DGX Spark freezes under load

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

Hi DGX Developers,

I’ve been having an issue of the DGX Spark becoming unreachable whenever the system is under load. I’ve tried various ways to fix this including setting gpu-utilization to 50%. I have couple of them and the other one works just fine. I’ve checked system configurations to make sure they are similar in all aspects. Based on the recommendations of Claude, I ran FieldDiagnostics test. here are the results

Removing Nvidia drivers and services...
Stopping 'docker.service', but its triggering units are still active:
docker.socket
Stopping 'systemd-udevd.service', but its triggering units are still active:
systemd-udevd-control.socket, systemd-udevd-kernel.socket
Installing MODS driver...
Starting OneDiag...
******************************************************************
*                                                                *
*                      DGX FIELD DIAGNOSTIC                      *
*                                                                *
******************************************************************

Version                  r9.257.3
Python                   /opt/nvidia/dgx-spark-fieldiag/dgx/python/python-3.10.7-glibc-2.17-aarch64/bin/python3
Build Date               Fri, 19 Dec 2025
Start time               Tue, 09 Jun 2026 23:36:39
Est. Completion time     Wed, 10 Jun 2026 02:39:09
Logs                     /opt/nvidia/dgx-spark-fieldiag/dgx/logs-20260609-233618
Inforom logs             /opt/nvidia/dgx-spark-fieldiag/dgx/logs-20260609-233618/inforom
Product                  NVIDIA_DGX_Spark
Product Version          A.7
Family                   DGX Spark
SKU                      0000
Serial Number            1983925015783

Testing GpuStress  -\|/-\|/-\|/-\|/-\|/-\|/-\|/-\|OK [ 3:19s ]
Testing C2CStress  -\OK [ 0:06s ]
Testing CpuStress1  -\|OK [ 0:08s ]
Testing CpuStress2  -\|/-\|/-\|/-\|/-OK [ 10:01s ]
Testing PowerStress  -\|/-\|/-\|/-\|/-\|/-\|/-\|/-FAILED [ 8:08s ]

Exit Code         | Virtual Id  | Test       | Subtest | Component | Component Id | Notes
===========================================================================================================================================================================
MODS-000000000000 | GpuStress   | custommods |         | GPU       |              | OK
MODS-000000000000 | C2CStress   | custommods |         | C2C       |              | OK
MODS-000000000000 | CpuStress1  | custommods |         | CPU       |              | OK
DGX-000000000000  | CpuStress2  | cpustress  |         | CPU       |              | OK
MODS-020000600139 | PowerStress | custommods |         | Power     |              | Acceptable temperature limits exceeded or the thermal sensor is broken or miscalibrated

#######     ####    ########  ###
#######    ######   ########  ###
##        ##    ##     ##     ###
##        ##    ##     ##     ###
#######   ########     ##     ###
#######   ########     ##     ###
##        ##    ##     ##     ###
##        ##    ##  ########  ########
##        ##    ##  ########  ########

Final Result: FAIL
End time: Tue, 09 Jun 2026 23:58:25 [ 21:45s elapsed ]
Copying logs...

Can anyone please check and suggest next steps? Note that I had tried to reach out to customer care for RMA and they recommend an Nvidia Developer look into this first.

2

I want to give you some broader context here because your situation is more than just a one-off bad unit, and I think it helps to understand what’s actually going on before you push forward with support.

This is a known, documented problem across the platform - not just your machine

The MODS-020000600139 PowerStress failure you’re seeing has been showing up across multiple threads on this forum for months, and the underlying issue goes deeper than a single defective unit. Since launch, the DGX Spark has had widely reported problems with its power delivery subsystem. John Carmack publicly noted back in October 2025 that his unit was capping at around 100W instead of the rated 240W and was experiencing spontaneous reboots under sustained load. Tom’s Hardware’s own review noted that under stress, heat concentrates heavily in the corner of the chassis housing the power-delivery inductor - which lines up exactly with what PowerStress is catching on your unit.

The fact that you have a second unit that passes the same test under the same conditions is strong evidence that this is a hardware-level fault on this specific unit - but it’s worth knowing you’re not alone and this failure mode is well understood at this point.

There was a firmware update pushed in late April 2026 that specifically addressed the USB Power Delivery Controller and Embedded Controller stability. If you haven’t applied it yet, it’s worth running before anything else:

sudo apt update && sudo apt dist-upgrade
sudo fwupdmgr refresh --force
sudo fwupdmgr upgrade

That said - if your unit is already on the latest firmware and still failing PowerStress reproducibly, firmware isn’t going to save it. The diagnostic result is telling you there’s a physical fault in the power subsystem.

Next steps

  1. Open a ticket at nvidia.custhelp.com (not the developer portal - that’s for software). Attach the full FieldDiag log bundle from /opt/nvidia/dgx-spark-fieldiag/dgx/logs-20260609-233618 and reference exit code MODS-020000600139 explicitly. Note that all other subtests passed and that your second unit runs cleanly under identical conditions.

  2. Mention your second unit. Support sometimes pushes back with generic troubleshooting steps. Having a control unit that passes the same test on the same setup removes the environmental variable argument immediately.

  3. Tag @aniculescu here with your case number once you have it. He’s been actively helping move stalled RMA cases through on this subforum and has confirmed that a FieldDiag PowerStress FAIL is eligible for RMA.

  4. Watch your case timeline closely. A few people here have reported RMA cases going quiet for weeks after approval - no shipping label, expiring timelines. Follow up proactively every few days if you don’t hear back. Don’t assume silence means things are moving.

One heads-up on what to expect from the RMA process itself: per NVIDIA’s warranty terms, the replacement may be a refurbished unit rather than new, and cross-shipment (advance replacement) is not something they currently offer for DGX Spark regardless of the circumstances. That’s the reality of the process as it stands, and it’s worth setting expectations accordingly.

Your diagnostic result is clean and the failure is reproducible. Push for the RMA - the evidence is on your side.

3

Thanks Marcel! That is very helpful.

I’ve indeed updated the firmware , so, I definitely feel like I’ve exhausted all options available on the software side. I’ll follow up with your instructions.

4

After your Spark freezes, reboot and run journalctl -k -b -1 -e to get previous boot logs and share with me please.