ASRock.com Homepage
Forum Home Forum Home > Technical Support > AMD Motherboards
  New Posts New Posts RSS Feed - Mixing Samsung and non-Samsung NVMe on X470 Taichi
  FAQ FAQ  Forum Search Search  Events   Register Register  Login Login

Mixing Samsung and non-Samsung NVMe on X470 Taichi

 Post Reply Post Reply
Author
Message
mapesdhs View Drop Down
Newbie
Newbie
Avatar

Joined: 08 Dec 2022
Location: Scotland
Status: Offline
Points: 65
Post Options Post Options   Thanks (0) Thanks(0)   Quote mapesdhs Quote  Post ReplyReply Direct Link To This Post Topic: Mixing Samsung and non-Samsung NVMe on X470 Taichi
    Posted: 08 Dec 2022 at 10:25am

I have a question about NVMe support on the Asrock X470 Taichi; mine has a 2700X, 1050 Ti, Win7/Pro/64, with BIOS 3.50. I have not updated the BIOS further since the support page states this is not recommended when using Pinnacle Ridge CPUs.

Until now I had a Samsung SM961 1TB in the M2_1 slot (drive M) and a Samsung 850 Pro 512GB as a scratch drive (normal SATA3 link, drive S). The main usage is image/video editing, so strong sustained sequential write speed is desired.

As the 1TB drive was nearing full, I decided to move it to the PCIE3 slot (held in a PCIe adapter card) and replaced the 850 Pro with an SM961 512GB fitted in the M2_2 slot; I then bought a WD Black SN850X 2TB to put in M2_1, but it didn't work, Device Manager shows a PCI Device for which Windows can't find the drivers. I tried forcing a driver search from the original AMD chipset drivers but that failed. I already have the Samsung NVMe driver 3.3 package installed.

I suspect what may be happening is the Samsung NVMe drivers are preventing Windows from correctly initialising any non-Samsung NVMe device. To check this, I tried the following instead which does work ok (also works if I swap round the two 1TB drives):

M2_1: SM961 1TB
PCIE3: 970 EVO Plus 1TB
M2_2: SM961 512GB

ie. when all the drives are Samsung, there is no issue with initialising a PCIE3 NVMe device, so the issue can't be simply trying to use a PCIe adapter at all as in this setup there's no problem. I have tried three different models of PCIe adapter with the same results.

I again checked this theory by testing an Adata SX8200 Pro 2TB and, just like the SN850X (whether in M2_1 or PCIE3) it does not work, the same PCI Device problem. It also fails if either non-Samsung is installed in M2_2, whereas both of these drives work ok when connected via TypeC USB at the rear port.

Ideally the config I would prefer is:

M2_1: SN850X 2TB
PCIE3: SM961 1TB
M2_2: SM961 512GB

Is this not possible because of a conflict between the Samsung NVMe drivers and the ability for Windows to correctly initialise non-Samsung NVMe devices? Or is there some obscure hardware issue at play? The PCIe link is of course set in the BIOS to be x8/x8.

Note that I know the SN850X is a 4.0 device and is thus limited to 3.0 speeds on this mbd, but I have sound reasons for the choice related to extended duration of sustained write performance (the drive has a huge 600GB cache layer, very useful when copying over large archives).

Maybe I could resolve the issue by removing the Samsung NVMe driver, but I'm loathe to do this for obvious reasons. If my theory is correct, is my only option to use all-Samsung drives if I want to keep the Samsung NVMe drivers installed?

Has anyone else had experience mixing Samsung and non-Samsung NVMe devices on this mbd? Does the same thing happen with X570 mbds?

Thanks!

Ian.

Back to Top
eccential View Drop Down
Senior Member
Senior Member
Avatar

Joined: 10 Oct 2022
Location: Nevada
Status: Offline
Points: 1440
Post Options Post Options   Thanks (0) Thanks(0)   Quote eccential Quote  Post ReplyReply Direct Link To This Post Posted: 08 Dec 2022 at 8:56pm
I have two different machines, both with Samsung NVMe (970PRO) and SK Hynix P31 NVMe, each. One is B550 Phantom Gaming-ITX and another is B450 Fatal1ty-ITX. These machines are running Win10 PRO.

Both work fine with Samsung driver installed. Device manager, Samsung Magician, and SK Hynix Drive Manager all indicate that Samsung drives are using Samsung's NVMe driver (shows up as a SCSI Disk Device, whereas SK Hynix drives are using Microsoft's default drivers.

So it's certainly possible.

Some possibilities I can think of are:
* Windows 7 quirk?
* WD drive specific quirk?

Is your Samsung NVMe driver the latest? Maybe they fixed it (or broke it) at some point.

If I were you, and I needed it to work, I'd just uninstall Samsung NVMe driver and use the Microsoft default drivers. Windows 7 got native NVMe support after SP1, I think. What are the obvious reasons you don't want to use the MS default drivers? I thought the performance differences weren't that much.
Back to Top
mapesdhs View Drop Down
Newbie
Newbie
Avatar

Joined: 08 Dec 2022
Location: Scotland
Status: Offline
Points: 65
Post Options Post Options   Thanks (2) Thanks(2)   Quote mapesdhs Quote  Post ReplyReply Direct Link To This Post Posted: 09 Dec 2022 at 3:39am

Thanks for the reply!! I think I figured out what was wrong, all working now, though quite how it came about beats me. Anyway, first to blether... :)


eccential wrote:
> ... These machines are running Win10 PRO.

I realised after posting that I do have an MSI B450 with a 2600X and Win10 I could use to compare, though in the event I didn't have to.


> * Windows 7 quirk?

I was wondering that too, but in end turns out no, at least not in the way I originally pondered.


> * WD drive specific quirk?

That was my biggest fear, having spent 170 quid on the thing, but thank grud no; seeing the same result with the Adata pushed that fear aside.


> Is your Samsung NVMe driver the latest? Maybe they fixed it (or broke it) at some point.

Yes, the usual 3.3.


> If I were you, and I needed it to work, I'd just uninstall Samsung NVMe driver and
> use the Microsoft default drivers. ...

After doing a fresh clone of the C-drive using Macrium, I decided that was the next best test. Imagine my surprise then when none of the NVMes showed up at all. :D

I did the original install using a Win7 USB slipstreamed to include the MS NVMe driver (KB2990941), even though the C-drive was just an 850 Pro. Well, turns out after it was done, the NVMe driver wasn't included in the install. Maybe it does a check and if it can't see any relevant devices at install time then it doesn't bother including the code, who knows.

Anyway, I installed the MS driver now and all the drivers are back, ran some tests, the usual expected performance loss vs. Samsung. Then I swapped the 970 for the SN850X and bingo, it was detected and runs ok, good speed as desired. I then reinstalled the Samsung driver and huzzah, the mixed drive config was properly detected, all working ok, Samsung device performance restored.

So my initial problem was caused because somehow the MS driver wasn't there, even though it was included in the USB inst. Weird. Oh well, cest la vie.

I think I'll edit my OP, add a starting note about the cause.


> ... What are the obvious reasons you don't want to use the MS default drivers? ...

Apart from the speed differences, Samsung's drivers offer a number of additional features, including the ability to continue reading from its devices even after all available P/E cycles have been used up. There's a summary of differences on Intel's site if you're interested (can I include a link in posts? Not sure).


> ... I thought the performance differences weren't that much.

I've read a great deal of test examples about this and certainly some people don't observe significant differences, but my own tests show substantial differences, at least for synthetic tests like AS-SSD anyway. Note I don't use ATTO or CDM because I think they both employ compressible data, whereas AS-SSD uses incompressible data which is more representative of working with encoded video.

Just for reference, here are some numbers for the Samsung vs. MS drivers (the presence of the Samsung driver didn't affect the SN850X results as expected), all in MB/sec (rounded), with the SM961 512GB in M2_2, SM961 1TB in M2_1, 970 EVO Plus in PCIE3, and remember I care far more about write performance even though the MS driver shows large read differences in some cases aswell:

970 E.P. 1TB:      Samsung         Microsoft
               READ    WRITE    READ    WRITE
        Seq:    2447     2415    2471     1634
        4K:    56.5    246.2    57.8    212.8
4K-64Thrd:    1714     2238     985      897
      Score:    2015     2726    1290     1273
   Combined:        5775             3195

==============================================

SM961 1TB:        Samsung        Microsoft
               READ    WRITE    READ    WRITE
        Seq:    2439     1803    2631     1774
        4K:    51.2    205.3    46.9    191.2
4K-64Thrd:    1655     1152     449      735
      Score:    1950     1538     759     1103
   Combined:        4457             2222

==============================================

SM961 512GB:       Samsung        Microsoft
               READ    WRITE    READ    WRITE
        Seq:    1223     1122    1124     1100
        4K:    53.5    152.1    39.0     99.7
4K-64Thrd:    1192     1062     491      731
      Score:    1368     1326     643      941
   Combined:        3399             1923



I have all the screenshots if desired.

Tbh I can't be certain how much the nuance matters for my real-world workloads, so far I've not had the time to conduct the kind of tests I'd like, but I do know that sometimes I have to copy from source (NVMe via TypeC) many tens or hundreds of GB in one go, which will get worse when I move up to a 4K camera next year.

It's an irony here that although the SN850X is running at half its normal potential speed, its massive cache layer means it can sustain sequential writes at Gen3 for much longer than any other drive I checked, including the 990 Pro (its cache layer is less than half that of the SN850X). Comparing dozens of sequential cache write profiles for different products, it's amazing how the exact performance can vary considerably because cache layer designs differ so much, ie. which product might be better after 1min of max speed writing may be totally different to which is best after 15 mins of writing, the total-data-written graphs cross over each other like dropped spaghetti. :D

Out of curiosity, I ran CDM on the SN850X, it gives 3649MB/sec read, 3577MB/sec write.

Thanks again!! :)

Ian.

Back to Top
 Post Reply Post Reply
  Share Topic   

Forum Jump Forum Permissions View Drop Down

Forum Software by Web Wiz Forums® version 12.04
Copyright ©2001-2021 Web Wiz Ltd.

This page was generated in 0.125 seconds.