AM625: SDK11 RT-Linux cyclictest performance is bad than SDK9.2, SDK10.0

Tony Tang

Part Number: AM625
Other Parts Discussed in Thread: AM62L

Tool/software:

I tested SDK11 on SK-AM6254 also, got similar result as AM62L

https://e2e.ti.com/support/processors-group/processors/f/processors-forum/1487312/am62l-processor-sdk-cyclictest-performance-poor-with--i1000

root@am62xx-evm:~# uname -a
Linux am62xx-evm 6.12.17-ti-rt-00771-gc85877d40f8e #1 SMP PREEMPT_RT Tue Mar 25 12:45:29 UTC 2 025 aarch64 GNU/Linux

root@am62xx-evm:/home# stress-ng -c4 --cpu-method=all &
root@am62xx-evm:~#cyclictest -m -Sp98 -i1000 -h400 -l50000000 -q > am62-sdk11-stress-methodall-i1000.txt

It is much bad than that of SDK10.0 max107us, SDK9.2 max 100us.

Should be something wrong in the RT Linux again.

2 months ago

0 Pekka Varis 2 months ago

TI__Mastermind 27050 points

It seems like this is the case, the SDK testing is not catching the benchmark degradation.

Pekka

0 Pekka Varis 2 months ago in reply to Pekka Varis

TI__Mastermind 27050 points

The automated test shows much better numbers, 41us for the best core, 73us for the worst core https://software-dl.ti.com/processor-sdk-linux-rt/esd/AM62X/11_00_09_04/exports/docs/devices/AM62X/linux/RT_Linux_Performance_Guide.html#system-benchmarks .

The difference is big enough that there must be something different in what you are running and what is run in the tests. Something else running as well or a different build.

What image are you running?

Pekka

Processors

Processors forum

AM625: SDK11 RT-Linux cyclictest performance is bad than SDK9.2, SDK10.0