my Gen4 nvme speeds don't match. Anything to try?

  • Thread starter Deleted member 143938
  • Start date
D

Deleted member 143938

Guest
I swapped my Gen3 WD Black SN750 to Gen4 WD Black SN850 but my sequential read and write benchmarks in CrystalMark don't match up other SN850 benchmarks online.

Here is what more or less every benchmark online looks like:
412507_cdm64.jpg

And here is mine:
Screenshot 2021-01-03 145217.jpg

-Could it be because they are using 1TB and I'm on 500GB version?

-Another odd thing is that on my first run my sequential read DID match up to 7000. But subsequent runs are at ~6200MB and no idea how to explain this.

(One thing to mention is that I did image the previous Gen3 drive onto this drive using Acronis True Image, instead of a fresh install. I'm not sure if that would cause this though)
My motherboard is Gen4 compatible Gigabyte B550i Aorus Pro AX and I made sure it is in Gen4 mode

Anything to try short of a fresh install?
 
Last edited by a moderator:
Benchmarking the OS drive (I'm assuming OS because of c: designation) while Windows is running from it can give pretty inconsistent results because the OS may be accessing it during the run. Also, yes smaller sizes are often slower and different mobos/systems can give different results as well.
 
Also, go under Settings and choose the NVMe SSD setting and run the test. While it may be different than what you are comparing it to, I feel that the tests it runs are more accurate to what the manufacturer are using to in their specification claims (specifically, the sequential 128K test).

I don't have the same drive as you, I have two Seagate Firecuda 120 2TB PCIE Gen 4 drives on an X570 board (in signature), so my numbers won't match the claims of the WD SN850 drive nor what you should expect on your system. But my benchmarks on both in the Sequential 128K test do reach pretty close to the manufacturer's claims (5,000MB/s write, 4,400MB/s read) with results of 4,954MB/s write and 4,247MB/s read. At least Seagate states in their specifications as to which benchmark setting they use to base their specs off of.
 
Yes. Smaller drives have fewer flash chips. And fewer chips mean less opportunity for parallel data transfers.
Specs for the 500GB are 7000MB/s and 4100MB, while the 1TB is listed as 7000MB/s and 5300MB/s. I'm not sure how full your SN850 is, but that as well as benchmarking on the OS drive can both reduce sequentials.


SN850 Specs - http://products.wdc.com/library/AAG/ENG/data-sheet-wd-black-sn850-nvme-ssd.pdf
that would definitely explain the lower write speed! That puts my mind at ease now knowing there isn't anything wrong with this drive/system as a whole
Also, go under Settings and choose the NVMe SSD setting and run the test. While it may be different than what you are comparing it to, I feel that the tests it runs are more accurate to what the manufacturer are using to in their specification claims (specifically, the sequential 128K test).

I don't have the same drive as you, I have two Seagate Firecuda 120 2TB PCIE Gen 4 drives on an X570 board (in signature), so my numbers won't match the claims of the WD SN850 drive nor what you should expect on your system. But my benchmarks on both in the Sequential 128K test do reach pretty close to the manufacturer's claims (5,000MB/s write, 4,400MB/s read) with results of 4,954MB/s write and 4,247MB/s read. At least Seagate states in their specifications as to which benchmark setting they use to base their specs off of.
I had tried and It gave me similar results for the first set of read/write with slightly greater figures for others. I chose the default setting as that's what the reviewers seem to pick so I wanted to show a setting-for-setting benchmark match, but thank you for thinking about this possibility too
 
I swapped my Gen3 WD Black SN750 to Gen4 WD Black SN850 but my sequential read and write benchmarks in CrystalMark don't match up other SN850 benchmarks online.

Here is what more or less every benchmark online looks like:
View attachment 315562

And here is mine:
View attachment 315564

-Could it be because they are using 1TB and I'm on 500GB version?

-Another odd thing is that on my first run my sequential read DID match up to 7000. But subsequent runs are at ~6200MB and no idea how to explain this.

(One thing to mention is that I did image the previous Gen3 drive onto this drive using Acronis True Image, instead of a fresh install. I'm not sure if that would cause this though)
My motherboard is Gen4 compatible Gigabyte B550i Aorus Pro AX and I made sure it is in Gen4 mode

Anything to try short of a fresh install?
yeah 500gb may have something to do with it. that and i think that's my bench on top, it's a 1TB and I just use that as a game drive and have my OS on a sata ssd. and on an x570 mb

at least your random 4k is faster! really from what i've read that more reflects real world performance? and still faster than xbox/ps ssd's so you should be good for directStorage.
 
yeah 500gb may have something to do with it. that and i think that's my bench on top, it's a 1TB and I just use that as a game drive and have my OS on a sata ssd. and on an x570 mb

at least your random 4k is faster! really from what i've read that more reflects real world performance? and still faster than xbox/ps ssd's so you should be good for directStorage.
lmao, yeah it probably is your benchmark! Yours was the first I compared the speeds to (remember, we had that talk in the Cyberpunk thread? ^.^), I then opened a bunch of the reviews for SN850 from Google's first results page and they looked more or less identical to yours so I just took a screenshot of one from a random tab of benchmarks I had opened and yours probably was the one I picked!

Previous (1TB) Gen3 SN750:
CrystalDiskMark_20201220033249.png

Current (500GB) Gen4 SN850:
Screenshot 2021-01-03 145217.jpg

The Gen3 SN750 was 1TB, it cost $135
The Gen4 SN850 is 500GB and cost $150
~250GB-300GB is the most I ever expect to use unless games grow in size significantly over the next 5 years, so I figured why not, especially with DirectStorage potential. Only $15 more, and since I didn't use the extra space from the 1TB, I was like eh, why not!
 
Last edited by a moderator:
Now you got me curious about my less than a week old 1TB SN850, and while I can't test CDM (ext4 format), I found something comparable enough and figured I might as well share. Drive has my OS on it and is maybe 3-4% full.
Also, I'd give Atto a try for measuring sequential reads and writes - https://www.atto.com/disk-benchmark/
1609722253702.png
 
MissJ84 Your random read and writes are crazy high (3rd row)..
The first bench screenshot in my opening post is from d3athf1sh who has the 1TB SN850 as you do, too, and yours are blowing ours out the water.
Did you test the right drive? in your post you say yours is 3-4% full but in the screenshot it says you are 93% full

ps is it correct to assume CPU shouldn't affect these results much? For example I am running 5600x with PBO disabled. According to your signature you are 5800x. Doubt it should matter though, right? I do know PBO affects L2 and L3 cache results significantly when benchmarking RAM in AIDA64.
 
MissJ84 Your random read and writes are crazy high (3rd row)..
The first bench screenshot in my opening post is from d3athf1sh who has the 1TB SN850 as you do, too, and yours are blowing ours out the water.
Did you test the right drive? in your post you say yours is 3-4% full but in the screenshot it says you are 93% full

ps is it correct to assume CPU shouldn't affect these results much? For example I am running 5600x with PBO disabled. According to your signature you are 5800x. Doubt it should matter though, right? I do know PBO affects L2 and L3 cache results significantly when benchmarking RAM in AIDA64.

It's not really an apples to apples test, being that I'm using a different app, although I believe ext4 tends to perform better in benchmarks like this compared to NTFS. The 93% is the empty space, though it's not showing up accurately. My root partition is 15GiB of 100GiB full, while my home partation is 20GiB of 831 GiB full. So hmmmm 35/931 = 3.75% :)

WDS100T1XHE-00AFY0 as seen in my screenshot can be found on the link above and it shows that it's the 1TB model with a heatsink.

Anyway, I agree and don't think CPU should really affect results very much, but running an NVME through say the X570 chipset can. With the exception of my memory (Micro E Die running @ 3600Mhz CL16), everything else is running stock. Needless to say I'm running Linux and my results should probably be ignored lol, but the next time I reinstall my OS, which I do somewhat often, I'll be sure to test the SN850 in Windows via CDM and reply here with the results.
 
Now you got me curious about my less than a week old 1TB SN850, and while I can't test CDM (ext4 format), I found something comparable enough and figured I might as well share. Drive has my OS on it and is maybe 3-4% full.
Also, I'd give Atto a try for measuring sequential reads and writes - https://www.atto.com/disk-benchmark/
View attachment 315634
is that in a desktop? and is it heatsinked? why don't you run crystal and post your score...? if you got time.

edit: oh and are you on x570 or b550?
 
Last edited:
is that in a desktop? and is it heatsinked? why don't you run crystal and post your score...? if you got time.
Yes, it's in my desktop and is indeed the heatsink model, but I can't test in Windows due to it being formatted ext4 and not NTFS. I do tend to reinstall my root directory semi-regularly and I'll run it next time I do this then reply with CDM results. I only have windows installed (different drive) as a backup in case some game I wanna play won't run in Linux, but the need for this is becoming more and more rare.
 
Yes, it's in my desktop and is indeed the heatsink model, but I can't test in Windows due to it being formatted ext4 and not NTFS. I do tend to reinstall my root directory semi-regularly and I'll run it next time I do this then reply with CDM results. I only have windows installed (different drive) as a backup in case some game I wanna play won't run in Linux, but the need for this is becoming more and more rare.
k
 
Yes, it's in my desktop and is indeed the heatsink model, but I can't test in Windows due to it being formatted ext4 and not NTFS. I do tend to reinstall my root directory semi-regularly and I'll run it next time I do this then reply with CDM results. I only have windows installed (different drive) as a backup in case some game I wanna play won't run in Linux, but the need for this is becoming more and more rare.
Have you compared a few games in Windows vs Linux in terms of graphical detail? I was curious to try switching to Linux and the very first game I ran (and only one in my very small game library at the moment) that was available for Linux - Layers of Fear, already had some textures that were either buggy or were low quality textures (all settings set to highest both in Linux and Windows). In Windows the few textures I noticed rendered correctly. It put me off the idea of considering Linux more seriously as a gamer. Have you noticed anything similar in your comparisons?
 
Have you compared a few games in Windows vs Linux in terms of graphical detail? I was curious to try switching to Linux and the very first game I ran (and only one in my very small game library at the moment) that was available for Linux - Layers of Fear, already had some textures that were either buggy or were low quality textures (all settings set to highest both in Linux and Windows). In Windows the few textures I noticed rendered correctly. It put me off the idea of considering Linux more seriously as a gamer. Have you noticed anything similar in your comparisons?
I run Manjaro KDE, but I'm mostly in Linux because it feels better to me psychologically as my work environment haha. Windows feels jagged and sharp and I prefer CLI over apps for most things, but more power to anyone who loves Windows 10.

I have compared a bit and while game graphical detail is mostly the same, there are some titles that run slightly better (higher frames) on one or the other. I don't play anything with high res texture packs, but friends of mine who play Skyrim as an example, have mentioned that it doesn't handle them as well and I supposed that confirms your observations regarding Layers of Fear. Moreover, I don't really play a ton of super new AAA games, so I'm probably not the best candidate to ask this stuff.

I play a lot of Civ VI and FFXIV and both feel and look similar in either OS with G-Sync enabled. The most recent title I've purchased would be Baldur's Gate 3 (early access) and the game worked natively on launch day in Linux whereas it took Larian well over a week to completely fix both Windows clients. I didn't purchase Cyberpunk, but it's another example of a game most weren't expecting to be compatible on day one, yet it was. Anyway, I'd be perfectly happy if I had to hop over to Windows to do all gaming, but it hasn't really been necessary thus far.

Linux isn't perfect. For example running two monitors can break things like G-sync unless one of them is first turned off and there are other quirks you might run into. Maybe also worth mentioning is that my Windows 10 install is on an SN750 1TB and all my games are on an EX950 2TB (Ext4). I actually debated getting a used Optane 900P for my workflow (bunch of VMs, compiling, etc) instead of the SN850, but I think I'm happy with my decision after using it for two weeks. My only dislike about the SN850 is that it uses 96-layer NAND and I may swap out for a new one if WD decides to upgrade them.
 
Back
Top