Two 960 Evo Drives, Something Kills 4K Reads/Writes

[+Warwagon MVC]

Just Tested my Macbook Pro 2011

 

Samsung 840 EVO SSD 

Windows Creator Update

 

[adrynalyne]

9 minutes ago, warwagon said:

Just Tested my Macbook Pro 2011

 

Samsung 840 EVO SSD 

Windows Creator Update

 

You jest, but earlier with an nvme SSD, I was geting slower 4k writes.

[DevTech]

 

Laptop:

http://www.storagereview.com/samsung_960_pro_m2_nvme_ssd_review

Samsung 960 Pro NVMe

Dell XPS 9550

Skylake i7 6700HQ

16 gig DDR4

NVIDIA 960M

3200 Touch Infinity

 

 

Desktop:

http://www.storagereview.com/samsung_sm951nvme_m2_ssd_review

Samsung 951 NVMe

Gigabyte GA-Z97X-SLI

Haswell i7 4770K @ 4.2 Ghz

32 gig DDR3

NVIDIA (Gigabyte) 1050

2 x ASUS ProArt 1920x1200

 

 

Edited April 15, 2017 by DevTech
Added drive links

[adrynalyne]

3 minutes ago, DevTech said:

 

 

 

Desktop:

 

Samsung 951 NVMe

Gigabyte GA-Z97X-SLI

Haswell i7 4770K @ 4.2 Ghz

32 gig DDR3

NVIDIA (Gigabyte) 1050

2 x ASUS ProArt 1920x1200

 

 

Are you short some PCIe lanes?

[DevTech]

2 minutes ago, adrynalyne said:

Are you short some PCIe lanes?

The mobo automatically disables 2 SATA ports for the one NVME drive slot if it detects the drive which I assumed was to provide lanes.

 

I don't have any real world perception of slowness but there is a strange parallel/inversion with your situation which might be Gigabyte related.

 

I have an unused ASUS Hyper card that uses a PCIe x 4 slot (effectively the second video slot on most mobos) but no spare drive so I'm not sure if it helps my test options.

 

[adrynalyne]

2 minutes ago, DevTech said:

The mobo automatically disables 2 SATA ports for the one NVME drive slot if it detects the drive which I assumed was to provide lanes.

 

I don't have any real world perception of slowness but there is a strange parallel/inversion with your situation which might be Gigabyte related.

 

I have an unused ASUS Hyper card that uses a PCIe x 4 slot (effectively the second video slot on most mobos) but no spare drive so I'm not sure if it helps my test options.

 

Gotcha.

Well I have been running all day and its been fine. I can only point the finger at Gigabyte because thats the oonly thing I could narrow it down to.

 

Which is a pity, because some of their utilities are nice to have.

[DevTech]

11 minutes ago, adrynalyne said:

Gotcha.

Well I have been running all day and its been fine. I can only point the finger at Gigabyte because thats the oonly thing I could narrow it down to.

 

Which is a pity, because some of their utilities are nice to have.

It would be nice to narrow your slowdown further simply because no sensible explanation has fallen out of this yet.

 

Obviously the 4K Random is far more serious than my weird sequential since in the real world that is what "feels" fast with a SSD and why the first gen SSD with 100 meg sequential still felt a lot faster than a HD. And WOW, Samsung has invented an amazing chip with the 3D TLC with your 4K times beating out the Samsung 960 Pro! Their write longevity on the 3D TLC is about 10X over previous TLC so I would have no worries using one. I once killed a SSD with C++ compiles so ever since I've been super careful about write longevity of flash RAM. (I still over provision all my SSD installs with a few gigs of extra free cells just in case)

 

I'm probably repeating myself but in 2017, overclocking by easy mobo utilities is still asking for trouble and the freeware monitoring utils seem as good as the mobo manufacturers these days so you probably won't be missing anything.

 

In Device Manager if you select "Show Hidden Devices" all of the various "fake" software device drivers will show up. There is a lot of them since hardware monitoring and anti-virus and a myriad of misc crap need kernel level access and that's how they get it. I suspect that is where Gigabyte is mucking up your system.

 

[sinetheo]

Ah Trim needs AHCI to work so what you have is write amplifcation. Go Google it?

 

I would backup your data and download partedcd or partedmagic and do a secure erase to wipe all the sectors. Then set it up in raid or AHCI in the bios and recreate your volume if it is in raid mode. Unfortunately the only other fix to run trim on is to change the protocol to AHCI or Raid but then Windows won't be able to boot without a re-image :-(

 

Reimage won't clear it either as the firmware will just add more virtual sectors of o's and keep the real ones still there. Secure Erase 

[adrynalyne]

27 minutes ago, Riva said:

use HWiNFO and inspect your SSDs to see if the current link width matches the maximum link width. If not you will need to check that is configurable through your UEFI.

 

Its not a hardware limitation and I've already gone through the link settings, plus given screenshots showing what the link is.

13 minutes ago, sinetheo said:

Ah Trim needs AHCI to work so what you have is write amplifcation. Go Google it?

 

I would backup your data and download partedcd or partedmagic and do a secure erase to wipe all the sectors. Then set it up in raid or AHCI in the bios and recreate your volume if it is in raid mode. Unfortunately the only other fix to run trim on is to change the protocol to AHCI or Raid but then Windows won't be able to boot without a re-image :-(

 

Reimage won't clear it either as the firmware will just add more virtual sectors of o's and keep the real ones still there. Secure Erase 

Did you just ask me to Google something when:

1. Referring to AHCI when my drive are NVME?

2. Without asking me if I ran TRIM (and I did on both drives)?

3. Ask me to wipe my drives and install Windows (like i did last night)?

 

If it  was a TRIM issue, reinstalling Windows would not fix it.

 

I appreciate your help guys,but  I feel like you and @Riva skipped to the end and didn't read the thread.  @DevTech and I already ran through all of this.

 

Edited April 16, 2017 by adrynalyne

[DevTech]

22 minutes ago, Riva said:

use HWiNFO and inspect your SSDs to see if the current link width matches the maximum link width. If not you will need to check that is configurable through your UEFI.

 

Thanks for your pointer.

 

Although 4X is listed as max, it seems like the max is actually 2X on the bus. Those little details that makers of tech everywhere expect the  consumer to not ever notice or complain about. Sigh.

 

Samsung 951:

 

 

The bus it connects to:

 

 

So do I live with it or boot from the ASUS Hyper card?

 

https://www.asus.com/Motherboard-Accessory/HYPER_M2_X4_MINI_CARD/

 

 

[adrynalyne]

32 minutes ago, DevTech said:

It would be nice to narrow your slowdown further simply because no sensible explanation has fallen out of this yet.

 

Obviously the 4K Random is far more serious than my weird sequential since in the real world that is what "feels" fast with a SSD and why the first gen SSD with 100 meg sequential still felt a lot faster than a HD. And WOW, Samsung has invented an amazing chip with the 3D TLC with your 4K times beating out the Samsung 960 Pro! Their write longevity on the 3D TLC is about 10X over previous TLC so I would have no worries using one. I once killed a SSD with C++ compiles so ever since I've been super careful about write longevity of flash RAM. (I still over provision all my SSD installs with a few gigs of extra free cells just in case)

 

I'm probably repeating myself but in 2017, overclocking by easy mobo utilities is still asking for trouble and the freeware monitoring utils seem as good as the mobo manufacturers these days so you probably won't be missing anything.

 

In Device Manager if you select "Show Hidden Devices" all of the various "fake" software device drivers will show up. There is a lot of them since hardware monitoring and anti-virus and a myriad of misc crap need kernel level access and that's how they get it. I suspect that is where Gigabyte is mucking up your system.

 

I kick myself for wiping the other drive without further experimentation. 

 

I will keep it mind for next time.

[DevTech]

8 minutes ago, adrynalyne said:

Its not a hardware limitation and I've already gone through the link settings, plus given screenshots showing what the link is.

Did you just ask me to Google something when:

1. Referring to AHCI when my drive are NVME?

2. Without asking me if I ran TRIM (and I did on both drives)?

3. Ask me to wipe my drives and install Windows (like i did last night)?

 

If it  was a TRIM issue, reinstalling Windows would not fix it.

 

I appreciate your help guys,but  I feel like you and @Riva skipped to the end and didn't read the thread.  @DevTech and I already ran through all of this.

 

I think he was replying to our discussion about my sequential slowdown and your observation that I might have reduced lanes

[adrynalyne]

8 minutes ago, DevTech said:

Thanks for your pointer.

 

Although 4X is listed as max, it seems like the max is actually 2X on the bus. Those little details that makers of tech everywhere expect the  consumer to not ever notice or complain about. Sigh.

 

Samsung 951:

 

 

The bus it connects to:

 

 

So do I live with it or boot from the ASUS Hyper card?

 

https://www.asus.com/Motherboard-Accessory/HYPER_M2_X4_MINI_CARD/

 

 

My OCD would require me to get it.

 

However your iops are pretty nice, so...not sure it would matter too much.

6 minutes ago, DevTech said:

I think he was replying to our discussion about my sequential slowdown and your observation that I might have reduced lanes

I apologize if so @Riva and @sinetheo, I wasn't sure who you two were replying to.

Edited April 16, 2017 by adrynalyne

[kjx]

I'm going to suggest something that goes against conventional wisdom.

 

Defragment your solid state drive(s) with a good defragmenter (not the one bundled into Windows).

 

I'd highly recommend UltraDefrag (it's free). Run a full optimization on the drive.

 

Try it - what have you got to lose from doing it once? I'm not suggesting this as an experiment - I've had much success with doing this on SSDs taking them back to a fresh Windows installation level of performance.

 

[adrynalyne]

3 minutes ago, K.John said:

I'm going to suggest something that goes against conventional wisdom.

 

Defragment your solid state drive(s) with a good defragmenter (not the one bundled into Windows).

 

I'd highly recommend UltraDefrag (it's free). Run a full optimization on the drive.

 

Try it - what have you got to lose from doing it once? I'm not suggesting this as an experiment - I've had much success with doing this on SSDs taking them back to a fresh Windows installation level of performance.

 

These were new drives.

Also, the problem was fixed. 

[DevTech]

20 minutes ago, sinetheo said:

Ah Trim needs AHCI to work so what you have is write amplifcation. 

 

I would backup your data and download partedcd or partedmagic and do a secure erase to wipe all the sectors. Then set it up in raid or AHCI in the bios and recreate your volume if it is in raid mode. Unfortunately the only other fix to run trim on is to change the protocol to AHCI or Raid but then Windows won't be able to boot without a re-image :-(

 

Reimage won't clear it either as the firmware will just add more virtual sectors of o's and keep the real ones still there. Secure Erase 

If you want to "Over Provision" a SSD you need a virgin drive or else you need to restore a virgin state.

 

My understanding is that you need a special low level util from the manufacturer to get back to "factory clean" point where the drive will accept unpartitioned space as part of the available Free Cells for Write Error.

 

Just curious if you have had any experience with this and a plain old "secure erase" util would do the job. It seems to me that secure erase is writing a pattern so the drive still won't see that as a "free cell"

 

[DevTech]

@adrynalyne respectfully wondering if you could just not reply to off-topic or wrong posts at this point since your issue is fixed but people are contributing weird and interesting ideas that I'm enjoying and threads are fragile so starting a new one would just kill the flow usually.

[DevTech]

8 minutes ago, K.John said:

I'm going to suggest something that goes against conventional wisdom.

 

Defragment your solid state drive(s) with a good defragmenter (not the one bundled into Windows).

 

I'd highly recommend UltraDefrag (it's free). Run a full optimization on the drive.

 

Try it - what have you got to lose from doing it once? I'm not suggesting this as an experiment - I've had much success with doing this on SSDs taking them back to a fresh Windows installation level of performance.

 

Can you hypothesize a reason for this?

 

Did you benchmark before and after to eliminate "placebo effect?"

[adrynalyne]

8 minutes ago, DevTech said:

@adrynalyne respectfully wondering if you could just not reply to off-topic or wrong posts at this point since your issue is fixed but people are contributing weird and interesting ideas that I'm enjoying and threads are fragile so starting a new one would just kill the flow usually.

Sorry, I didn't think about that.

[goretsky Supervisor]

Hello,

 

One thing that's always good to check is whether or not the latest Samsung NVMe driver is installed.  Windows Update may not always install it.

 

Regards,

 

Aryeh Goretsky

 

[DevTech]

On 4/15/2017 at 9:03 PM, adrynalyne said:

Sorry, I didn't think about that.

Yeah, just talking about it appeared to either Collapse the Wave Function or else break the Rules of Fight Club.