Hi ,
Have anybody tried the i.MX8M Mini's PCIe performance? I installed a RTL8125 which is a 2.5G Ethernet card and tried with WiFi, I can get 1.8G at both send and receiver on TCP test. I fixed on one CPU core, I found if I don't fixed to one CPU core, the performance is lower.
CPU: I.MX8M Mini
NIC: RTL8125 / 2.5G Ethernet
OS: Linux 5.10.52
Iperf3: 3.12
Test command: iperf3 -c SERVER_IP -A 3 -w 32M
I bind the test to core 3, if I remove this parameter, the performance will be 1.7xG, a little lower. I use top to check the CPU usage, it shows about 20%.
Does this the performance I can get? Where is the limitation? Since I only use one CPU core, it's not the CPU for sure.
Does anybody tried with any device can exceed this?
BTW, I tried with Intel's i225V, but somehow it can't be recognized by CPU at all, I mean lspci shows nothing.
Thanks in advanced.
Hi @Dhruvit, thanks for your reply. The i.MX8M Mini is GEN 2.0 while RTL8125B is GEN 2.1. I used the mini PCIe connector for RTL8125B NIC. RTL8125B connects to a 2.5G port of the switch while my Linux PC server connects to 10G port of the same switch. Linux PC running a iperf3 server, the test command is iperf3 -c SERVER_IP -A 3 -w 32M
Regards,
Yanjun Luo.
Hi @Yanjun_Luo,
I hope you are doing well
Thanks for your patience. We are working on your query.
The performance of PCI is based on the number of lanes. How many PCI lanes you are using? One can try increasing the number of lanes to get better throughput.
Thanks & Regards
Dhruvit Vasavada
Hi @Dhruvit ,
As we know, imx8m mini only supports one lane, I can't use more. Also I still don't know what is the bottleneck of the PCIe. imx8m mini has 4 CPUs, I bind the test to one CPU I got 1.8G, I think it should get 2.2G ~ 2.3G, where is the restriction?
Regards,
Yanjun Luo.
Hi @Yanjun_Luo ,
Thanks for your help, it's useful information.
Hi @Yanjun_Luo