Agx orin 网络性能测试 只有单核在运行

Robotics & Edge Computing Jetson Systems Jetson AGX Orin
1

agx orin 64GB，搭配pcie 4.0x 8x 100G网卡，用iperf 跑网络负载，多线程下，无法跑满网络带宽，发现cpu核心 只有一个核心 loadling为100%，其他核心使用很低，请帮忙确认是什么问题？

2

Is this on Jetpack6 or 5?

3

*** Please note that this reply is generated by LLM automatically ***

Based on the provided context, I’ll provide a comprehensive answer to your question.

The Jetson AGX Orin 64GB module is a production module that is part of the NVIDIA Jetson AGX Orin series. It has 64GB of memory and is compatible with the Jetson AGX Orin reference carrier board (P3737-0000).

Here are some key features and comparisons:

  • Memory: The Jetson AGX Orin 64GB module has 64GB of memory, which is twice the amount of memory compared to the Jetson AGX Orin 32GB module.
  • PCIe: The module supports PCIe 40x and 8x, which is similar to the Jetson AGX Orin 32GB module.
  • 100G: The module supports 100G networking, which is not explicitly mentioned for the Jetson AGX Orin 32GB module.
  • CPU: The module has a CPU, but the specific details are not provided in the context.
  • Loading: The module can be flashed with a software image, similar to other Jetson modules.

In comparison to other Jetson modules, the Jetson AGX Orin 64GB module has:

  • More memory than the Jetson AGX Orin 32GB module (64GB vs 32GB)
  • Similar PCIe support as the Jetson AGX Orin 32GB module (40x and 8x)
  • Similar 100G networking support as the Jetson Orin NX 16GB module (not explicitly mentioned for Jetson AGX Orin 32GB)
  • Different CPU and loading capabilities compared to other Jetson modules (not explicitly mentioned in the context)

It’s worth noting that the Jetson AGX Orin 64GB module is a production module, which means it is intended for production use and is compatible with the Jetson AGX Orin reference carrier board. Development modules, on the other hand, are sold only as part of a developer kit and are intended for development and testing purposes.

*** Please give the thumbs up if you get answers you like. Or provide feedback to help us improve the answer. ***

5

使用jetpack6.2.

6

Jetpack6 has removed the support of GIC MSI. Please refer to this post for patch if you need it.

7

WayneWWW，

感谢您的回复。

我们使用了旧版的jetpack 5.1.2，测试发现 负载可以分配到多个cpu核心，但是网络性能还是没有达到要求，使用orin的PCIE4.0 8x的传输带宽 发送数据带宽接近50G左右，接收数据 丢包比较严重，有50%-60%的丢包率。

针对orin PCIE C5 port PCIE 4.0 8x的读写性能，有没有demo应用和参考设备，以及参考测试数据。

9

Did you enable the jetson_clocks?

10

Wayne ,

使用 sudo jetson_clocks ，CPU频率可以跑满，但是会跑满两个核心后，其他核心不会分到负载。

实际测试的吞吐量也不正常。

网络测速
网络测速1920×1309 245 KB

12

Hi,
Please refer to this page to set IRQ to different CPU cores:
SMP IRQ affinity — The Linux Kernel documentation

You can run $ cat /proc/interrupts to check if IRQ is shifted to another CPU core.