WD SN850 Slower than expected performance

[monkeyboy46800]

Dear All,

Hope you are all well

I think we can finally put this thread to sleep.

WD admitted to me in a call that there is an incompatibility with this drive and the X570 chipset lanes.

No fix is in the pipeline:

Picked up a Corsair MP600 a couple of weeks ago and it works pretty damb close to advertised speeds in the same slot where my SN850 was crippled.

Put the SN850 in my freenas box and using it as a cache drive, no room for dodgy drives in my workstation and its the last time I will buy western digital.

Also pretty gutted Techpowerup didnt chime in and test the drive they have in their possesion considering the number of people with the issue, maybe they are under agreement with western digital with regards to the negative points of which they are allowed to communicate.

 

[Hachi_Roku256563]

Dear All,

Hope you are all well

I think we can finally put this thread to sleep.

WD admitted to me in a call that there is an incompatibility with this drive and the X570 chipset lanes.

No fix is in the pipeline:

View attachment 196633

Picked up a Corsair MP600 a couple of weeks ago and it works pretty damb close to advertised speeds in the same slot where my SN850 was crippled.

Put the SN850 in my freenas box and using it as a cache drive, no room for dodgy drives in my workstation and its the last time I will buy western digital.

Also pretty gutted Techpowerup didnt chime in and test the drive they have in their possesion considering the number of people with the issue, maybe they are under agreement with western digital with regards to the negative points of which they are allowed to communicate.

Uhhhh you had me to the last paragraph they likely dident test it cause thats not the kind of thing it does also that drive is not faulty or dogy it just had ONLY 3000mb also I I to hate company's that give me a free tb

 

[monkeyboy46800]

Uhhhh you had me to the last paragraph they likely dident test it cause thats not the kind of thing it does also that drive is not faulty or dogy it just had ONLY 3000mb also I I to hate company's that give me a free tb

The drive has 5000+ write on a cpu lane and only 3000 on a chipset lane, this dis-advantage should have been picked up by reviewers. Simple as that to be honest.......

 

[R0H1T]

The drive has 5000+ write on a cpu lane and only 3000 on a chipset lane, this dis-advantage should have been picked up by reviewers.

True but tbf that is not always picked up by reviewers unless they encounter it personally or some (other) users report it. That's how it is mostly for all kinds of product reviews.

 

[Space Lynx]

Dear All,

Hope you are all well

I think we can finally put this thread to sleep.

WD admitted to me in a call that there is an incompatibility with this drive and the X570 chipset lanes.

No fix is in the pipeline:

View attachment 196633

Picked up a Corsair MP600 a couple of weeks ago and it works pretty damb close to advertised speeds in the same slot where my SN850 was crippled.

Put the SN850 in my freenas box and using it as a cache drive, no room for dodgy drives in my workstation and its the last time I will buy western digital.

Also pretty gutted Techpowerup didnt chime in and test the drive they have in their possesion considering the number of people with the issue, maybe they are under agreement with western digital with regards to the negative points of which they are allowed to communicate.

wow... I'm glad I didn't buy one... I almost did... maybe I will go with samsung 980 pro after all... hmm

 

[Hachi_Roku256563]

The drive has 5000+ write on a cpu lane and only 3000 on a chipset lane, this dis-advantage should have been picked up by reviewers. Simple as that to be honest.......

Yeah but it's only on this specific chipset the reviewers don't test in everything

 

[raziel77]

I just bought this disk, installed windows and all my applications, but when doing tests I noticed that I am with this speed...asus z490-a cpu i7 10700k

my second disck is even worse is announced 3400 MB/s read 3000 MB/s write

Team Group Cardea Zero Z340

 

[monkeyboy46800]

I just bought this disk, installed windows and all my applications, but when doing tests I noticed that I am with this speed...asus z490-a cpu i7 10700kView attachment 196977

my second disck is even worse is announced 3400 MB/s read 3000 MB/s write

Team Group Cardea Zero Z340View attachment 196979​

Thats crap, what slot you in ? PCH Lanes I think on Z490 are still PCIE 3.0 , but even still thats is very slow !

 

[raziel77]

Thats crap, what slot you in ? PCH Lanes I think on Z490 are still PCIE 3.0 , but even still thats is very slow !

Thanks for the reply....yes i shouldn't have spent so much money on a disc that is not fully compatible, i was tricked in a store, however after installing my whole system i don't want to return the disc, anyway the speed is very low and the second disc is half speed so i have another unknown problem here. maybe something in the bios that has to change .... any ideas ??( the TG cardea is pcie3 and the writing speed is faster than reading)
what do you mean, "what slot you in" ?i only have two M2 slots on the board they are both with abnormal speeds

 

[Sihastru]

A PCIe 3.0 4x slot MAXIMUM theoretical bandwidth is 3940MB/s. NVMe SSDs incur quite a lot of overhead, especially during write operations. So, 3500/3300 R/W is the absolute best you can get on PCIe 3.0 NVMe drives. In other words, that's EXCELLENT. Calling it slow is disingenuous.

As for the drive connected to the chipset lanes, the overhead of the chipset is even greater, but you must have other devices taking up bandwidth. The chipset is connected to the CPU through a DMI 3.0 interface, that's more or less equivalent to a PCIe 3.0 4x interface. So you should, in perfect conditions, be around 3000-ish.

I would also look into installing drivers for the chipset from the Intel site.

 

[jesdals]

Well I am still happy with mine

 

[Tobiman5591]

I'm experiencing the same issue on an ASUS impact VIII. It has a stand-off card for the nvme drives. I don't know if that's part of the problem. It took me quite some time to get this drive to work with my system, even without it being my main drive. I was getting multiple BSODs over and over again. I had to reinstall windows to get this drive to show up in drive management. I bought this drive because of the advertised speeds and nothing more.

 

[raziel77]

Thanks for the explanation, the chipset drivers are update but I have a 1660 super 4g and a UAD solo plugged in, maybe that's it, I'll try to turn off the UAD

 

[W1zzard]

Also pretty gutted Techpowerup didnt chime in and test the drive they have in their possesion considering the number of people with the issue, maybe they are under agreement with western digital with regards to the negative points of which they are allowed to communicate.

I haven't talked to anyone at WD for years. I bought a retail SN850, so slow down with the accusations

What do you want me to test?

Thanks to @TheLostSwede for bringing this post to my attention

 

[Lindatje]

STILL cant belive 3000mb/s writes is BAD

Because you pay for 7000 MB/s and not 3000 MB/s.

 

[Sihastru]

@W1zzard It seems that on X570 not all PCIe lanes are created equal. "Gen.4" NVMe PCIe 4.0 4x drives will perform to spec in CPU connected PCIe lanes, but not great on chipset connected lanes. We're not sure if it's a WD problem or an AMD problem (generalized or just limited to a few boards), but I've seen reports of the Samsung 980 Pro having similar issues. We're speculating there are chipset latency issues that result in less then stellar performance for NVMe drives.

It would be great if we can have a semi-official TPU test with the WD SN850 Black, the Samsung 980 Pro and maybe a couple of last gen "Gen.4" drives (Phison E16) and a couple of current gen "Gen.4" drives (InnoGrit IG5236, Phison E18) in these two situations (connected to the CPU, then connected to the chipset).

Even some "Gen.3" NVMe PCIe 3.0 4x drives seem to suffer when connected to the chipset lanes.

 

[W1zzard]

NVMe PCIe 4.0 4x drives will perform to spec in CPU connected PCIe lanes, but not great on chipset connected lanes

I assumed that everyone knew that?

It would be great if we can have a semi-official TPU test with the WD 850 Black, the Samsung 980 Pro and maybe a couple of last gen "Gen.4" drives (Phison E16) and a couple of current gen "Gen.4" drives (InnoGrit IG5236, Phison E18) in these two situations (connected to the CPU, then connected to the chipset).

Great idea, this should be fairly easy to test, because I can use my existing data for "connected to CPU". Already working on 3 SSD reviews at the same time, so it'll be a few weeks

 

[Sihastru]

Well, there's knowing... and then there's kind of knowing but finding out on your own that the situation is so much worse and then feeling cheated. If it was a couple of hundred MB/s, that would've been fine. But it seems that switching to the X570 chipset lanes just kills performance. You lose more than 2GB/s (just above 5GB/s on the CPU, under 3GB/s on the chipset). And it seems to be affecting PCIe 3.0 drives as well. First time I've seen this I assumed it was user error, but I couldn't test, I only have a B550 board. Now I'm not sure anymore. There might be more to this story.

 

[Lindatje]

@Sihastru
Here we get the performance as indicated by WD from the SN850 with different X570 motherboards. It is not really the X570 motherboards or AMD or WD, it is more the user who does something wrong on her / his side. If it is up to AMD / WD or x570 everyone is bothered by it and I know several people with this combination who just get the WD specifications (performance) from the SN850.

 

[monkeyboy46800]

I haven't talked to anyone at WD for years. I bought a retail SN850, so slow down with the accusations

What do you want me to test?

Thanks to @TheLostSwede for bringing this post to my attention

On the German review site WD requested certain parameters not to be tested, let me dig up the link.
Good to know you bought the drive yourself and not at accusing anybody, just me annoyed with myself for spending this amount of dosh on a terrible performing drive and coming accross a bit funny with others on this forum who doubted me even after all the extensive testing I did which included purchasing other drives and making RMA.

The issue Lots of people around the interwebs are having slow writes when this drive is run on X570 chipset lanes, it bottlenecks at 3000MB/s but other manufacturers drives work just fine at close to full speed when running on X570 chipset lanes (take for example Corsair MP600).
So this SN850 drive bugs out when it encounters the X570 chipset lanes which is a shame and should be easy to fix in software but WD just replied (after extensive testing) that they have no solution at this time. Asrock also reached out to WD and WD explained to Asrock what the issue is but due to NDA Asrock were unable to elaborate further.

If somebody could test who has a bit of status in this community then it would be great.

The only positive thing that has come out of this so far is that when I did my RMA, they replaced my 1tb with a 2tb model which also performs crappy on X570 chipset lanes.

 

[Sarajiel]

The issue Lots of people around the interwebs are having slow writes when this drive is run on X570 chipset lanes, it bottlenecks at 3000MB/s but other manufacturers drives work just fine at close to full speed when running on X570 chipset lanes (take for example Corsair MP600).

Judging from your signature your board also has a SATA SSD, a TB3 port and a WiFi module attached to the PCH. Personally I'm surprised you even get that much from a NVMe drive connected via the chipset. Keep in mind that PCIe 3.0 devices still read and write to the CPU at 3.0 speeds even if the PCH is connected with 4.0 speed.
Since my new SN850 is still on its seemingly endless trip around central Europe, I can't test it in my new system yet. I'll keep you guys posted once I get my 5950X machine completed and hopefully finally up and running this weekend.

What do you want me to test?

May I suggest that you run a dedicated series of tests with at least one B550, X570, Z490 and Z590 board and both PCIe 3.0 and 4.0 NVMe SSDs attached to the PCH to get a general overview how the chipsets handle the bottleneck to the CPU. Maybe throw in a test with an active SATA connection or an active WiFi module as well.
There might be something odd going on with some X570 boards that also have PCIe 3.0 or SATA devices connected to the PCH.

 

[Space Lynx]

On the German review site WD requested certain parameters not to be tested, let me dig up the link.
Good to know you bought the drive yourself and not at accusing anybody, just me annoyed with myself for spending this amount of dosh on a terrible performing drive and coming accross a bit funny with others on this forum who doubted me even after all the extensive testing I did which included purchasing other drives and making RMA.

The issue Lots of people around the interwebs are having slow writes when this drive is run on X570 chipset lanes, it bottlenecks at 3000MB/s but other manufacturers drives work just fine at close to full speed when running on X570 chipset lanes (take for example Corsair MP600).
So this SN850 drive bugs out when it encounters the X570 chipset lanes which is a shame and should be easy to fix in software but WD just replied (after extensive testing) that they have no solution at this time. Asrock also reached out to WD and WD explained to Asrock what the issue is but due to NDA Asrock were unable to elaborate further.

If somebody could test who has a bit of status in this community then it would be great.

The only positive thing that has come out of this so far is that when I did my RMA, they replaced my 1tb with a 2tb model which also performs crappy on X570 chipset lanes.

Are you using a Zen 3 CPU? You have to have a 5xxx series CPU AND an x570 mobo for the gen4 speeds to work correctly.

EDIT: I was wrong, apparently W1zzard's own test for the review of this nvme drive was done on a slower chip and his speeds worked just fine... hmm no idea how to help you then, maybe something on your mobo just isn't working right. Most people seem to be getting advertised speeds.

 

[Sihastru]

@Sihastru
Here we get the performance as indicated by WD from the SN850 with different X570 motherboards. It is not really the X570 motherboards or AMD or WD, it is more the user who does something wrong on her / his side. If it is up to AMD / WD or x570 everyone is bothered by it and I know several people with this combination who just get the WD specifications (performance) from the SN850.

On the chipset lanes? I think you're misunderstanding the problem. If you connect the NVMe drive on the primary M.2, the one nearest to the CPU socket, that's CPU connected lanes. It works fine there.

May I suggest that you run a dedicated series of tests with at least one B550, X570, Z490 and Z590 board [...]

In this list, only X570 has PCIe 4.0 chipset connected lanes.

 

[Sarajiel]

In this list, only X570 has PCIe 4.0 chipset connected lanes.

I'm fully aware of this. Maybe I wasn't clear enough what I meant. It was kinda late at night when I made my suggestion. Let me elaborate a bit then to make my reasoning a bit clearer.

The issue of a bandwidth bottleneck with NVMe drives connected to the PCH is nothing new in general. One of my systems runs both SATA (Evo 850) & NVMe (Evo 970) SSDs on an old Z170 board and I've often observed fairly bad performance during long running concurrent operations, especially if those are independent of each other, e.g. downloading files to one drive while copying large amounts of data from the other drive to my NAS. Usually one doesn't notice the effect that much during typical activities like gaming or office work since the drive access patterns are short and often don't interfere with each other even with multiple devices connected to the PCH.

Therefore it would be interesting to see if there is a difference in handling concurrent workloads between the X570 and B550 in itself. That could hint towards a problem with the chipset. Maybe the X570 throttles the drive I/O to 3.0 speeds under certain conditions?
A comparison with Z490 might be interesting to see if there is a difference of the performance penalty between PCIe (AMD) and DMI (Intel) connections to the CPU.
The Z590 comparison is mostly for completeness sake since one would expect that a wider DMI connection would allow the usage of a x4 NVMe drive without incurring the bottleneck compared to the others.

Since there wasn't much outcry of people noticing performance issues with the first generation of PCIe 4.0 drives on X570 at the time when the chipset became available, it could just be an issue limited to the SN850 or Asrock boards. However if it's a bigger problem with the newer and faster drives, it could also affect other boards or other fast drives like Samsung's PM9A1 and 980 Pro.

 

[Sihastru]

Well, most people don't do sanity check tests. They read the spec sheet or inform themselves from a couple of reviews, then they buy the product and they assume all the benefits are there. Reviewers try to standardize their testing procedures and we end up with the best case scenario. In the real world, the product they tested and made a recommendation for will be subjected to so much more. And here we are.

It would be nice to know. People buy X570 boards because of the benefits a PCIe 4.0 chipset is suppose to offer. We all know about resources being shared in a system, it's the nature of the beast, it doesn't matter if it's AMD or Intel. But what if X570 isn't that great even when resources aren't shared? What if there's an inherent PCIe 4.0 instability in this AMD chipset? What if those pesky USB problems are connected to all this?

If it's a problem only WD or Samsung have with certain SSD models, then fine, we can buy something else. But if it's an AMD problem, there is no one else, there's only one PCIe 4.0 chipset.