[35.4.1] LAN Tx performace is getting lower when HDMI is plug out -EE

Hi @DaneLLL ,

I am HW engineer.
About the topic “[[35.4.1] LAN Tx performace is getting lower when HDMI is plug out]”([35.4.1] LAN Tx performace is getting lower when HDMI is plug out) problem, maybe I can help to compare the PCB version between ASUS and NVIDIA side.
Please help us to provide the way to check we are using the same PCB.

BR,
VincentCJ_Chen

Hi,
For design of Xavier NX developer kit, please refer to
https://developer.nvidia.com/embedded/downloads#?search=carrier%20board

Jetson Xavier NX Developer Kit 3D CAD STEP Model (P3518)
Jetson Xavier NX Developer Kit Carrier Board Design Package (P3509 A01)
Jetson Xavier NX Developer Kit Carrier Board Specification (P3509 A01)

Hi DaneLLL,

Thanks alot~!

The module are show as picture ( Mode : P3518 and P3668 ).
Are they same as your PCB ?

BR,
VincentCJ_Chen

Hi @DaneLLL,

更新我這邊的測試, 我使用同一塊底板測試Xavier NX的板子會有HDMI拔掉後LAN會有掉速的問題.
但同一塊底板插上TX2的板子測試HDMI拔掉後LAN的速度是正常的.

我這邊驗證起來的結果看起來跟硬體的板子比較沒關係, 你那邊有建議的其他測試方式或是覺得哪邊需要再確認的嗎?

Hi,
Please check the status in
[35.4.1] LAN Tx performace is getting lower when HDMI is plug out - #49 by DaneLLL

It is expected performance drops for a short while during hotplug. And then it shall go back to normal.

Hi @DaneLLL ,

我有試過若HDMI不回插回來的狀況下, throughput是會一直降速的.
如若插拔後的幾秒內速度會恢復的話我們也可以接受.
如下圖擷取部分的資料所示, 我下一次command是跑60秒, 狀況會是從15秒拔掉後就一直掉速到60秒結速.
可以理解你說的插拔過程中可能會有因切換帶來的瞬間掉速, 不過這個狀態似乎是不一樣的.

Hi,
What we observe is like:

[  5]   8.00-9.00   sec   112 MBytes   941 Mbits/sec    0    550 KBytes       
[  5]   9.00-10.00  sec   112 MBytes   941 Mbits/sec    0    550 KBytes       
[  5]  10.00-11.00  sec  96.9 MBytes   813 Mbits/sec    0    768 KBytes       
[  5]  11.00-12.00  sec  93.4 MBytes   784 Mbits/sec  534    551 KBytes       
[  5]  12.00-13.00  sec   112 MBytes   942 Mbits/sec    0    551 KBytes       
[  5]  13.00-14.00  sec   112 MBytes   941 Mbits/sec    0    551 KBytes       
[  5]  14.00-15.00  sec   112 MBytes   941 Mbits/sec    0    551 KBytes 

Please share your steps in detail so that we can set up Xavier NX developer kit(with emmc module) and try to replicate the phenomenon.

Hi @DaneLLL ,

是的, 這部分就是我想先釐清兩邊在測試上是哪裡有差異.
如果你的測試也是拔了HDMI以後就沒有再插上去, 裝置的throughput會在1秒內恢復原本的速度的話比較正常.
因為我們SW Feng跟我說他現在燒在機台上的Code是原廠的code所以理論上應該是要跟你的結果一樣才對, 是否兩邊在確認的時候有什麼差異的地方可能miss掉了?
另外你那邊在跑的時候會有這麼多retry嗎?
image

Hi Vincent,

Previously there are lots of ethernet issue reported from your sw engineer Feng. We checked every of them but we cannot reproduce any of them on our side.

There was also no other similar issue reported by other users before.

I would like to know did you ever try other Xavier NX modules in your test before. But not just swapping a carrier board.

Hi @WayneWWW,

是的, 我有先去了解你們之前的討論內容.
但由於我們手上只有一塊Xavier NX module, 所以目前只能以手上現有的資源盡可能的幫兩邊先把硬體的相關因素都先釐清.
如果只有一套Xavier NX的狀態下, 是否還有我能確認的地方?
不確定這樣是否可行, 或許我們可以把SW燒好code的SD卡寄給你們, 提供給你們插在你們的機台上測試結果是否相同?

Hi,
The production module is with emmc. The module with SD card is for development. Do you observe the issue when using production module? If yes, please share the steps in detail so that we can test on developer kit. Seems like some misunderstanding in the steps so we are not able to replicate the issue.

Hi @DaneLLL

我們目前的測試手法如下:

  1. SW將source code燒錄到SD卡.
  2. 機台使用SD卡開機並進到OS.
  3. 機台與電腦對傳測試.

因為我們手上只有1台NX的module可以測試, 是否我這邊可以寄一塊SW燒好code的SD卡給你們測試.
這樣也可以一併驗證我司SW在燒的code是否有問題.

Hi,
We would suggest you buy a Xavier NX module with emmc and give it a try. It is better to have a production module for validating real use-cases.

Hi @DaneLLL ,

有的, 我司PE1000N的機種是使用貴司MP版本的SOM, 目前產線上每台都是相同問題.
只是目前判斷不影響先生產, 待我們解決問題後更新, 所以產線也是在等我們處理中.
Version : Jetson Linux 35.4.1 using nvidia sdkmanager

Hi,
Please share your steps in detail so that we can set up Xavier NX developer kit(with emmc module) and try to replicate the phenomenon. What we observe is short drop during hotplug. Seems like some misunderstanding in the steps so we are not able to see the persistent drop in throughput.

We have worldwide teams working on Jetpack releases and have to confirm the issue occurs on the developer kit. So that our teams can reproduce it in the same way and debug further.

Hi @DaneLLL ,
vincentcj_chen is my colleague.
We used Xavier NX devkit (with emmc module) with the official image Jetpack 5.1.2 by sdkmanager.
We found that without HDMI connection, the TX performance of LAN will be lower than that with HDMI connection.
With HDMI, the TX performance can be above 949 Mbits/sec.
Without HDMI, the TX performance will reach to about 788 Mbits/sec.

The command we used on the Xavier NX devkit(with emmc module): iperf3 -s
The command we used on my PC side: iperf3.exe -c 192.168.100.100 -B 192.168.100.101 -t 120 -R

We have tested Intel I210 and Asix Ax88179 and both of them will have the same issue.
All Xavier NX EMMC modules we have have the same problem.

The following shows the iperf logs.
0.00-7.00 sec: HDMI is plug in.
7.00-67.00 sec: HDMI is plug out.
67.00-79.00 sec: HDMI is plug in.
79.00-120.00 sec: HDMI is plug out.

iperf3_20231024.txt (9.2 KB)

Hi,
Thanks for the steps. We have some questions:

  1. Do yo connect Xavier NX devkit to a HDMI out and boot? Or the devkit is booted without connecting to a HDMI device?
  2. Do you connect ethernet port of Xavier NX devkit and the host PC directly through a cable? Or you connect devkit and host PC to a router to be in local area network?
  3. Is the host PC in Ubuntu 18.04?

Hi @DaneLLL

  1. Yes. Xavier NX devkit connects to a HDMI out and boot. The devkit is booted with connecting to a HDMI device.
    我們只有接HDMI銀幕,開機,測試,拔掉HDMI,就會看到現象。

  2. directly through a cable without any router.
    我們用一條網路線對接。沒有用router等第三方裝置。

  3. windows OS. By the way, we also used orin nano devkit as iperf3 RX. Xavier NX has the same problem as well.
    沒有用過ubuntu 18.04, 我們測試使用windows OS,也使用過orin nano devkit當作對接端,Xavier NX的TX performance都有相同現象,會掉到700多。

  4. 如同之前所提到,我們發現xavier nx在kernel 4.9沒有此問題。

感謝。

Hi @DaneLLL

我們找了一塊eMMC的SOM測試狀況如下方附件影片:

V_20231024_174354_ES2.zip (20.7 MB)

提供給您參考, 看是否跟您那邊的手法相同.

Hi,
We can observe the issue without executing sudo jetson_clocks. It is because when HDMI output is absent, emc clock is dropped since the loading of HDMI is not present. Please execute sudo jetson_clocks to keep the clock at maximum.