DGX Spark Performance Degradation - GPU Power Draw Issue

Tried restoring the system or reaching for post-sale? I got my first unit with the NIC totally down and have to replace one. I don’t think MSI have a very great QC. But the spark here in Hong Kong is quite scarce, replacing one takes weeks. Hope you can find a solution!

Firmware Component Name: FLASH
Firmware Version: SBP:R:2.152.15
Firmware ID: Not Specified

–

Firmware Component Name: UEFI
Firmware Version: R:1.108.20
Firmware ID: Not Specified

Firmware Component Name: EC Firmware
Firmware Version: 3.3.2
Firmware ID: Not Specified

–

Firmware Component Name: PD Firmware
Firmware Version: PD0 FW1: 5.7, FW2: 5.22
Firmware ID: Not Specified
Release Date: Not Specified

Firmware Component Name: PD Firmware
Firmware Version: PD1 FW1: 5.7, FW2: 5.22
Firmware ID: Not Specified
Release Date: Not Specified

Adding data from an MSI EdgeXpert MS-C931 for comparison.

sudo dmidecode -t 45 output:
FLASH: SBP:R:2.148.24
UEFI: MSI_UEFI_40_1.6.0
EC Firmware: 4.1.60
PD Firmware: PD0 FW1: 5.0, FW2: 4.10
PD1 FW1: 5.0, FW2: 4.10

Comparing against post #44 (Founders Edition, working):

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚ β”‚ EdgeXpert β”‚ Founders Edition β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚ SOCFW β”‚ 2.148.24 β”‚ 2.152.15 β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚ PD FW2 β”‚ 4.10 β”‚ 5.22 β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚ EC β”‚ 4.1.60 (MSI scheme) β”‚ 3.3.2 β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

Symptoms: GPU power limit N/A, GPU model Unknown, default clock 2418 MHz, periodic SW power capping events. GPU
computes at ~70W during inference but mailbox communication to EC appears broken.

Current state: EdgeXpert is on GA2 OTA2 (2026-03-04) β€” the only BSP MSI has published. SOCFW 2.152.15 has not been
released for the EdgeXpert. nvidia-spark-ota-check reports EC/SOCFW/USBPD as because the tool cannot read
MSI’s firmware version scheme (EC uses 4.1.60 not 3.3.2).

The fix appears to require MSI publishing an updated BSP with SOCFW 2.152.15 equivalent. Have filed a request with MSI
support.

The GX10 received an update:

        Firmware Component Name: FLASH
        Firmware Version: SBP:R:2.148.24
        Firmware ID: Not Specified
--
        Firmware Component Name: UEFI
        Firmware Version: ASUS_UEFI_0104
        Firmware ID: Not Specified
        Release Date: Not Specified
--
        Firmware Component Name: EC Firmware
        Firmware Version: 2.78.24
        Firmware ID: Not Specified
--
        Firmware Component Name: PD Firmware
        Firmware Version: PD0 FW1: 5.0, FW2: 5.7
        Firmware ID: Not Specified
        Release Date: Not Specified
--
        Firmware Component Name: PD Firmware
        Firmware Version: PD1 FW1: 5.0, FW2: 5.7
        Firmware ID: Not Specified
        Release Date: Not Specified

The EC firmware and the UEFI firmware were updated. Let’s see if they fix the problems.

I’ve seen one unit power itself on after having its power brick plugged in again. This happened after the update was installed and the system was shut down cleanly from the OS using the shutdown command. I’ll check if it happens again.

Ok, I’m able to reproduce this. The GX10 powers itself on automatically when plugged in again. I haven’t configured this in the UEFI firmware. I’ll do more unpaid QA work for Asus and Nvidia later today.

Auto power on has always been the default behaviour.

I had it disabled before the upgrade. It is indeed set by default now to active.

Maybe they decided to make it the same on the GX10 as the FE model?

I made a firmware update yesterday and got stuck in low-power mode again. My Spark was switched off overnight and booted normally this morning. However, while working, I noticed that the performance had dropped again and it was back in low-power mode.

	Firmware Component Name: FLASH
	Firmware Version: SBP:R:2.148.24
	Firmware ID: Not Specified
--
	Firmware Component Name: UEFI
	Firmware Version: ASUS_UEFI_0104
	Firmware ID: Not Specified
	Release Date: Not Specified
--
	Firmware Component Name: EC Firmware
	Firmware Version: 2.78.24
	Firmware ID: Not Specified
--
	Firmware Component Name: PD Firmware
	Firmware Version: PD0 FW1: 5.0, FW2: 5.7
	Firmware ID: Not Specified
	Release Date: Not Specified
--
	Firmware Component Name: PD Firmware
	Firmware Version: PD1 FW1: 5.0, FW2: 5.7
	Firmware ID: Not Specified
	Release Date: Not Specified

In the screenshot, you can see that it is only drawing 10 W at 95% utilization.

After unplugging the power cable from both the wall outlet and the Spark, it draws 97 W again.

        Firmware Component Name: FLASH
        Firmware Version: SBP:R:2.148.24
        Firmware ID: Not Specified
--
        Firmware Component Name: UEFI
        Firmware Version: ASUS_UEFI_0104
        Firmware ID: Not Specified
        Release Date: Not Specified
--
        Firmware Component Name: EC Firmware
        Firmware Version: 2.78.24
        Firmware ID: Not Specified
--
        Firmware Component Name: PD Firmware
        Firmware Version: PD0 FW1: 5.0, FW2: 5.7
        Firmware ID: Not Specified
        Release Date: Not Specified
--
        Firmware Component Name: PD Firmware
        Firmware Version: PD1 FW1: 5.0, FW2: 5.7
        Firmware ID: Not Specified
        Release Date: Not Specified

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.142                Driver Version: 580.142        CUDA Version: 13.0     |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GB10                    On  |   0000000F:01:00.0 Off |                  N/A |
| N/A   47C    P0             11W /  N/A  | Not Supported          |     95%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+


Attached GPUs                                          : 1
GPU 0000000F:01:00.0
    Performance State                                  : P0
    Clocks Event Reasons
        Idle                                           : Not Active
        Applications Clocks Setting                    : Not Active
        SW Power Cap                                   : Not Active
        HW Slowdown                                    : Not Active
            HW Thermal Slowdown                        : Not Active
            HW Power Brake Slowdown                    : Not Active
        Sync Boost                                     : Not Active
        SW Thermal Slowdown                            : Not Active
        Display Clock Setting                          : Not Active
    Clocks Event Reasons Counters
        SW Power Capping                               : 338072605 us
        Sync Boost                                     : 0 us
        SW Thermal Slowdown                            : 0 us
        HW Thermal Slowdown                            : 0 us
        HW Power Braking                               : 0 us
    Sparse Operation Mode                              : N/A

The updates didn’t fix the problem.

============================================================
Spark GPU Throttle Check

GPU state at idle:
Clock: 208 / 3003 MHz
P-state: P8
Power: 4.8 W

Warming up GPU (2.0s)…

Collecting 20 samples under load (0.5s interval)…
Threshold: 1400 MHz

  #  Clock (MHz)  Max (MHz)  PState  Power (W)

───── ─────────── ───────── ────── ─────────
1 624 3003 P0 12.5
2 624 3003 P0 12.4
3 624 3003 P0 12.5
4 624 3003 P0 12.5
5 624 3003 P0 12.5
6 624 3003 P0 12.5
7 624 3003 P0 12.5
8 624 3003 P0 12.5
9 624 3003 P0 12.4
10 624 3003 P0 12.4
11 624 3003 P0 12.4
12 624 3003 P0 12.4
13 624 3003 P0 12.5
14 624 3003 P0 12.5
15 624 3003 P0 12.5
16 624 3003 P0 12.6
17 624 3003 P0 12.5
18 624 3003 P0 12.5
19 624 3003 P0 12.5
20 624 3003 P0 12.5

────────────────────────────────────────────────────────────
RESULTS
────────────────────────────────────────────────────────────
Samples: 20
Peak clock: 624 MHz
Average clock: 624 MHz
Avg power draw: 12.5 W
Below threshold: 100% of samples < 1400 MHz

β–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆ
β–ˆ FAIL β€” GPU IS THROTTLED β–ˆ
β–ˆ Clock never exceeded threshold under load. β–ˆ
β–ˆ Likely cause: bad USB PD power negotiation. β–ˆ
β–ˆ Try: disconnect power brick from wall and Spark, β–ˆ
β–ˆ wait a minute, then reconnect. β–ˆ
β–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆ

This unit didn’t crash to end up in this state. It was powered off for approximately 20 hours with the power brick plugged in. It started in this state. It was running properly before shutdown.

Happening to me on my Asus Ascent GX10. Unplugging from the mains for a minute or so fixed it. Annoying.

    Firmware Component Name: FLASH
    Firmware Version: SBP:R:2.148.24
    Firmware ID: Not Specified
--
    Firmware Component Name: UEFI
    Firmware Version: ASUS_UEFI_0104
    Firmware ID: Not Specified
    Release Date: Not Specified
--
    Firmware Component Name: EC Firmware
    Firmware Version: 2.78.24
    Firmware ID: Not Specified
--
    Firmware Component Name: PD Firmware
    Firmware Version: PD0 FW1: 5.7, FW2: 4.10
    Firmware ID: Not Specified
    Release Date: Not Specified
--
    Firmware Component Name: PD Firmware
    Firmware Version: PD1 FW1: 5.7, FW2: 4.10
    Firmware ID: Not Specified
    Release Date: Not Specified

The throttling happens daily now with all the updates. It’s always throttled when it powers on the next day after being powered on for use the previous day.

This is how my GX10 gets throttled

  • unplug it and plug it back in to get it out of the throttled state
  • the GX10 starts automatically when power is restored
  • use it for a number of hours and power it off
  • power it on again the next day β†’ it’s limited to 4-12 W and performs extremely poorly

Is someone from Nvidia still working on this?

Another observation is that the GB10 chip from this GX10 never manages to go above 85-86 W. It doesn’t hit 95-96 W as other units do. It may or may not be relevant based on the workload.

I’ve been unplugging my GX10 every evening after power off. Haven’t encountered the throttling since.

That approach won’t help much with long running tasks. It’s a pain to have to jump through hoops with such hardware. I still hope this issue is caused by a bug which can be fixed in the firmware.

Hi All,

Just a quick note to anyone experiencing power issues. I had throttled speed issues and it was actually related to my UPS Battery Backup that was gating the Power supply. Something to rule out if possibly similar circumstances.

Any update on the issue?

To understand this, I was capturing GX10 (ASUS) telemetry, and it seems like I am capped at GPU T.Limit Temp : 50 C and 15W. While GPU utilisation seen 100%

The system was performing diffrently a week ago, and now it feels like rubbish.

@nilayparikh Install all updates (those from the web interface 11000) + after updates, power off + then remove the type-c power for 5+ minutes. It helped me, the GPU started to burn well

@dobs What are you using to get those pretty graphs?

The GX10 has ended up in a throttled state a few more times since my last reply. It happens almost daily.

Unplugging the power cable from the unit seems to fix it just as unplugging the power brick does. The power brick doesn’t appear to get into a bad state which prevents it from providing the full power to the GX10.

A different GX10 unit doesn’t throttle. It boosts to a higher 2548 MHz than this GX10 which throttles. The broken GX10 can’t sustain 2500 MHz in the throttling test or under load. Both run the same firmware and software. Both are fully updated.

I’ve swapped the broken GX10’s power brick with the other unit’s power brick. The broken GX10 still throttled. The other not broken GX10 unit doesn’t throttle at all, not even with the broken unit’s power brick.

I’ve reset the GPU and rebooted the unit when the GX10 was throttled to 5-14 W. It didn’t make any difference at all.

I now think this is a hardware problem with the units which throttle. It seems to throttle under load as well. Its temperature is about 77 degrees Celsius at a power consumption of 47 W shown by nvidia-smi.

You probably want to wait for Nvidia to provide a solution before you buy a GB10 system if you haven’t already put your money into this.