Question Problems with my 4TB Samsung 990 Pro SSD ?

Toggle sidebar Toggle sidebar
adirna

adirna

Feb 27, 2025
12
0
10
I just got my 4TB Samsung 990 Pro SSD 2 days ago and i have few problems ,

First of all, when i do benchmark on the Samsung Magician software
I'm getting way lower results than what Samsung publish for this drive.
They say its should give:

RANDOM READ (4KB, QD32) 4TB: Up to 1,600,000 IOPS​

RANDOM WRITE (4KB, QD32) 4TB: Up to 1,550,000 IOPS​

But I'm getting way lower than that , so I wonder if there is a problem with my drive ?
I'm adding photo via this link since its not letting me to add it normal:
https://prnt.sc/Os0y00Uvnqmg

As you can see, I'm getting like half of what I should get on random read/write , why is that ? how to fix that ?

my spec is 12600k ,rtx 3080 ,32gb ram and ofc i have gen 4 M.2 NVMe slot.
 
Read this:

Question - How to use IOMeter like Tom's for pSLC testing ?

If there's a page already written about this, apologies and I'd like to read it. The "how we test" article doesn't mention it. I found a post from 2023 here, but nobody actually gave information on IOMeter as Tom's uses it. What are the test methods used by Tom's for IOMeter to test the pSLC...
forums.tomshardware.com forums.tomshardware.com
"tl;dr: Splitting an SSD into multiple partitions results in the pSLC cache size for each partition being split proportionally as well."

I did not know this happened.
But strongly suggests not to partition an SSD into smaller parts.
 
USAFRet said:
Read this:

Question - How to use IOMeter like Tom's for pSLC testing ?

If there's a page already written about this, apologies and I'd like to read it. The "how we test" article doesn't mention it. I found a post from 2023 here, but nobody actually gave information on IOMeter as Tom's uses it. What are the test methods used by Tom's for IOMeter to test the pSLC...
forums.tomshardware.com forums.tomshardware.com
"tl;dr: Splitting an SSD into multiple partitions results in the pSLC cache size for each partition being split proportionally as well."

I did not know this happened.
But strongly suggests not to partition an SSD into smaller parts.
Click to expand...
so you think that if i will remove all the partitions and put it on 1 drive only i will get the full speed? i mean to do it i will need to format the drive complitly and lose all the work i install windows and programs and games on it , if it will fix the problem i dont mind but if its not then all my work going to waist and will need few hours to do it again
 
USAFRet said:
Read this:

Question - How to use IOMeter like Tom's for pSLC testing ?

If there's a page already written about this, apologies and I'd like to read it. The "how we test" article doesn't mention it. I found a post from 2023 here, but nobody actually gave information on IOMeter as Tom's uses it. What are the test methods used by Tom's for IOMeter to test the pSLC...
forums.tomshardware.com forums.tomshardware.com
"tl;dr: Splitting an SSD into multiple partitions results in the pSLC cache size for each partition being split proportionally as well."

I did not know this happened.
But strongly suggests not to partition an SSD into smaller parts.
Click to expand...
I deleted one of the partition and resize the other one to be full and only partition on the drive and then i did a new benchmark , the results are exactly the same as before with no change
 
adirna said:
yes its come on last firmware out of the box , also my MB bios on last version
Click to expand...
from the web site:
  1. 2 x M.2 connector (Socket 3, M key, type 2260/2280/22110 PCIe 4.0 x4/x2 SSD support) (M2P_SB, M2Q_SB)
  2. 1 x M.2 connector (Socket 3, M key, type 2260/2280/22110 SATA and PCIe 4.0 x4/x2 SSD support) (M2M_SB)
So try a different slot. I would avoid #2 (the combo SATA/PCIE slot) because its most likely sharing the same pcie lane as the on board SATA controller.
 
IOPS on QD32 do relly on your CPU multithreading, its 32 threads afterall, and even if you get 16c/32t CPU, you still may get into 80/90% of IOPS posted by samsung on conventional PC, as those numbers (not sure if you noticed) are written as UP TO (which means in best/specific lab condition)..in real world scenario, if you dont have data center built PC, its just smoke number, or do you use your drive with 32 threaded storage software? torrent might be capable of achieving that
 
Last edited:
adirna said:

RANDOM READ (4KB, QD32) 4TB: Up to 1,600,000 IOPS​

RANDOM WRITE (4KB, QD32) 4TB: Up to 1,550,000 IOPS​

Click to expand...
usually these hardware tests OEMs do are done with a boot usb instead of an os loaded on the same drive. So your mileage may vary especially with a Windows OS.
 
dev_cyberpunk said:
usually these hardware tests OEMs do are done with a boot usb instead of an os loaded on the same drive. So your mileage may vary especially with a Windows OS.
Click to expand...
I did tests on another windows aswell that installed on another ssd, and the results are the same
 
It's better to compare those numbers with reviewer benchmarks instead of what the company says. Who knows in what conditions you can get those "perfect" scores.
 
JayGau said:
It's better to compare those numbers with reviewer benchmarks instead of what the company says. Who knows in what conditions you can get those "perfect" scores.
Click to expand...
I saw some other people on forums that are getting close to this results that samsung say you should get.
and i saw other with same results as me, i dont know if some drivers are less good then others or why is that , ‏i even saw a guy on one of the forms that said he have 3 samsung 990 pro drives on the same pc and each one of them perform different on benchmarks
 
Last edited:
adirna said:
One second ,how do you run this test with crystaldiskmark? Since its not show to me random read/write test results there
Click to expand...
go to settings, set it to nvme and then configure queu depth/threads and block size
4k block 256 queue 16 threads, then on main page where it whows your drive letter and used/available space,click on MB/s and switch it top IOPs
heres how it did test my old 970 evo
https://imgur.com/cj5OJ3p
View: https://imgur.com/cj5OJ3p

looks better than paper specs
read up to 500 000, write up to 480 000
guess some firmware update bumped it up a little
 
Last edited:
kerberos_20 said:
go to settings, set it to nvme and then configure queu depth/threads and block size
4k block 256 queue 16 threads
Click to expand...
you need to check it on

4KB, QD32​

thats what samsung say will give

RANDOM WRITE Up to 1,550,000 IOPS​

RANDOM READ Up to 1,600,000 IOPS​

 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom