• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

Low random writes Samsung 990 Pro

I was having the exact same issue and I read in another forum that you have to change the mode to NVE and that gave me normal results.
As user The_Enigma pointed out, you can tell that you have it in the wrong setting because the label reads "Q32T1" instead of "Q32T16". Switch the mode, and the label will switch as well

In CrystalDiskMark, go to "Configuration" and select "NVMe SSD".
 
It hit 100% during RND4K Q32T16 whilst reading. But it was only 20% during writing (which is the lowest expected result I got)
So you saturating CPU, is this on your 7600X? I would reconfigure that test to run to match the CPU capability so T12, can still be Q32, so Q32T12, instead of T16. Score seems low for a 7000 series Ryzen still, but the 100% is confirmation you was at least bottlenecked on the rand 4k reads.

Was it actually running at Q32T16? as your screenshot was blurred.
 
Hi,
Here's bone stock acer m.2 that came with in my new laptop and a new WD sn850x 4tb I added
Both have heatsinks now

1702473410317.png
 
I also would advice to upgrade your 990 PRO firmware, your are two versions behind. And this firmware is known to degrade your 990PRO after some time.
I had the same drive as you, it did upgrade right out the box to 3B2QJXD7. Can u show the screen of Magician, i think your are wrong. All the ones i got in the shop did upgrade to this new version. You say it is latest version now, that would be the first i hear then. It's possible if Samsung changed the NAND. But all new 990PRO from 1TB, did upgrade out the box to this new firmware.

With version of Magician do you use? What production date has your 990pro?

Screenshot 2023-12-13 220145.png


There are new versions of the 990PRO available, but they have the firmware 0B2QJXG7. This is the latest NEW firmware and there is no newer firmware available for these series. See the difference in the firmware number? The letter D changed in an G. Only drives with the new V8 process get the firmware 0B2QJXG7, again there is no newer firmware for these new SSD!!
You have the older 990PRO version SSD and firmware, so it should upgrade to 3B2QJXD7. If it does not work, you either use an old version of Magician, your drive is affected by the locked error or you are doing something wrong. It could be you have a new NAND, but i have not seen any here or on the internet. If Magician keep saying it is the latest, you should download the ISO file and boot with this Linux version to upgrade.
 
Last edited:
I have the 4TB 990 Pro..the one with the new V8 process.

My random writes are a bit slow. Q32T16. Not sure why. Everything else is fine. My random writes have never been good. Not sure what else to do. Like I said this is a brand new drive on the new firmware and new V8 process drive. Samsung magician is all correct. Using this as my C: though. Still random Q32 writes should not be this low. I am on NVME in crystaldiskmark. Peak performance profile gives worse results actually. I could change the thread count to 22 or something I guess. Odd though.

CrystalDiskMark_20231213174948.png
 
But your first screenshot sows a 1TB 990PRO with the OLD firmware? This one should be upgraded ASAP!
 
Ah yes, i need glasses... My bad... :)

The 4TB 990 Pro isn’t as quick as the 2TB or 1TB, and this might be chalked up to the flash on the 4TB, which, while new, actually has a higher listed 4KB read latency as it is documented. This is not a bad trade-off for twice the flash density, which will increase latency from an extra addressing cycle.
Like the WD Black SN850X uses 4 dies, whereas Samsung only uses 2 dies for the same, so only single sided versus WD double sided and also needs more space.

Samsung is the only that need two chips for 4TB capacity, while all other's need 4 chip's for that, with consumes more power en generates more heat.
But RND speed's are a bit slower for the bigger capacity ones.
Anyway even when it is much faster, you will not see any difference when working with it... And it stays the fasted PCIE 4.0 drive out there.
 
Last edited:
So the funny thing is I just bought my 4TB drive 3 weeks ago and it was on firmware G7. Now today I got an update in magician software to update my firmware. And now it is at firmware rev D7.

Not sure how my drive went from G7 to D7 today.
 

Attachments

  • Screenshot 2023-12-14 113803.png
    Screenshot 2023-12-14 113803.png
    73.1 KB · Views: 143
That letter plays no role really, the most important is that this a firmware update for the older AND new versions of the 990PRO. The firmware goes now from 3B2QJXD7 to 4B2QJXD7.
Told you earlier that you not had the latest version Firmware. You're good to go now!

For people with a NEW 990PRO, Samsung made this firmware compatible with the versions made in V8 process!
So for anyone with an 990PRO and firmware 0B2QJXG7, this updates it to 4B2QJXD7!


Screenshot 2023-12-14 224127.png
 
Last edited:
So I had firmware with the G revision. Which means my drive is the NEW drive, not the old one.

The firmware that was released 2 days ago changed the letter from G to D. G was the new process drive. Which I had. Now the firmware just lists D.

I did have the latest version. This firmware was just released. Not sure if you know that. Samsung magician just found this firmware 2 days ago. I had 0B2QJXG7. And the firmware that was just released 2 days ago changed it to 4B2QJXD7.

4B2QJXD7
this firmware was just released!
 
Did you update the firmware via samsung magician? 990pro needs a fw update out of the box to correct a couple things.
lol, it's good for any ssd.
 
So I had firmware with the G revision. Which means my drive is the NEW drive, not the old one.

The firmware that was released 2 days ago changed the letter from G to D. G was the new process drive. Which I had. Now the firmware just lists D.

I did have the latest version. This firmware was just released. Not sure if you know that. Samsung magician just found this firmware 2 days ago. I had 0B2QJXG7. And the firmware that was just released 2 days ago changed it to 4B2QJXD7.

4B2QJXD7
this firmware was just released!
Samsung has merged both the older 3B2QJXD7 and newer 0B2QJXG7 firmware into one for all versions of the 990PRO!

So now for any version the latest firmware is now 4B2QJXD7.

Screenshot 2023-12-14 225151.png
 
Last edited:
Even if it was double fast speed, you won't see or feel it anyway in reality when working with it. You will see no difference at all.
And random writing almost NEVER happens in Windows or Linux for a very long time as in this test... It's mostly just read and writes but not gigabytes at random, even in games this does not happen.
 
Last edited:
It is not performing per spec of the drive and what other reviewers are getting in the same benchmarks I am running. I don't care about woulda coulda shoulda. My drive is underperforming in Q32 random writes in multiple benchmarks. That is a problem and it should not be happening at all. The scores should beat 6k or 5k...not at 3k. That is an issue no one knows why. 850k random write IOPS is WRONG. The drive is capable of 1,550,000 random write IOPS. This is a problem no one is addressing. Other reivewers are not getting those slow speeds at all. I am wondering if I should RMA my drive.

CrystalDiskMark_20231216051908.png


The highest random write IOPS I am getting with 22 threads is 902,000k....not 1,550,000 IOPS as the DRIVE SPEC. Everything else is working fine.

I am thinking either my drive is not working right or further firmware updates are needed.
 
Last edited:
Keep in mind that SSD speeds are also dependent on CPU, RAM speed, system etc

In the past, Intel systems would to bench faster than AMD storage wise, this might still be the case.

I have a feeling that a single CCD AMD could have slower speeds than a dual CCD.

Don’t forget, the drive specs are all “up to”.

Differences in motherboard, firmware, power management implementation, power plan, C-states, CPU PCIE lanes, chipset PCIE lanes, drivers, VMD / RAID enabled (even if not used or configured), Windows VBS, if Windows installed to the same drive, temperature / cooling and much much more could impact SSD bench speeds.
 
Last edited:
There can also be a huge difference in performance from using an old install of Windows compared to a fresh install.

My Samsung 990 PRO 4TB for example, became a lot faster from just re-installing Windows 10.


Old install of Windows 10 (from November 2021):

Performance Benchmark.png


And here is a fresh install of Windows 10 (from December 2023):

Benchmark (Clean Install of Windows 10).png


In both cases the 990 PRO 4TB is installed as a secondary drive, and not as the Boot-drive (drive is also empty in both screenshots).

But despite it being a secondary drive, there was still a huge difference in performance after a fresh install of Windows 10.
 
Last edited:
Capture d'écran 2024-01-08 152315.png


I have the same low random writes issue. Was getting arround 1 000 000 IOPS after the initial windows setup, now after a few updates and softwares installed I’m at 650 000 iops without XMP (Magician bench, CDM with NVME parameter, AS SSD same results). Sequential speeds seems normal (read 7100 / 6820mo/s write)
  • MSI Z790i, DDR5 32gb 6400, 13700K (no PL/UV)
  • Samsung 990 pro 4To, initial firmware was 0B2QJXG7, updated to 4B2QJXD7. It's running on the cpu M.2 chipset of the board at pcie 4.0x4
  • Windows 11 pro 23H2, last to date drivers and bios.
- Enabled « full performance mode » in Magician → Didn’t change anything.
- Closed most of the background process and network for the benchs
- Enabling XMP gives even worse random writes 450 000 IOPS (both confirmed in Magician and CDM)
- Processor at 100 % during sequential writing, arround 50 % during random writing. Disk temps are normal at 35 – 45°C.

I don’t know what’s wrong and what the lower score with XMP enabled even means. I haven’t run tests for the memory stability with xmp yet, is it bad sign somehow?
I guess since there are differents users reporting the same issue for the drive, it might not be hardware related but a firmware issue ? My random read speeds are equal to the advertised ones.
Also, not sure if that’s normal but I was at 400go written on the ssd after windows install and a few programs. I’m now at 4,5To after maybe 10-12 benchmarks runs, do they write that much data?
 
I am also unsure why reads and writes are so high as I barely used the PC yet. And I didn't test/benchmark for more than 30 minutes total.
I run the test (3x, 1GiB) and count the writes/reads...
A single run can add 150~200GB writes and 200~250GB reads
If you run the bench like 10 times... well...
 
There are users on other hardware sites reporting the same issue. Not sure what is going on. We should not be getting sub 1million random write IOPS at all. When the drive is rated at 1,550,000 random writes (4TB version which we have). We shoud be at least getting 1,3 or 1,200,000 million random write IOPS. I have increased my thread count to 22 and that helped get me 900,000 but it is still low. This is on a completely fresh, latest drivers install of win 11. And the latest bios on my motherboard. But like I said, other people are getting the same poor performance. Maybe a firmware update will help this.

And regardless of whether or not the drive is a system drive with an OS on it or a secondary drive, the drive is rated at 1,5500,000 million random write iops. Wost case we should be at least getting 1,100,000 as a system drive. There is something not right. Sorry. Other users I see getting the speeds on lower sized drives from benchmarks about a year ago. But now is now. We need to figure it out.

And my system is not the issue. I have a 7900X cpu, 32gb of 6000mhz rated ram with xmp enabled, 4080 video card, 1000watt psu and a good asus motherboard. My 990 pro is plugged into my gen5 nvme slot and it's set on auto. HWINFO32 reports everything pristine. Everything in my system is rock stable and should be no issue to this 990 pro getting the full speeds. Others have the same issue as here. There is something wrong with the 4TB version and random write iops. I've stopped running benchmarks because I no longer care but sucks we are not getting the full potential of the specs advertised.
 
Last edited:
The firmware that was released 2 days ago changed the letter from G to D. G was the new process drive. Which I had. Now the firmware just lists D.
You read a version number left to right. Start with the first digit, not some random letter second to last.

And regardless of whether or not the drive is a system drive with an OS on it or a secondary drive, the drive is rated at 1,5500,000 million random write iops. Wost case we should be at least getting 1,100,000 as a system drive. There is something not right. Sorry. Other users I see getting the speeds on lower sized drives from benchmarks about a year ago. But now is now. We need to figure it out.
The difference of having an os on the drive or not, is that the os can be doing all sort of things in the background while you are benchmarking. The rated max value is measured in a lab with a factory fresh drive. The second you put data on it this value starts dropping. The spec sheet listed by Samsung says UP TO, not minimum/average/guaranteed/whatever. At QD1 the drive is rated at UP TO 80,000. Your results are between those two extremes, leaning heavily towards the max value listed at QD32. I'd be satisfied. (And truthfully, what workloads are you running that will need >900,000?)

I don't agree with how they report drive specifications, but this is the status quo. Marketing speak wins the day. If they measured 1,550,000 once in a lab, they can 'prove' that UP TO is correct.
 
It's rated at 1,550,000 NOT WE DID THIS ONCE PER DAY IN A LAB. There is something wrong with the drive with the low writes. Sorry you don't believe that. It should not be that low. And historically drives rated get close to their rated speeds. All my ssds did. Now is that suspect as well? No.

The fact that the 2tb version have benchmarks all over the internet getting close to the rated speed doesn't tell you anything? 4TB has issues. Look up other lower capacity 990 pro drive benchmarks. They are working fine and getting the speed.

I don't need to be running anything doing anything at all. Samsung needs to sell the drive operating at specs they advertised. I am not going to comment any longer here. The drive has issues. Sorry.
 
It's rated at 1,550,000 NOT WE DID THIS ONCE PER DAY IN A LAB.
No it is not. It is rated at UP TO. But go ahead, blow a gasket over this. If you are this unsatisfied with your new drive, I strongly suggest that you return it and buy a different product.
 
I run the test (3x, 1GiB) and count the writes/reads...
A single run can add 150~200GB writes and 200~250GB reads
If you run the bench like 10 times... well...

So this is normal then, i was getting worried when i noticed reads/writes that were that high from just a single run.

Even made a whole Thread about it lol


It's rated at 1,550,000 NOT WE DID THIS ONCE PER DAY IN A LAB. There is something wrong with the drive with the low writes. Sorry you don't believe that. It should not be that low. And historically drives rated get close to their rated speeds. All my ssds did. Now is that suspect as well? No.

The fact that the 2tb version have benchmarks all over the internet getting close to the rated speed doesn't tell you anything? 4TB has issues. Look up other lower capacity 990 pro drive benchmarks. They are working fine and getting the speed.

I don't need to be running anything doing anything at all. Samsung needs to sell the drive operating at specs they advertised. I am not going to comment any longer here. The drive has issues. Sorry.

I have the 4TB model myself, and i get over 1,000,000 IOPS on both Reads and on Writes as you can see in post #43

I did have to reinstall Windows to get those numbers though
(you can see just how much of a difference it made too).
 
Back
Top