we used Nvidia’s file JP511-orin-nano-sd-card-image.zip to flash to our Jetson Orin Nano Devkit with Intel I210 M.2.
We used iperf3 to test throughput, and we found that the on-board-LAN is ok and its throughput can reach to around 849 Mbits/sec.
However, the Intel I210 has very low throughput with only around 9.45 Mbits/sec.
The steps I did as following:
- “iperf3 -s” for a Notebook
- on the other hand, in jetson orin nano devkit, I did the command: iperf3 -c “192.168.100.102” -t 120 -i 10 -P 4
Could you please check this issue for us? Thanks a lot.
我們使用Nvidia’s file JP511-orin-nano-sd-card-image.zip燒到jetson orin nano devkit，並接上intel I210 Lan M.2。
在做iperf3傳輸時，intel I210的throughput只能達到約莫9.45 Mbits/sec。然而onboard LAN並沒有這問題，可以到約莫849 Mbits/sec。
- 一台筆電iperf3 -s
- orin nano devkit: iperf3 -c “192.168.100.102” -t 120 -i 10 -P 4
能幫忙看一下Intel I210在orin nano devkit的效能低的問題嗎，感謝呢。
intel_i210.txt (19.4 KB)
onboard_lan.txt (22.9 KB)
dmesg.txt (85.9 KB)
log.7z (132.9 KB)
boot_folder.7z (207.4 KB)
Could you reproduce this directly on Orin Nano devkit without a pcie switch?
Unfortunately, we cannot get rid of the pcie switch on the Intel I210 M.2 by hardware reworking.
Is there anything we can do further?
Please just try to reproduce on devkit.
Is this still an issue to support? Any result can be shared? Thanks
Does the r35.3.1 support PCIE switch? We have a problem on the Intel I210 M.2 with a pcie switch. The throughput is very low on it.
If r35.3.1 doesn’t support PCIE switch, will it support in the future release?
Thanks a lot.
There may be concurrent access with PCIe switch. It would be better if you can connect single device to each PCIe slot to have maximum throughput.
Hi @DaneLLL ,
Is it possible have a solution in the future release?
From the other post: [35.2.1] I210 throughput slow on Xavier NX
we use 32.7.1 and it does not have this kind of issue on I210 with PCIe switch.
If Nvidia does not have a plan to solve it, please let us know. Thank you very much.
I know this thread is pretty old, but I figured id post this in case it helps out someone else.
Another user on the forum sent me a patch that fixes this issue.
Its due to a errata in the Pericom PCIe switches where cut through mode does not work when the speed on each lane is different and if ACS is enabled. Setting the switch to store and forward mode fixes it.
It can be fixed in two ways. Either by a manual setpci command or by a patch to the kernel pci quirks.
The setpci operation is done against the downstream port of the bridge (connection between Jetson RP and switch). in my case 0007:01:00.0. Changing register 0x74 bit 0 to 1 fixes the problem.
sudo setpci -s 0007:01:00.0 0x74.w # take note of the value, mine was 0x0102
Write the new value
sudo setpci -s 0007:01:00.0 0x74.w=0x0103
Applying this patch is a more permanent and correct solution. It appears to have been included in kernels 5.16+
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.