System throttled due to over-current?

I just got a new Xavier NX and flashed the latest image from nvidia website. Next step, I cloned yolo, changed to mode 2 (15W 6 core), but then this error message popped up whenever I try to run yolo. Is that normal or have I got a bad board? I could understand the error if I am using a nano and powering thru usb, but I am using a NX which is powered by the DC barrel provided in the box. How could I went over-current with the default power brick in the box? Right now I can only run 10W modes. Any help would be appreciated, thanks.

xe

2 Likes

I too have the same issue as described with new Xavier NX.
Any suggestions NVIDIA?

@zhg65671
Could you run the sudo tegrastatus and also check with others mode by nvpmodel?

1 Like

I have the same issue as described with new Xavier NX too. and I use 10W MODE, this warning disappear. But I need use maximum performance. I use the original power adapter, witch is 19V 2.37A. I try to remove all usb device, it doesn’t work.

We see the same issue. In JetPack 4.4.1, putting the same stress on the system, we do not get the system throttling message. In JetPack 4.5, we get the message that the system is throttled. It seems as if in the new JetPack Version, throttling starts already at 10W even for the 15W mode. Tegrastats shows loads around 10-12 W.

Could you have detail step to us to reproduce the issue.

Thanks

We run jetson_clocks and then use the tool “stress”:

stress --cpu 6 --io 4 --vm 2 --vm-bytes 128M

Then we run the visionwork-sfm sample. After that we get the throttling message. Tegrastats shows:

RAM 2434/7773MB (lfb 967x4MB) SWAP 0/3887MB (cached 0MB) CPU [100%@1420,100%@1420,100%@1420,100%@1420,100%@1420,100%@1420] EMC_FREQ 40%@1600 GR3D_FREQ 16%@1109 NVDEC 166 NVDEC1 204 VIC_FREQ 0%@115 APE 150 MTS fg 4% bg 0% AO@35C GPU@38C PMIC@100C AUX@35.5C CPU@40.5C thermal@37.9C VDD_IN 10385/10358 VDD_CPU_GPU_CV 6386/6373 VDD_SOC 1576/1602

All 15W modes exhibited the same issue. Here’s the tegrastatus

Hi,
Can’t repo it by using 19v 2.37A adapter.

List test steps:

  1. Run on 15W 6core mode
  2. Enable cpu stress:
    $ stress --cpu 6 --io 4 --vm 2 --vm-bytes 128M
  3. Run VisionWork samples:
    nvx_demo_feature_tracker
    nvx_demo_feature_tracker_nvxcu

Tegrastats:

RAM 2750/7766MB (lfb 171x4MB) SWAP 0/3883MB (cached 0MB) CPU [100%@1420,100%@1420,100%@1420,100%@1420,100%@1420,100%@1420] EMC_FREQ 47%@1600 GR3D_FREQ 33%@1109 NVDEC 665 NVDEC1 665 VIC_FREQ 5%@115 APE 150 MTS fg 4% bg 0% AO@38.5C GPU@39.5C PMIC@100C AUX@38C CPU@41.5C thermal@39.7C VDD_IN 9651/6891 VDD_CPU_GPU_CV 5152/2700 VDD_SOC 1919/1774
RAM 2584/7766MB (lfb 191x4MB) SWAP 0/3883MB (cached 0MB) CPU [100%@1420,100%@1420,100%@1420,100%@1420,100%@1420,100%@1420] EMC_FREQ 50%@1600 GR3D_FREQ 39%@1109 NVDEC 652 NVDEC1 627 VIC_FREQ 28%@115 APE 150 MTS fg 4% bg 0% AO@38.5C GPU@39.5C PMIC@100C AUX@38.5C CPU@41.5C thermal@39.7C VDD_IN 9774/6892 VDD_CPU_GPU_CV 5275/2701 VDD_SOC 1959/1774

Hi ShaneCCC,

From the tegrastats output, it seems that you don’t reach the critical range of the power, as VDD_IN is below 10W. Can you put more load on the system with other samples? I reached it by running “./nvx_sample_sfm”.

hello sevm89,

may I know what’s the actual use-case for putting CPU with extreme usage and running VisionWork?

Hi JerryChang,
We are just doing a temperature verification under heavy load to verify our thermal solution.
However, we think it is a bug of the new JetPack 4.5, totally independent of any use case. There should be no throttling when the power is below 15W and the temperatures are ok. Am I wrong?

1 Like

hello sevm89,

just for confirmation,
did you had exactly same testing setup with JetPack-4.4.1 and not able to reproduce the issue?

Yes, in JetPack 4.4.1, we never got such a message.

1 Like

Just want to update on this matter: I sent the original NX back and received a new one, but the same problem persists.

Hi

Is there any update from NVidia?

I got the same issue on deepstream. In previous version (JetPack 4.4.1), I did not encounter any warnings. I disabled notification as a workaround but there should be a solution

I can confirm similar issue, using JetPack 4.5, see my thread:

1 Like

Can NVidia confirm this problem? Could you reproduce it?

I have the same issue. Running on jetpack 4.5 as well.