[SOLVED] NVME Drive speed problems, WD Black and Blue, help?

[Dr Bunsen Honeydoo]

HI all, I am having a bit of trouble.

500gb M.2 WD Black NVME SN750 (tried 2)
500gb M.2 WD Blue NVME
3 computers.
3 days of tests.
Both onboard and with a PCI-Ex4 to NVME card.
Windows 10 Home and Pro (not that that should matter)
Doesn't matter if it is the boot drive or the second (or third actually)

Test drives with HDTune - speed is fine
Fresh install of Windows OR clone an install that is running fine.
Now HDTune reports great speed for 1 second, then slow for 4, and that repeats for the entire drive.
Speed test with userbenchmark, on the left where it shows the read tests, the first test is fine, then the next 4 are 1/2 the speed (on the black) or 2/3 (on the blue).
Crystal Disk Mark shows the drive speed as great in every case.
Performance by experience, it is the speed of a spinning disk, or a slow SSD.

If I format the drive - problem persists.
If I secure erase the drive - the problem goes away.
I do the install or clone again and the problem comes back.

Tried with a 1TB WD Black NVME and an intel NVME, no issues.
Gamer mode in the toolbox makes no difference.
Temp reports seem a bit high, 55c on the last check, but it is cool to the touch.

Does anyone have any idea what is going on? (or how to fix it?)

 

[Darkbreeze]

Have you checked for updated firmware for that drive?

Is the BIOS firmware for the motherboards in EACH of those three systems fully up to date?

What are your FULL hardware specifications including CPU, motherboard and ACTUAL Windows build number? If you are unsure, type winver into the start menu run/search box and hit enter.

 

[Calvin7]

Problems may be encountered if your doing some kind of legacy installation instead of UEFI only with a bootable UEFI installation media with Compatibility Support Module disabled. AMD and Intel are a little different. Your post lacks your system specifications.

 

[Dr Bunsen Honeydoo]

All 3 drives, WD Dashboard reports the drives are up to date.
All 3 systems have been flashed to current bios.
All 3 systems (and the 4 other drives I used to test the blanks) running V1909
1 x X570 Aorus Master with a 3900x (air cooled)
1x Z170 with a 6700k (water cooled)
1x Asrock B450M Pro 6700k with DDR4 (air cooled)
All systems running UEFI, although I did try switching to raid on the X570
Install Media is UEFI Flash drive 1909. Clones were done with Acronis and Macrium Reflect

 

[Darkbreeze]

You have the EXACT same issue with the Black 500GB NVME as you do with the Blue 500GB NVME, but that DOESN'T happen with the 1TB Black NVME?

 

[Dr Bunsen Honeydoo]

That is correct. The exact same issue.
2 x 500gb SN750 WD Black NVME -problem occurs
1 x 500gb WD Blue NVME - problem occurs
1TB WD Black NVME - no issue.
1TB Intel 660 - no issue,
250GB Samsung Pro - no issue.
I have been trying to figure out this problem for 3 weeks now. I have gotten support from 1 other tech in my store, 1 amazon tech, and 1 microsoft tech I know, and none of them have any idea.

 

[Darkbreeze]

That's weird, because the WD Black and the WD Blue NVME drives use entirely different protocols and IIRC, totally different controllers as well. If it is a DRIVE related issue, based on model, what affects one should not affect the other. That leads me to believe that perhaps it is NOT a matter of the drive model but maybe some other factor. What factor, I am unsure of at this point, but it is very strange that both these drive models experience the exact same issues on three different OS installations in three different systems, when these drive models not only don't share much in common but also the additional confusing factor that three other drives don't have this problem.

I'll do some research but as of right now, no idea other than if there is a shared, but borked, firmware revision. Seems unlikely that drives with different controllers would have the same firmware though.

 

[Dr Bunsen Honeydoo]

I completely agree with you, that is why this is so frustrating. By using process of elimination, this shouldn't be possible. My only wild guess is that almost nobody does a full drive speed test, and this is a common problem that hasn't been noticed. Crystaldiskmark says the drives are all fine, but, the speed difference is noticeable in usage. If I hadn't been familiar with the machines, I might not have noticed. And I bet my customer would not notice either, but that isn't the point. I would think HDTune itself was the problem, but if that were the case, why does the problem go away with a secure erase? Note, diskpart/clean does not fix the problem.

Thanks for the help, and not assuming I am just an idiot .

 

[Darkbreeze]

I've invited somebody that knows a good deal more than me when it comes to storage components, so he may be able to offer some further insights.

 

[Maxxify]

There's multiple types of WD Blue SSD: WD Blue (2D) and WD Blue 3D, which are SATA-based drives, and the WD Blue SN500 or SN550 which are NVMe-based. There's also three Blacks: Black (pre-2018), Black (2018), and the SN750, which are all NVMe.

Okay so if you're using only NVMe drives, all of these drives have a small, static SLC cache, which is different from all other NVMe drives on the market. Static cache resides in the reserved/overprovisioned space of flash (outside user-accessible) and thus is limited in size based at least partially on capacity. So it's easier to fill this cache at 500GB than 1TB for example. The SN500/SN550 and Black/SN750 use the same controller (more or less) and thus cache layout - specifically they try to avoid folding (a slower state of compressing SLC into TLC), so it's very easy to outrun the cache. You can see that here as an example - it's hitting TLC speeds which are far below SLC ones.

Not sure if this is related to your issue or not but maybe that information will be useful to you. Be aware the smaller drives will drop lower than as on the graph as at 500GB you have inferior controller saturation, at least on the Black/SN750 (512GiB of flash = 8x2 dies versus 1TiB = 8x4).

 

[Darkbreeze]

We are talking about NVME drives. Not SATA M.2.

WD support staff say that the Blue and Black NVME drives do NOT use the same controller. Do you have some factual evidence to the contrary? If so, I'd like to see that, to be sure. You understand of course.

 

[Dr Bunsen Honeydoo]

I am sorry, I never said that they had the same controller, only that I had the same problem.
SN750 and SN550 are the drive models. Uploading 2 pictures to show what is happening. Partner came up with an idea that we are about to try. We bought Acronis, we have not tried with the Acronis from the WD website, we will try that and post what happens.

Edit: I also updated which motherboard and cpu are on the third system on the earlier post.

 

[Dr Bunsen Honeydoo]

CrystalDiskMark
HDTune Same Drive

 

[Maxxify]

I am sorry, I never said that they had the same controller, only that I had the same problem.
SN750 and SN550 are the drive models. Uploading 2 pictures to show what is happening. Partner came up with an idea that we are about to try. We bought Acronis, we have not tried with the Acronis from the WD website, we will try that and post what happens.

Edit: I also updated which motherboard and cpu are on the third system on the earlier post.

The SN550 uses WD's proprietary SSD controller but it lacks an IMC (no DRAM) and is only half the channels (four vs. eight) in comparison to the SN750's. The SN500 and SN750 have the same flash, and the SN550, SN500, and SN750 have the same SLC cache design (the SN550 uses 96L/BiCS4 instead). In any case, I was explaining how this design can have high/low speed bumps show up on some benchmarks like HDTune, but perhaps I misunderstood what you were asking.

 

[Maxxify]

We are talking about NVME drives. Not SATA M.2.

WD support staff say that the Blue and Black NVME drives do NOT use the same controller. Do you have some factual evidence to the contrary? If so, I'd like to see that, to be sure. You understand of course.

They're the same basic architecture. The SN500/SN550's has no IMC (no DRAM or HMB) and only half the channels (4 vs. 8). It's a cutdown version of the SN750's. The controller is proprietary specifically so it can be scaled in this manner.

 

[Darkbreeze]

You are right, the SN500 and SN750 DO use the same controller, so it seems WD support staff don't know WTF they are talking about, and that shouldn't surprise me at all.

I'd have to agree that the fact these are DRAM-less and have only a small amount of memory ON the controller itself, is likely the entire reason why these drives act differently than the other drives being compared because those drives would all seem to HAVE a DRAM cache.

Honestly though, unless you are noticing this same behavior in normal non-benchmark usage, or unless benchmarking is your goal, I probably wouldn't even worry about it. It's EXTREMELY rare that your storage devices encounter situations that are similar to those conditions imposed by benchmark utilities.

 

[Maxxify]

Yeah, I can understand why WD support wouldn't understand the distinction - they are technically different but with similar architecture. In any case they both have the same SLC cache design which is different from other consumer NVMe drives. Most consumer NVMe drives have a large, dynamic cache, even the Samsung drives are hybrid (static + dynamic). Because static SLC uses spare TLC it's limited in size based on drive capacity, so you can easily exceed the cache which makes results drop to TLC speeds. In no way should that be considered wrong or slow though...in consumer usage your working set (amount of data being juggled at any one time) is relatively small, which is why the DRAM-less SN550 actually performs pretty well thanks to the NVMe protocol.

Sean Webster here at TH shows as much in his recent 665p review as he compares it to the SN550: here you can see the SN550's (light green) tiny SLC cache and the performance drop, for example.

 

[Darkbreeze]

Makes complete sense, once you understand the reason behind it.

 

[Dr Bunsen Honeydoo]

Fair enough.
Do you have any idea why secure erase would get rid of the "HDTune test failure" but a diskpart clean or format would not?

 

[Maxxify]

Fair enough.
Do you have any idea why secure erase would get rid of the "HDTune test failure" but a diskpart clean or format would not?

Secure Erase specifically wipes the mapping table. You need a sanitize to do both that and wipe the drive. SSDs work differently than HDDs which might be some of the confusion. If you take the unpartitioned drive and run a write test in HDTune - settings for Benchmark should be set to 128 KB block size, Full test, Accurate - you should see how the SLC cache works on the drive.