Samsung 950 pro m.2 raid 0 sleep problem

Windows 10 Pro 64-bit

Edited by nme - 16 Apr 2016 at 9:39am

A curious but obvious issue, I can't say I ever checked the performance of two 950 Pro's in RAID 0 after waking from Sleep when I had that setup a while ago.

I couldn't stand the failure of the RAID 0 array if the UEFI/BIOS was cleared, or the same failure after a UEFI update. Which is why I stopped using RAID 0 with these SSDs.

I don't know what is going on after waking from Sleep, but one thing I want you both to check is the following... Hmm, just remembered the Magician software does not like RAID mode, and I can't remember if this will work when the SSDs are in a RAID 0 array. I'll mention it anyway...

On the main, Disk Drive screen of the Magician software, with one of the 950 Pro's selected (if possible) in the System Information area, the PCIe Slot information, looks like this with a single 950 Pro:



We want to confirm the Link Speed Current, and the Link Width Current, after waking from Sleep. 10Gbps is DMI3/PCIe 3.0 speed, and x4 is the lane connections, as you know.

If the Link Speed or Link Width changes, we then have a hint about what is happening.

I had a thought, put my Z170 Exteme7+ PC, with my single 256GB 950 Pro into Sleep mode, wake it, and see what I get. This is the result:



That's what it should be, so no problem after waking from Sleep on my system. You'll have to trust me on this, I would have posted the result regardless of what it was. I'm using RAID mode, but this is a single drive, no RAID array.

Since I imagine you both are using the M.2 slots, and Z170 systems use the chipset DMI3 lanes for the M.2 slots, you cannot try the UEFI option available for the PCIe slots. That is, setting the PCIe slots to Gen3 (PCIe 3.0) in Chipset Configuration. Z97 and X99 boards use the CPU's PCIe 3.0 lanes for the M.2 slots.

Grant Rogers wrote: Hi Mike, I have just stumbled across this message in search for an answer and have exactly the same problem as you and a similar system set-up. . . . I have a strong suspicion this is being cased by the intel raid drivers.  They just don't seem to handle sleep properly.  I have also noticed in the event viewer I am getting event ID 129 storahci (Reset to device, \Device\RaidPort0, was issued.) when waking up from sleep.  I am going to report this to Intel and ASRock although I am not holding my breath.

That's an interesting event, but something is strange about it. storahci is the native Windows 10 AHCI driver. The Intel IRST driver is iaStorA.sys, which you can confirm in the Device Manager entry, under Storage Controllers.

Regardless, the reference to \Device\RaidPort0 seems to be related to this situation. You should check the Event Log for anything about iaStorA.sys, if you manually installed the IRST 14.6 or 14.8 version driver, which I believe you must have done.

We also have the Samsung NVMe driver to potentially add to the list of reasons for this issue. A question that I am unsure about is, when using these SSDs in RAID 0, is the Intel IRST driver in complete control, and the only one that is used? The Samsung NVMe driver not being used at all?

Or, is the Windows 10 native NVMe driver, whose name I cannot recall right now.

OHH great, more Windows Sleep issues... why am I one of the few that does not have any...

I'll be thinking about this, and watching this thread.

I don't use sleep modes on my system but will test it later to see if I have the issue and report back tomorrow.

Parsec that clear CMOS is really f'd on this board,...that's why I stay away from it and have no issues with my RAID or bios updates.

clubfoot wrote: I concur with your findings.  IRST version 15. write back cache disabled, BIOS loaded with defaults and balanced power plan in Os. I don't use sleep on my computer as it is used for gaming,...but good to know. Before sleep: 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) :  3192.719 MB/s   Sequential Write (Q= 32,T= 1) :  1916.382 MB/s   Random Read 4KiB (Q= 32,T= 1) :   820.143 MB/s [200230.2 IOPS]  Random Write 4KiB (Q= 32,T= 1) :   640.631 MB/s [156404.1 IOPS]          Sequential Read (T= 1) :  3062.996 MB/s         Sequential Write (T= 1) :  1923.757 MB/s    Random Read 4KiB (Q= 1,T= 1) :    45.345 MB/s [ 11070.6 IOPS]   Random Write 4KiB (Q= 1,T= 1) :   162.830 MB/s [ 39753.4 IOPS]   Test : 1024 MiB [C: 36.8% (158.4/430.5 GiB)] (x5)  [Interval=5 sec]   Date : 2016/07/11 20:38:37     OS : Windows 10 Professional [10.0 Build 10586] (x64) After sleep: 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) :  1637.564 MB/s   Sequential Write (Q= 32,T= 1) :  1575.108 MB/s   Random Read 4KiB (Q= 32,T= 1) :   824.710 MB/s [201345.2 IOPS]  Random Write 4KiB (Q= 32,T= 1) :   662.792 MB/s [161814.5 IOPS]          Sequential Read (T= 1) :  1606.940 MB/s         Sequential Write (T= 1) :  1380.449 MB/s    Random Read 4KiB (Q= 1,T= 1) :    51.131 MB/s [ 12483.2 IOPS]   Random Write 4KiB (Q= 1,T= 1) :   140.071 MB/s [ 34197.0 IOPS]   Test : 1024 MiB [C: 36.8% (158.4/430.5 GiB)] (x5)  [Interval=5 sec]   Date : 2016/07/11 20:57:59     OS : Windows 10 Professional [10.0 Build 10586] (x64)

It looks like we have another data point where the wake from Sleep changes something that causes a performance loss for the 950 Pro in RAID 0.

But I find your before and after Crystal results (in text mode) easier to read than the screenshots, and I can analyze them better. Thanks for posting this

Clearly the sequential, large file read and write speeds (with one file, or with a Queue depth of 32 files) are reduced by almost 50% for the Read speed, but not as bad for Write speed.

What is curious is the 4K small file results, (with one file, or with a Queue depth of 32 files) are either virtually identical for Q=32, and actually improved for single file 4K reads! The 4K random write takes a small hit in performance.

I'd really like to know what happens here, given these results.

The reduced sequential file speeds imply a bandwidth reduction, but to not have that reflected in the 4K results is strange.

The increase in the 4K random read speed (Q=1) reminds me of the difference between one of these drives and two in RAID. In RAID 0, you loose some 4K random performance, which is normal and the biggest downside of RAID 0 performance. But a single 950 Pro should not be able to match the Q=32 4K results. So that is again, strange.

Do you see any difference in your CPU speed after waking from Sleep? Reduced CPU speed, or the loss of an OC is a classic wake from Sleep bug that many boards have. I do NOT see that on my ASRock Z170 Extreme7+ board, with an i5-6600K at 4.4GHz. I also do not see that happen on my ASRock X99 board.

So what is happening? Are the DMI3 lanes changing to DMI2 speeds? Are we loosing a DMI3 lane for some reason?

clubfoot, if you run the Magician software after waking from Sleep, what do you see about your 950 Pros, as I asked about in my earlier post with the Magician screenshot?

Also, I am not blaming the UEFI clear for the problem with the failure of the RAID 0 array of the PCIe NVMe SSDs. A RAID 0 array of SATA III SSDs will not be in a failed state after a UEFI clear. That has always been that way.

IF the PCIe Remapping option must be enabled, or the PCIe SSD RAID 0 array will fail when the PC POSTs, I also cannot call that a UEFI bug. The PCIe Remapping option is not something that should be enabled by default for the majority of users. If it was enabled by default, most users would need to disable it before normal operation of their SATA drives would happen.

Again, IF the PCIe Remapping option must be enabled, or the PCIe SSD RAID 0 array will fail during POST, that is a bad situation that may not be able to be solved with an IRST driver update. I have more theories, but they are no worth mentioning.

There is an option to lock the buses at GEN3,...but I'm thinking that is only for the PCI graphics bus as changing it to GEN3 lowers the graphics scores :( Is there a UEFI option to lock the DMI for the m.2s?

I've always enabled the remapping for my two m.2s,...are you saying I do not need to?

In my "normal" setup i.e. o/c at 4.5Ghz with C States disabled, Speed Step disabled,...my comp will not sleep :) Tomorrow I can retest the cpu speed drop from sleep to see if that is what is causing it.