Odd X25-M NCQ Issue

[Al Koholic]

Hey all, my first post here but I'm an Anandtech addict!

Anyway here's the issue:

I did a build back in December with a 160GB X25-M G2. I mistakenly installed Win7x64 while the mobo was in IDE mode instead of AHCI. Eventually I ran into issues with programs that needed AHCI and changed some registry settings to be able to switch it to AHCI in BIOS and not boot to bluescreen. Everyone knows about this with a quick search.

The interesting thing is that since doing this, I have gone into the Intel SSD Toolbox and entered the Drive Information tab. I took a look at all the features supported on SATA and see NCQ with a value of 1, indicating that it is supported.

When I scroll down a bit to the section titled "Serial ATA Features Enabled" I do NOT see NCQ. How can I know it is working?

The thing that clued me into checking this is that I have seen some stellar Crystalmark benchmark screenshots on various websites. I am getting ABOUT the right numbers but all of them are between 10 and 25 percent lower than what I see online. 4k is the worst hit.

I get that there is some variability but this seems to be a bit much especially given my strong suspicion that NCQ is off.

I postulate that in the process of switching from IDE to AHCI post-install, not all of the AHCI features were enabled so this would STRONGLY suggest people should do a reinstall anyway if it wasn't done right the first time.

What do you all think? Can't find any mention of this via searching so it could be a new issue with performing the switch.

Could anyone help by checking their drive info in SSD toolbox? I'd love some confirmation that people see NCQ in the enabled Serial ATA features section if you did a native AHCI install.

Al

 

[Emulex]

run atto with like 64 depth instead of 4 and you'll see if ncq is there.

 

[Al Koholic]

What sorts of numbers indicate that it is on or off?

 

[sub.mesa]

Use CrystalDiskMark or AS SSD; look for the random read benchmark.

With NCQ enabled, the QD32/64 random read should be ~10 times higher than the normal (QD1) random read benchmark. If the numbers are about the same, then NCQ is not working; likely due to AHCI not being enabled.

 

[Al Koholic]

Alright well, I got to the bottom of it. Sort of. I ended up doing a secure erase and reinstalling.

Turns out the QD32 test revealed that NCQ was on pre-erase. It is not supposed to indicate this in the area I thought it would in the SSD Toolbox. So we can conclude that doing the SWITCH from IDE to ACHI does not produce any problems with NCQ.

I did discover a couple of other things though that other users may want to know.

If you set up a RAID array AFTER doing the IDE-->AHCI switch and are running the latest RST (9.6.0.1014) you may be having some issues with the program not showing you all your disks, all your disk info, and so on. After the native AHCI install my RST is now working fully. It even tells me NCQ is enabled on the SSD now. Before, the SSD didn't show up, the optical drive didn't show up, and I couldn't view stats of any RAID member disks.

The second thing I noticed was that my 4K random read improved by approximately 35%. The BEST number I ever got out of CrystalMark was about 45MB/s both in IDE mode and after the switch from IDE to AHCI. Now that I have a native AHCI install, I am up to 62MB/s.

The result after running the toolbox optimizer after the switch from IDE to AHCI:

Sequential Read : 244.766 MB/s
Sequential Write : 109.373 MB/s
Random Read 512KB : 183.013 MB/s
Random Write 512KB : 108.242 MB/s
Random Read 4KB (QD=1) : 13.567 MB/s [ 3312.4 IOPS]
Random Write 4KB (QD=1) : 45.513 MB/s [ 11111.5 IOPS]
Random Read 4KB (QD=32) : 163.342 MB/s [ 39878.5 IOPS]
Random Write 4KB (QD=32) : 100.326 MB/s [ 24493.8 IOPS]

Now the result with a native AHCI after the toolbox optimizer:


Sequential Read : 246.337 MB/s
Sequential Write : 109.422 MB/s
Random Read 512KB : 174.647 MB/s
Random Write 512KB : 108.155 MB/s
Random Read 4KB (QD=1) : 14.818 MB/s [ 3617.6 IOPS]
Random Write 4KB (QD=1) : 62.602 MB/s [ 15283.7 IOPS]
Random Read 4KB (QD=32) : 162.793 MB/s [ 39744.4 IOPS]
Random Write 4KB (QD=32) : 100.475 MB/s [ 24530.0 IOPS]

IOPS are higher in both read and write on 4k random as well. Clearly there is a difference here, and when you are enterprising to fine tune a system don't keep running with the switch, do a native AHCI install. One of my native AHCI benchmark runs even hit 4.2k IOPS on random read.

So ok, issue solved. Hope it helps someone in their google searches.

I should also mention that this wasn't entirely scientific. I changed one other HUGE variable in that I underprovisioned the drive to 128GB usable space. Should be something like 22% free space on the drive now that is used by the controller to do erase-block management. According to some slides I saw on another thread here dealing with "160GB SSD Slowing Down" (or something like that), doing this up to 27% will get you increased IOPS and decrease write amplification. I like that since I'm NOWHERE NEAR/EVER getting close to even NEEDING this much space. However, this could be the real explanation for the 4k performance increase. We'll never know b/c I'm not going back to be all scientific about it...

Al