The iops data for nvme SSDs is very different from those declared by the manufacturer, what is the reason?

Question

Solid State Drives

The new server has a Samsung PM1725b MZPLL1T6HAJQ SSD, it has a Seq read. Read 5400 MB/s, Ran. Read 750k iops, but in fact it gives out only 607 mb sec, 155337 iops with Fio mixed test.
Why such a difference from the stated reading? Is this normal or is something not working right? Maybe we're measuring something wrong?
Server performance characteristics:
mother SuperMicro H11SSL-NC-B
processor AMD EPYC 7551P
Debian axis, latest proxmox distro, measurements were made on the host.

Answer 1

Bathroom, because random read + write. With such a load profile, you will never get a speed equal to the linear one. And so, that it would make sense to compare, find the specification of how the measurements were taken from the manufacturer and in the same conditions, notice at home.
The manufacturer claims

ran. Write 135K IOPS

that 607mbps / 4k block ~=155k iops
Not knowing what read / write ratio you had when testing - it seems to be true

Answer 2

Why such a difference from the stated reading?

Tested incorrectly.
It must be understood that the maximum possible value is declared under certain ideal conditions.
If the conditions are not ideal, it will be lower.
For example - such a high IOPS for this SSD is possible only with a large queue length, and only for reading.
If the queue length is less - and IOPS will be less, if instead of reading you have a mixed load - also less.
It also depends on the correct setting of the test bench - which processor, memory, OS, file system, test program, read block size, etc.
I can, if desired, get 10 IOPS on this SSD - tweak some settings and that's it.