thanks for this update. On my side the RMA (Europe) procedure was a disaster because one month later I received 960EVO unit with the wrong firmware "3xx" again as in fact no stock does exist with the non faulty "2xx".
Moreover, due to lack of coomunication between Tech Support and RMA service this service did retruned me back a first time (15 days delay) my own units telling they are working... A very non professional organisation.

Ok finally Samsung has released new FW 4B6QCXP7.

Official Statement

I've recently updated my 960Pro firmware from 2B6QCXP7 to 4B6QCXP7. And I deeply regret that. With new FW, I get 5-10s freezes (even mouse is not moving) under high load (Crystall Disk Mark is the easiest way to reproduce this) there was nothing like that with the old FW.

1. After some googling, found this post. So.. it looks like someone was able to revert the FW on 960pro.
2. Next, I found this funny "hot swap SSDs" trick, people use to fool the firmware update tool with 850-series SSDs. But ... Does it mean that it would work if I hot-swap M.2 SSD? :smileysurprised: . Anyway, I don't have second one...
3. However, here's another guy, who modified the updater, so that it skips the original firmware version check. Perhaps, any Samsung-ssd-firmware-updater-iso could be "modified" like that? Interesting chance...
4. Here's another similar story, describing slightly different "modify fw version" approach.
5. Finally, "NVMe SSD-960 PRO Firmware ISO 2B6QCXP7 | 15MB" (and only this version) could still be downloaded here.

@i3v:
You may want to RMA your SSD so that Samsung can analyze the issue and produce a better firmware and ask for a replacement with firmware 2B6QCXP7.
In their forum posts they state that they used the returned units for testing firmware 4B6QCXP7.

@i3v

Can you do some tests with MS native NVMe driver instead of Samsung NVMe driver v2.3.0.1709 ?

@Ethaniel :

1. Sure, that's one of the possibilities. And it's good to know that they do use them to study the issue. Still, it would take considerable amount of time. And there's a risk that they would just say "it's functioning OK". I prefer to avoid this process where possible. I'm sure I'm not the only one here.
2. I'm puzzled why Samsung is not officially supporting it, if it's that easy, actually. Higher chance to loose data? OK, just add "only with Secure Erase before and after".
3. I was quite angry at Samsung the day I've updated the firmware, after running all those synthetics. Now, after a couple of days with the new firmware, I'd say that the issue is probably almost invisible during my normal daily usage. SSD usage proved to be not that high, as in synthetics. So, maybe, after all, I would just keep firmware "as is".

Zitat von 100PIER im Beitrag #20

---

@i3v

Can you do some tests with MS native NVMe driver instead of Samsung NVMe driver v2.3.0.1709 ?

---

-----------------------------------------------------------------------
CrystalDiskMark 5.1.2 x64 (C) 2007-2016 hiyohiyo
                           Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
* KB = 1000 bytes, KiB = 1024 bytes
 
   Sequential Read (Q= 32,T= 1) :  3298.137 MB/s
  Sequential Write (Q= 32,T= 1) :  1268.830 MB/s
  Random Read 4KiB (Q= 32,T= 1) :   145.379 MB/s [ 35492.9 IOPS]
 Random Write 4KiB (Q= 32,T= 1) :    45.840 MB/s [ 11191.4 IOPS]
         Sequential Read (T= 1) :  2017.021 MB/s
        Sequential Write (T= 1) :  1728.685 MB/s
   Random Read 4KiB (Q= 1,T= 1) :    35.876 MB/s [  8758.8 IOPS]
  Random Write 4KiB (Q= 1,T= 1) :    34.813 MB/s [  8499.3 IOPS]
 
  Test : 32768 MiB [N: 0.0% (0.2/1907.6 GiB)] (x9)  [Interval=5 sec]
  Date : 2018/05/14 2:57:59
    OS : Windows 10 Enterprise [10.0 Build 15063] (x64)
 
 

Samsung NVMe driver v2.3

----------------------------------------------------------------------- 
CrystalDiskMark 5.1.2 x64 (C) 2007-2016 hiyohiyo                        
                           Crystal Dew World : http://crystalmark.info/ 
----------------------------------------------------------------------- 
* MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]             
* KB = 1000 bytes, KiB = 1024 bytes                                     
                                                                        
   Sequential Read (Q= 32,T= 1) :  3356.968 MB/s                        
  Sequential Write (Q= 32,T= 1) :  1287.030 MB/s                        
  Random Read 4KiB (Q= 32,T= 1) :   154.831 MB/s [ 37800.5 IOPS]        
 Random Write 4KiB (Q= 32,T= 1) :    49.808 MB/s [ 12160.2 IOPS]        
         Sequential Read (T= 1) :  2258.803 MB/s                        
        Sequential Write (T= 1) :  1677.746 MB/s                        
   Random Read 4KiB (Q= 1,T= 1) :    40.405 MB/s [  9864.5 IOPS]        
  Random Write 4KiB (Q= 1,T= 1) :    41.219 MB/s [ 10063.2 IOPS]        
                                                                        
  Test : 32768 MiB [N: 0.0% (0.2/1907.6 GiB)] (x9)  [Interval=5 sec]    
  Date : 2018/05/13 21:16:43                                            
    OS : Windows 10 Enterprise [10.0 Build 15063] (x64)                 
                                                                        
 
  
 

Samsung NVMe driver v3.0.

                                        
-----------------------------------------------------------------------
CrystalDiskMark 5.1.2 x64 (C) 2007-2016 hiyohiyo
                           Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
* KB = 1000 bytes, KiB = 1024 bytes
 
   Sequential Read (Q= 32,T= 1) :  3459.968 MB/s
  Sequential Write (Q= 32,T= 1) :  1393.466 MB/s
  Random Read 4KiB (Q= 32,T= 1) :   154.855 MB/s [ 37806.4 IOPS]
 Random Write 4KiB (Q= 32,T= 1) :    49.916 MB/s [ 12186.5 IOPS]
         Sequential Read (T= 1) :  2493.605 MB/s
        Sequential Write (T= 1) :  1731.700 MB/s
   Random Read 4KiB (Q= 1,T= 1) :    40.205 MB/s [  9815.7 IOPS]
  Random Write 4KiB (Q= 1,T= 1) :    40.488 MB/s [  9884.8 IOPS]
 
  Test : 32768 MiB [N: 0.0% (0.2/1907.6 GiB)] (x9)  [Interval=5 sec]
  Date : 2018/05/14 2:12:16
    OS : Windows 10 Enterprise [10.0 Build 15063] (x64)
 

 

Old test with 2B6QCXP7 firmware (just for reference)

-----------------------------------------------------------------------
CrystalDiskMark 5.1.2 x64 (C) 2007-2016 hiyohiyo
                           Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
* KB = 1000 bytes, KiB = 1024 bytes
 
   Sequential Read (Q= 32,T= 1) :  3454.034 MB/s
  Sequential Write (Q= 32,T= 1) :  1955.108 MB/s
  Random Read 4KiB (Q= 32,T= 1) :   664.880 MB/s [162324.2 IOPS]
 Random Write 4KiB (Q= 32,T= 1) :   439.239 MB/s [107236.1 IOPS]
         Sequential Read (T= 1) :  2626.996 MB/s
        Sequential Write (T= 1) :  1913.946 MB/s
   Random Read 4KiB (Q= 1,T= 1) :    55.179 MB/s [ 13471.4 IOPS]
  Random Write 4KiB (Q= 1,T= 1) :   212.360 MB/s [ 51845.7 IOPS]
 
  Test : 8192 MiB [N: 0.0% (0.1/1907.6 GiB)] (x9)  [Interval=5 sec]
  Date : 2017/11/19 19:32:26
    OS : Windows 7 Enterprise SP1 [6.1 Build 7601] (x64)
 
 

(You might have noticed that this one uses 8GB file instead of 32GB, but that doesn't matter much - I've just did one test with 8GB file again - the difference is below 3%.)

It looks like newer Samsung driver is a tiny bit (~10%) faster. Nothing to actually care about. Old firmware was 4x (and even ~8x) faster in some tests. And there was no "hangs", AFAIR.

Finally, I've tested each driver with and without Resource Monitor & Task Manager running. I've noticed, that if it is running - I get about 3-4 hangs per full CrystallDiskMark run, up to 5-10s. Otherwise, I get only about the same number of ~0.5s hangs. That's just a visual assessment - I've scrolled a rather long document in notepad++, up and down, by holding an arrow key. This way all non-smoothness is quite visible.

Maybe there's also some thermal throttling here. Like shown here. It looks tests finish before temps get that high, and 5sec pause between them allow SSD to cool down a bit, but I'm not sure. I don't have any active cooling there. Maybe I should add some.

@i3v
Here I describe the configuration I have:
I do use two 960 PRO 1TB devices with the last firmware version 4B6QCXP7 and do observe no problems on my ASUS Sabertooth x99 running W10 x64 RS4 v1803 Build 17134.48.
I do use Samsung NVMe driver v3.0.0.1802 to handle four NVMe SSD devices (2x960PRO and 2x960EVO) plugged onto a HighPoint PCIe x16 Gen 3.0 add in card put on a x16 Gen 3.0 slot of the machine.
All of these devices are pure data storage devices.
My system SSD is a Intel i750 NVMe 400GB.
To get the optimal data storage performance I have applied Meltdown Protection and not applied Spectre protection.
When applying Spectre protection you loss about 16% storage performance (i have measured).
So, I have no idea why you get abnormal hangs.
May be more a hardware problem within your configuration ? PCIe add in card ?

Hi,

I am on Windows 10 entreprise 1803 with 960 pro latest NVMe 3.x driver and latest 4B6QCXP7 and i've no issue so far, even with the previous 1709 Windows and 2.x driver and previous NVMe firmware, maybe it depends on mobo and bios settings used or manufacturer production lines.
I ve just test my configuration with CrystalDiskMark and emsisoft anti-malware real-time protection, spectre and meltdown patched.
(specific configuration : no pagefile, no prefetch, no superfetch, no hiberfil.sys, write cache buffering off, boot time defrag off).

Zitat von 100PIER im Beitrag #23

---

@i3v
When applying Spectre protection you loss about 16% storage performance (i have measured).

---

System is Meltdown protected: YES
System is Spectre protected: YES
Microcode Update Available: YES
Performance: GOOD
CPUID: 306F2
 

With both Spectre and Meltdown protection disabled (still using "Samsung NVMe driver v3.0"), benchmark results for 4K blocks are indeed up to ~2x better (but that's still ~2x slower than in those "Old test with 2B6QCXP7 firmware") :

-----------------------------------------------------------------------
CrystalDiskMark 5.1.2 x64 (C) 2007-2016 hiyohiyo
                           Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
* KB = 1000 bytes, KiB = 1024 bytes
 
   Sequential Read (Q= 32,T= 1) :  3477.422 MB/s
  Sequential Write (Q= 32,T= 1) :  1923.192 MB/s
  Random Read 4KiB (Q= 32,T= 1) :   369.032 MB/s [ 90095.7 IOPS]
 Random Write 4KiB (Q= 32,T= 1) :    99.435 MB/s [ 24276.1 IOPS]
         Sequential Read (T= 1) :  2681.701 MB/s
        Sequential Write (T= 1) :  1870.449 MB/s
   Random Read 4KiB (Q= 1,T= 1) :    50.813 MB/s [ 12405.5 IOPS]
  Random Write 4KiB (Q= 1,T= 1) :    78.885 MB/s [ 19259.0 IOPS]
 
  Test : 32768 MiB [N: 3.8% (72.7/1907.6 GiB)] (x9)  [Interval=5 sec]
  Date : 2018/05/19 18:36:50
    OS : Windows 10 Enterprise [10.0 Build 15063] (x64)
 

Zitat von Walls im Beitrag #24

---

Hi,
maybe it depends on mobo and bios settings used or manufacturer production lines.

---

So, for me still having the 1B6QCXP7 on my 960 Pro....am i missing something ? I could tell from reading the posts in this thread that i have the best performing version of firmware for my nvme drive.

Is that right or have i misunderstood ? 2nd version had loss of performance, 3rd is crap and 4th has regained the performance from the 2nd version ?

@felix
The current Firmware I do use for 960 PRO 1TB is: 4B6QCXP7

Hello 100Pier!

Thank you for your feedback. Is there any case you have the same benchmark with the 1st firmware so I can directly compare?

As far as I remember, my best Crystal benchmark had on sequential 3300/2000 R/W, but that is on a laptop, so I cannot compare with yours.

Thank you in advance

@felix
Yes, i have a set of tests done from nov 2016, starting from firmware 1B6QCXP7 up to 4B6QCXP7.
I don't see significant impact from v1 to v4 firmware. It does seem NVMe driver version is a more sensitive factor for performance.
My current best and stable firmware/NVME driver couple for 960PRO 1TB is: Firmware 4B6QCXP7/NVMe Samsung v3.0.0.1802.