slow memory I/O on AMD EPYC 9334

[Gero <gputzar@beck.engineering>]

Dear experts,

thanks a lot for your work and commitment on the Debian system. I'm 
using Debian for years and I am generally very pleased with it.  :-)

With my company we do numerical simulations and recently did some 
benchmarking tests on new AMD EPYC 9334 processors that showed a 
significant performance loss of a current Debian system compared to an 
older Red Hat or Rocky Linux. We could narrow that down to the following 
finding:

Running these commands:

cd /dev/shm
echo 3 > /proc/sys/vm/drop_caches
fio --name=random-write --ioengine=posixaio --rw=randwrite --bs=4k 
--numjobs=1 --size=4g --iodepth=1 --runtime=60 --time_based --end_fsync=1

on an old Rocky Linux 8.7 results in ~1000MB/s. On a current stable 
Debian 13.1 the same results in ~250MB/s, on the very same computer. I 
tested it a couple of times, also a year ago with the then current 
stable Debian with the same outcome. (So I'm pretty sure this is real...)

The computer has an Asus K14PP-D24 Mainboard with two AMD EPYC 9334 
32-core CPUs and 24x 32 GB RAM. The Rocky Linux 8.7 kernel is 4.18.0.

I would like to know if you have an explanation or an idea. And I wonder 
if you would be interested in investigating the issue any further. Or if 
you have a suggestion who I might address preferably.

Thanks in advance, kind regards,

Gero Putzar