSH connection and NVIDIA Sync are actively refused on port 22 (“target machine actively refused it”). I have no way to reach the unit because the display does not function.

This matches the earlier field diagnostic result: FAIL on GpuStress, “GFW boot reported a failure.” The GPU firmware boot failure persists after a full factory System Recovery, and now also prevents display output needed to complete setup. Per your guidance that a fault persisting after recovery indicates a hardware issue, please proceed with the RMA.

Serial: Field diagnostic logs (summary.json) and nvidia-bug-report.log.gz are available to upload.

5

Thanks for the update, please share your failed fieldiag logs with consumer support and request an RMA

6

Same symptom on my DGX Spark (Founders Edition, Serial 1983925007599).
OS 7.5.0, kernel 6.17.0-1021-nvidia, driver 580.159.03 — identical to yours.

Identical kernel error on every boot:
NVRM: ksec2PrepareBootCommands_GB20B: SEC2 secure boot partition timed out
NVRM: RmInitAdapter: Cannot initialize GSP firmware RM
NVRM: GPU 000f:01:00.0: RmInitAdapter failed! (0x62:0x65:2028)

GPU enumerates on PCIe (000f:01:00.0) and kernel modules load fine, so this is
a GSP firmware bootstrap failure, not a driver/OS config issue. It started
after a forced power-off when a large LLM load wedged the GPU under memory
pressure. Already tried: 10+ min full power-disconnect cold boot (multiple
times), kernel module unload/reload, normal reboots — none fixed it.

Since your thread concluded with an RMA after System Recovery also failed,
I’d like to proceed straight to RMA rather than erase my SSD. My
nvidia-bug-report.log.gz is ready to share. How should I submit it for RMA?

7

Hello Soojin,

I had to create the post, have the moderator deem it an RMA and there was an email that I generated from the live chat support. I would say that you have to run the fielddiag and the bug-report available for them to review.

Best,

8

Thanks wbrokaw — that matches the path I’m on.

NVES — I ran the field diagnostic as requested. Result: partnerdiag cannot run because its MODS diagnostic driver is blocked by Secure Boot:
insmod: ERROR: could not insert module mods.ko: Key was rejected by service

The MODS driver compiles and installs fine — it’s Secure Boot rejecting the unsigned module. The documented fix is to disable Secure Boot in UEFI (mokutil confirms: SecureBoot enabled), but the GPU failure also killed all video output (monitor shows nothing past the firmware splash), so I cannot enter UEFI to disable it. There is no BMC/remote console on DGX Spark.

So the hardware failure blocks the diagnostic on two fronts:

  1. GPU fails to initialize — GSP firmware: SEC2 secure boot timeout, RmInitAdapter failed 0x62:0x65:2028 (nvidia-smi: “No devices found”)
  2. No video output, so UEFI is unreachable to disable Secure Boot for fieldiag

This isn’t a driver/setup issue — the diagnostic driver builds correctly; the hardware fault itself prevents running it. Identical signature to thread 373394, which concluded in an RMA.

Attached: full fieldiag logs (driver.log shows the Secure Boot rejection). nvidia-bug-report is on my first post. Could you confirm RMA?

Serial 1983925007599, Under Warranty (06/03/2027), Customer Care case #260620-000330.

fieldiag-logs.zip (5.5 KB)