ASUS Ascent GX10 (GB10) – shutdowns under load, now powers off 2s after power-on; ASUS won't honor warranty

Hi all,

I’m reporting another GB10 unit failure, and I’m hoping NVIDIA can help escalate with ASUS.

Hardware: ASUS Ascent GX10 (GX10-GG0003BN), DGX OS, bought in August 2026, in use for about 1 week.

What happened:

During the first week, the unit shut down unexpectedly about 5 times under normal AI workloads.

The last time, it froze completely during a long-context inference run, and I hard-powered it off.

Since then, it powers on for about 2 seconds, then shuts down. There’s no display output and no BIOS screen.

It uses the original power adapter, has never been opened, and has no physical damage. A 4-hour full power drain didn’t help.

This looks like the same failure pattern as other threads here about units that froze under load and wouldn’t boot afterwards.

The warranty problem: the ASUS warranty check shows the unit active until 2028 and registered, but ASUS Moldova, where I live, offers only paid diagnostics, ASUS Poland says the unit is from the Netherlands and won’t handle it, and the seller sends me to ASUS. The hotline quoted 2–3 months with no replacement.

Questions:

Is this a known GB10 hardware or firmware failure mode, and is there a recovery procedure?

Can NVIDIA escalate GB10 warranty cases with OEM partners like ASUS?

Thanks.

Update, unit is going to RMA, initial diagnostics indicates board damage but they don’t know exactly what failed.

I am interested to hear about Asus’s handling of your case. Thanks for the updates.

After me starting a charge back through ZEN care, and starting to post on forums they have decided to call me in today and take the device for RMA. That’s how they are handling the case. Absolute nightmare. Damage is on board level, so it is a replace or refund candidate. I want to actually know what went wrong, if I ever get unit repaired I’ll share logs here. I don’t have logs as the unit unexpectedly froze and powered down. So far we have one confirmed fact that something in the board is fried. They don’t know what. I’ll keep everyone posted.

I’d like to update the topic, they have done the diagnosis, they said PMIC and VRM modules completely fried.

They ordered a new motherboard. It’s a great unknown when it will arrive. So most likely I’m out at least for a month or two, and there is a chance that I will just get a refund instead of a new board.

I will keep everyone updated. On a one week old unit such heavy damage is a bit interesting to me.

That would be rough, considering the lack of availability and price hikes.

I told to the service center I do not want that option since I will end up buying the same device anyway. But I might be out of luck unfortunately. Let’s see. If not this I’ll buy a similar box like Lenovo or MSI they are essentially the literal same box in different packaging anyway with same OS.

Follow-up: our GX10 died with no logs at all. Defective board, replaced under warranty

Sharing what we found in case it helps others with sudden power-offs on the GX10.

Setup

  • ASUS Ascent GX10 (GB10), about three weeks after first boot
  • BIOS GX10DGX.0105 (2026-05-05) at the time of failure, NVIDIA driver 580.178.04
  • Local LLM inference with Ollama (llama.cpp backend), models up to ~98 GB, contexts up to ~76k tokens
  • Only the original power adapter and cable connected, nothing else

What happened

  • 2026-09-21 03:36 UTC: the unit stopped responding. Power LED stayed on, no network, no SSH.
  • I forced it off with the power button. After that it never booted again: press power → LED on for ~2 s, fans spin up, a brief HDMI signal, then it turns off. It never reached the BIOS.
  • Tried: a different power source (an EcoFlow power station; the unit drew ~16–22 W before cutting off), unplugging and draining power for several hours, and just the original adapter. No change.

What the logs show (the SSD survived, so after the repair I read the journal from the failed boot)

  • The last journal entry is at 03:36:05 UTC, mid-activity. There was no shutdown sequence, no kernel panic, no thermal / NVRM / Xid / AER / MCE message, and nothing in pstore.
  • The system was essentially idle when it died. Ollama’s last request finished at 22:45 UTC the evening before, so the GPU had done nothing for ~5 hours (a model stayed loaded). Our health snapshots every 2 minutes looked normal right up to the end: 37–45 GB of memory available, no swap, all services up.
  • So in our case it was not a load spike or overheating. It was a sudden hardware power cut or a freeze with no software trace.
  • Looking back, the journal also shows the same kind of silent, unlogged stop on Sep 6 (right after a model finished loading) and several on Sep 8 (mostly idle). These may have been early warning signs.

Outcome

  • The authorized ASUS service center confirmed the motherboard was damaged and replaced it under warranty. The SSD and all data were fine.
  • Note: the replacement board came with an older BIOS (0101, 2025-10-14) instead of 0105, so check your BIOS version after any board swap and update it.

The case is closed for now. I am not able to get further details or logs. It points to a motherboard defect rather than me running the thing to it’s limits, because when it died it was simply idle. Doing nothing.