Jump to content
Corsair Community

Experiments with a peir of X32s


Cadencia

Recommended Posts

I bought a couple of 32GB extreme SSDs for no other reason than that I wanted to answer some questions for myself. Here are my conclusions. The evidence, and the long story may follow, although I probably won't upload ALL the benchmarks, nor detail all the false starts and failing motherboards I passed through. Suffice it to say that I punished these two SSDs on 4 different motherboards, in various configurations, over 4 days of concentrated testing.

 

The SSD devices purport to be at firmware level 2.0, and declare that they suppport TRIM. As far as I can tell, they don't. There is no published information about the differences between firmware 1.0, 1.1, and 2.0. I wish there were!

 

Question 1: Is there any benefit in using a non-standard NTFS allocation unit?

Answer: No. New straight out of the box, both disks gave ATTO benchmarks with a 4K write speed of 79198 and 77982, and a 8MB write speed of 93858 and 94187 respectively. One of the disks was formatted with 32K NTFS allocation unit, the other had the 4K default. Several successive benchmarks showed only minor differences, with no pattern of increased performance for either formatting scheme.

 

Question 2: How much degradation occurs as the device gets "used".

Answer: Not that much. Maybe it's this level of firmware. The slowest I could get the disks to benchmark repeatedly (excluding weird results when some other factor influenced the benchmark) was 4K write 68642, 8K write 81241, 8M write 91929. I did have a couple of runs where the disk slowed markedly, down to 55296, 69808, 76368. This was after I had been using Acronis to write all over the disks. At the same time I had to upgrade the BIOS of the motherboard, so I think those tests should be ignored. So, fastest repeatable 8M rate: 97083 write, 212622 read. Slowest repeatable 8M rate: 91929, 191739.

 

About 5% for the big writes. The 4K writes showed a much greater variance - about 30% at the extremes. The 0.5K writes showed no significant variation.

 

Question 3: Does using Acronis to write all x'00' or x'ff' all over the disk restore performance?

Answer: No. Either causes the disk to slow down, as does all the other variations of 'erasing' done by Acronis

 

Question 4: Does "wiper.exe" help?

Answer: No. I could not get it to recognize any of my drives (the 2 x32s and my system x256, firmware 1.1) on any of the four computers I had available.

 

Question 5: Does using hdparm to do an ATA security erase restore performance?

Answer: Yes. It restores performance to exactly as new. It takes very little time. The bulk of the time is spent waiting for the Ubunto CD to boot itself up, then the Acronis CD to boot itself up for the restore.

 

Question 6: Is there a performance advantage in 'aligning' the partition to a 4K boundary?

Answer: Yes, but not as much as I expected. 4K writes 74555 aligned, 68472 unaligned, 8M writes 96213 aligned, 93045 unaligned. For this test, the disks were aligned by doing a security erase, then installing windows XP to one drive, Windows 7 to the other, and allowing Windows to choose the alignment. I then ran ATTO in the respective operating system, and followed up by moving the disks to a 'non-system' situation and rerunning the tests. Conclusively, aligning helps performance a bit.

 

Question 7: Is there a performance advantage from running 64 bit Win7?

Answer: There's no disadvantage. I did a few tests, but saw no particular increase, although it 'felt' faster.

 

Qestion 8: What happens if you put 2 drives into a stripe RAID?

Answer: About double the speed on write, a bit more on read, using the default stripe size. Using a 4K stripe made reads slower than writes,, max write speed 133883 MB/s. Using a 128K stripe size made reads much faster, max write speed 184280, max read speed 288020. But 4K writes and reads slower than single disk. Given the difficulty of restoring performance as usage increases, I will not be using RAID for myself.

 

If anyone has another test they would like me to try, ask and I'll give it a go.

786009251_XPformat.PNG.8fc3988361e219ccc78124f5a023de7d.PNG

161715861_Win7Format.PNG.b8cf4f13e135efb8852c57acdbd938f5.PNG

Link to comment
Share on other sites

I finally found a method of restoration that appears to work for me. I use Parted Magic and boot with the SSD disconnected. After fully booted up I reconnect the SSD and run secure erase which appears to take a second or two.

It take longer to take the case apart to disconnect the SATA cable and then reconnect it after the boot than secure erase takes itself.

Without performing any other type of format, etc., I boot with the Acronis boot CD and restore the drive, selecting both W7 partitions and restore MBR. This takes just a few minutes, maybe 7 or 8. The drive is now back to where it was before running the procedure, except the performance numbers are exactly the same as the ones in your screen shot.

Through all of the trials and errors this is the single, only method that appears to work.

Link to comment
Share on other sites

Question 1: Is there any benefit in using a non-standard NTFS allocation unit?

Answer: No. New straight out of the box, both disks gave ATTO benchmarks with a 4K write speed of 79198 and 77982, and a 8MB write speed of 93858 and 94187 respectively. One of the disks was formatted with 32K NTFS allocation unit, the other had the 4K default. Several successive benchmarks showed only minor differences, with no pattern of increased performance for either formatting scheme.

 

Ditto. No matter how much I buggered with it or what allocation I used the results came back identical.

 

In the end I realised write speeds were a bit tom and read speeds were astronaught.

Link to comment
Share on other sites

Question 4: Does "wiper.exe" help?

Answer: No. I could not get it to recognize any of my drives (the 2 x32s and my system x256, firmware 1.1) on any of the four computers I had available.

------------------

 

Use the standard MS ahci driver!

In that case wuper.exe will run.

But I don't think the wiper will run in raid.

Link to comment
Share on other sites

Ram Guy makes the point that you should delete your experimental partition after initial testing, and restore your system from a backup. That's important, because there has been plenty of suggestions made that you should retain the 32K allocation unit.

 

But using the 32K NTFS allocation WILL NOT IMPROVE PERFORMANCE. I can't see why theoretically it should be faster, and I have proven experimentally that it makes ZERO performance difference. What it WILL do is to use up a lot of your disk drive in unused allocation 'slack'. Using a large NTFS allocation is only a good idea on HDD drives if your normal file size will be very large, because a large allocation will decrease the tendency to file fragmentation. Fragmentation on SSD drives makes no difference to performance.

 

The best performance is achieved by making sure your format is aligned to a 4K boundary. This is theoretically the best alignment, and I have experimentally proven it to be true. You can achieve this by letting Win7 install anew onto the SSD, or by using diskpart (available in the Win7 installation disk under repair tools) and creating an aligned partition ("create partition align=4096"). If your installation is already Win7, just use Acronis to restore to the SSD. If your installation is WinXP or earlier, restore to an aligned partition.

 

The thing is: the Corsair guys, and a lot of 'overclockers' are really interested in getting the best benchmark figures - because that's what the marketing hype is based on. And to do that, you don't use the SSD for your system drive, or if you do, you take all the little files off it. Then you get a really whizzy benchmark. But, for the average user who doesn't care about benchmarks that much, using the SSD for the system disk is just about the BEST thing you can do to make your PC really responsive and 'fast feeling'. Leave all the temp files and paging and everything on the SSD. The benchmarks may be slower, but the computer in actual use will be fantastic.

 

I apologize for the diatribe. Touch of food poisoning makes me grumpy.

Link to comment
Share on other sites

  • Corsair Employees

I have split this off to your thread for several reasons.

I do thank you for trying to help others but if you are giving conflicting information that does not help any one.

And as I posted on another post we gave testing results to back up our suggestions. That was posted in Corsair Shows Users How to Set Up RAID-0 Arrays and our results Please see Raid Configuration and Performance! If you still have questions or an issue I would encourage you to call in and talk to me by phone 800-205-7657 and ask for Ram Guy I will be happy to talk to you in person.

Link to comment
Share on other sites

The conclusion I come to here is that we should all stop worrying about firmware, and just enjoy what we have. I do NOT mean to be contrary or give conflicting advice. Indeed, here is no advice, just opinions.

 

For the sake of completeness, I am posting this final set of test results. My X32s are firmware 2.0, so they may act differently from earlier models - but both of mine are identical in performance. I spent a lot of time running ATTO benchmarks, and came to the personal conclusion that the idea of 32K NTFS allocation unit was irrelevant from a performance point of view, and damaging from a disk space point of view. And that the alignment of the partition in the SSD does make a slight difference in the benchmark.

 

The performance differences show up in the benchmarks, however, they are so trivial that in a real usage situation I doubt if they would be noticed. The ATTO benchmark (which is the baseline used on this forum) measures sequential read/write speed - which is what the marketing folk like to stress. But unless you are moving vast quantities of data from one disk to another, what is important above all is random access time. That is my opinion, anyway. I get a benchmark result from a 300GB Velociraptor that in many respects is better than the best benchmark from the X32. And that is clearly nonsense. The computer is much much more responsive with the X32 than with the Raptor. I've uploaded screenshots of these tests.

 

So, I tried using the AS SSD benchmark tool they like on another forum. And it gives a much better picture of what's going on. I spent today running benchmarks on a 'used up' X32 - I mean, not security erased - as slow as I could make it. And I put the results in a spreadsheet, uploaded. I ran each test twice, to give an idea of the variation between runs. Tests 1 and 2 are on the disk as I left it after my previous efforts. The partition is properly aligned, and the allocation unit is 4K. Then I reformated the disk and mis-aligned the partition and re-ran the tests - tests 3 and 4. Then tests 5 and 6 have the partition re-aligned, and tests 7 and 8 have the partition re-mis-aligned. Then I reformatted the disk using 32K allocation unit and repeated the whole set.

 

Finally, I did a diskpart clean all (writes x'00' over everything), and did test 15, then did a security erase with hdparm and did test 16.

 

And lastly, I put the WD Velociraptor 300GB on the computer, and ran the benchmark twice on that.

 

What you see from all this is that the sequential write speed of the X32 is not that crash hot, compared to the velociraptor, but the sequential read is better, and the random performance is miles and miles better. The worst read access time of the X32 was 1/38 of the best raptor access time. The raptor fared much better in the random write test - it was only 7 times slower. This simply does not show up in the ATTO benchmark, but it makes a HUGE difference to the responsiveness of the computer. At the very worst, the X32 can read a floppy disk's worth of data - 1.12MB - before the fastest available spinning disk has even got the read/write head into position to start the job.

 

The X32 performance, as benchmarked by both AS and ATTO is significantly worse than the performance of my X256 - but still streets ahead of the fastest available spinning disk. I hold to the view that 32GB is a bit small for a system disk - especially if you have it configured for system security and a page file. And each ATTO benchmark writes over 4GB of data into the thing, so it doesn't take long to slow it all down. It's just that the slowdown is more noticeable in a benchmark than in real life, I think.

 

Anyway, for what it's worth, I've done what I set out to do. Now I'm left with a couple of X32s I don't quite know what to do with. Probably give 'em to the grandchildren.

 

Thank you for reading this far.

AS Benchmarks.pdf

raptor.PNG.cccbf0ecd6baf8f64ee995820964c9ab.PNG

X32.PNG.6b2499e3f5698f770f115361c2160300.PNG

Link to comment
Share on other sites

  • 8 months later...

Archived

This topic is now archived and is closed to further replies.

×
×
  • Create New...