Performance problem on Force 90GB SSD

Hi,

 

I have a newly built/installed PC using an Intel DP67BG motherboard, 16GB Corsair 1600MHz RAM, Corsair Force 90GB SSD on SATA II port 2 of the motherboard. Windows 7 Pro 64-bit.

 

The SSD is the system drive but the page file is on a hard disk (port 1).

 

The PC has 2 disks set up as RAID 1 (port 0) and one disk not-Raided (port 1). The BIOS is set up for RAID because of the 2 RAID 1 drives. I have noticed that Corsair recommends having the SSD on AHCI, but this is not an option for me.

 

Nothing has been done to the BIOS which is not Intel-recommended - standard clocking at 3.4GHz etc.

 

The performance of the SSD seems OK using the ATTO benchmark test. However CrystalDisk, HDTune and PassMark show a very different story, especially on write.

 

Here is the disk info which shows that TRIM is enabled:

 

http://www.irelandupclose.com/downloads/CrystalDiskInfo.jpg

 

Here is info on the disk utilisation:

 

http://www.irelandupclose.com/downloads/SSD_Space.jpg

 

Here is the ATTO result. The left hand one is from http://thessdreview.com/our-reviews/reviewed-corsair-f90-force-series-90gb-ssd-review/ and the right-hand one is the drive on my PC. My drive looks fine here.

 

http://www.irelandupclose.com/downloads/ATTO%20Disk%20Benchmark.jpg

 

Here is the Crystal Disk Mark test (again with the thessdreview test on the left). In this case the performance on my drive is terrible.

 

http://www.irelandupclose.com/downloads/CrystalDiskMark.jpg

 

Here is the HTTune benchmark which again shows very poor performance on my drive:

 

http://www.irelandupclose.com/downloads/HDTune%20Benchmark.jpg

 

HDTune Extra Tests shows very poor performance again:

 

http://www.irelandupclose.com/downloads/HDTune%20Exrtra%20Tests.jpg

 

As does the HDTune Random Access Test:

 

http://www.irelandupclose.com/downloads/HDTune%20Random%20Access.jpg

 

And finally, here is the PassMark benchmark which shows very poor sequential write:

 

http://www.irelandupclose.com/downloads/PassMark%20Benchmark.jpg

 

I am not 100% certain of this, but I did run the PassMark benchmark immediately after O/S install and I think it showed both Read and write over 250MB/s.

 

I would appreciate your help/advice!!

Due to the drop in sequential write speed in CrystalMark that I saw it is clear TRIM is not working.

 

Did you run the tests as you have presented them here? Obviously a lot of benchmarking, especially without TRIM, will degrade the drive's performance. That makes perfect sense if ATTO has the correct results but the tests later on start to show decrease in speed.

 

CrystalDiskInfo cannot tell you if TRIM is working or not. That is a list of the drive's features only. Given that win7 has TRIM enabled by default I doubt you have to check fsutil, but it is worth a shot. Most likely, the TRIM command is not being passed because you are running the controller in RAID mode. If you want to use a RAID array then you should use another controller. I don't like how motherboards only have one controller these days.

 

If you want to confirm this hypothesis, you will have to clone your drive and store the image on the mechanical drive. Secure erase the drive, and then test it while it is blank. You should get the performance numbers you want. Keep in mind, that the performance tests those reviewers (and corsair) perform while the drive is completely empty.

 

Also what driver are you using and have you tried another driver?

The tests shown here were all taken one after the other. I haven't run many performance tests at all and I have tried to keep the drive as clean as possible (for example by putting the page file on a hard disk). There is virtually no write to this disk - it's principally a boot disk.

 

The driver is the Microsoft driver:

 

http://www.irelandupclose.com/downloads/ssd_Drivers.jpg

 

'fsutil behavior query DisableDeleteNotify' returns 0 for the drive which means, I guess, that Windows has Trim enabled for the drive. But I do agree that it seems that the performance of the drive is dropping, which would indicate that Trim is not working.

 

Is there any way of cleaning dirty blocks to see if the problem is that Trim isn't working?

 

Also - even if Trim is working, this doesn't explain the drop in performance on read.

TRIM isn't instantaneous. How long between tests?

Sorry I meant the driver on the SATA controller. The drive doesn't really use a driver.

 

Try idling the computer (log off) and let it sit for a while (not sure how long) then run ATTO again to see if the performance is still in spec. ATTO is the most accurate for sequential reads and writes out of these tools because it tests it at varying data sizes. If you see abysmal performance in ATTO then you'll know that TRIM is definitely not working.

The tests were run one after the other - however there is plenty of free space on the drive and I ran the short version of the tests. I will leave the PC for a few hours and try one of the tests again.

 

The SATA driver is in fact out of date. I checked using the Intel Driver Update utility and it shows this:

 

http://www.irelandupclose.com/downloads/RAID_Driver.jpg

 

This could be the problem so I will update the driver and then run a performance test in a few hours.

 

I also found this driver for the DP67BG:

 

http://www.irelandupclose.com/downloads/IntelSATADriverDownload.jpg

 

(link to the page here: http://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&DwnldID=19632&lang=eng)

 

However this driver was not found by the update utility and I don't know if I should install it or not.

Motherboards don't need drivers. What I see in that link are the latest RST drivers, the 10.x series.

 

The Intel RST 8.6 drivers are very old! Even the 9.6 are not the newest. 10.x series was released some time ago. I don't know what program you are using to check driver versions, but it looks like crap (or is that the intel website?), eitherway.

 

You don't want the F6 versions. The F6 are for just that..hitting F6 before installing windows.

 

Also, try rolling back the driver or pointing it to the Micorosft AHCI 1.0 standard driver. You can get some pretty good performance with that one (that's what I use).

I've installed the Intel 9.6 driver and left the system running for several hours. The PassMark test still shows sequential write at around 67MB/s ... so no improvement there.

 

I checked for the latest driver download using the Intel Driver Update Utility. I've looked for more up-to-date drivers on the Intel download site but can't find one ... if you have a link I would appreciate it.

 

I can't go back to AHCI because I need to use RAID 1 on my hard drives for security reasons.

 

Is there a program available that can 'clean' the SSD without wiping the O/S? If I could do this manually then at least I would be able to be sure that the problem is with TRIM not working ... at the moment I'm pretty much in the dark.

Thanks for that info - I am trying to find out if I can get the Diskeeper with HyperFast as a trial. I would have to buy the Pro Premium version as I have disks over 1TB ... a bit expensive, so I would like to try it out first.

 

One rather strange thing is that CrystalMark Disk Info shows a power on count of 212 ... which is WAY more than the SSD has ever been powered on. It may have been powered on 50 times perhaps, but no more.

 

Is it possible that this disk is defective?

Well in your sticky, on the first screen shot, you show the power on count and power on time and you seem to think these particular figures are accurate :biggrin:

I have run Diskeeper with defragmentation and Hyperfast and there is absolutely no difference in performance:

 

http://www.irelandupclose.com/downloads/Post_Diskeeper.jpg

 

What now? Suggestions most welcome!

It does seem that the different test utilities report very different results. Perhaps a more valid test would be to read / write a file (or files) from a Ram drive and use a stopwatch?

 

If I need to install Diskeeper before the degradation starts, what do I now need to do to get the disk back to its pristine condition? Should I for example take a disk image, wipe the disk clean (using what utility?) and then restore the system from the disk image?

 

Do you have a procedure for doing this?

 

And if TRIM has been working all along and I have done very little writing to this disk (except for the O/S install), why would it have degraded so quickly (in a matter of days)?

Well I've set up a RAM disk and written the following to the SSD:

 

#1: 2.53GB (11 files): around 9 seconds or 250MB/s. Not bad!

 

#2: 3.33GB (14 files): around 22 seconds or 150MB/s. Not brilliant, but OK.

 

#3: 2.12GB (224 files): around 3 seconds or 706MB/s. Clearly partly cached.

 

#4: 3.43GB (284 files): around 22 seconds or 155MB/s. Not brilliant, but OK.

 

These figures are quite repeatable. Is seems pretty obvious (to me that is) that system caching is hiding the poor performance that can be seen as soon as the transfer size goes above a certain figure. BTW - I'm just using Windows explorer. I've also tried it from the command line.

 

It's easy to see that in the case of a transfer with 14 files that the first 8 or so seem to go very fast (~ 1 second) and then the next 5 get progressively slower for an overall transfer rate of over 20 seconds.

 

The RAM disk itself (QSoft) is very fast with a transfer to itself of 3.4GB taking 2 seconds or less (that is read 3.4GB and write 3.4GB).

 

Now for something worse really:

 

#5: 3.57GB READ to Ram Disk: 21 seconds or 160GB/s

 

compared to same 3.57GB read from RAID 1 Caviar Black at 20 seconds!

 

I'm going to try a 4GB ISO image as per the post after this from Wired, who corrected my miserable arithmetic ... thanks!

Those should be ~150 MB/s, not 15 MB/s. Also note that the more files you have the more OS overhead there is. Try copying a 4 GB iso image and see what happens.

What's a RAM disk? A disk made of RAM modules or a paging file?

 

I would not recommend putting a paging file on a SSD.

OK - 1 4GB write takes about 11 seconds, or 360GB/s

2 simultaneous 4GB writes take about 28 Seconds, or 280GB/s

3 simulaneous 4 GB writes take about 45 seconds, or 265GB/s

 

All very good and all clearly showing that data is cached (it can be seen quite clearly with the second write stopping the first, then the 3rd stopping the second, and when these complete the 1st resumes and completes ... but of course I don't when when the 1st actually completes - that is, all the data in the cache has been written).

 

At any rate it would seem to me that on the whole the performance of the drive is OK.

 

I would be interested to know why Corsair uses the ATTO benchmark. Does this do low-level un-cached reads and writes? Does it give an accurate measurement of the drive performance, bypassing O/S overheads as much as possible? Is there a version of this test that can be used standalone so that Windows is taken out of the equation?

OK - 1 4GB write takes about 11 seconds, or 360GB/s

2 simultaneous 4GB writes take about 28 Seconds, or 280GB/s

3 simulaneous 4 GB writes take about 45 seconds, or 265GB/s-

 

That was one 4 GB file?

Did you notice in thessdreview cyrstal disk mark results that their data is zero filled and completely compressible? By default crystaldiskmark does random incompressible data. You are comparing your incompressible results to compressible ones. Turn zerofill on in your crystaldiskmark and you will get the same results as the thessdreview. The sandforce SSD compresses data and does not do as well with incompressible data.

 

You have to be aware of what data the various bencmarks are using. ATTO and hdbench both use zero filled data which really does not tell you anything about real world performance with SSD. By default, crystaldiskmark uses random uncompressible data. Real world performance is somewhere in between , but much closer to the crystaldiskmark results.

Hi, I have got a similar problem with my F90.

 

I get 149ΜΒ/sec seq read and 127mb/sec write and at 4K i get 19mb/sec read and 57mb/sec write using AS SSD benchmark. The disk is brand new.

Hi, I have got a similar problem with my F90.

 

I get 149ΜΒ/sec seq read and 127mb/sec write and at 4K i get 19mb/sec read and 57mb/sec write using AS SSD benchmark. The disk is brand new.

 

Read my post above yours. Your numbers are consistent with how the drive will perform with incompressible data.