OCZ Vertex Drives

[Rhys]

Here is my X25-M after one run of IOmeter. Less than 2 hours after doing a clean install on it too.


Happens everytime I do a large Iometer run. It took me a while to figure out what was causing it.

Hi Lowfat
Thanks, I will be sure not to run Iometer then!
What method do you use to get your drive back up to speed?

[Ao1]

@ Rhys, bottom line is that synthetic benchmarks don't really mean jack with SDD's. It's user experience that matters and if your SDD's aren't being affected I wouldn't be too bothered with benchmark results. If your user experience is being affected the good news is that a quick format and re-image should get you back to full performance and it's much quicker than defragging a HDD, so no big deal

As far as I know Intel gave out HDDErase 3.3 to reviewers to enable them to format X-25-E after each test. They asked reviewers to format after each benchmark because benchmarks affect the way the Intel drives perform due to the way they adapt to usage. I don't know if HDDErase is the most applicable way to format on an OCZ drive, but I'm sure the guys at the OCZ forum could advice you.

@ lowfat, out of interest have you tried not running synthetic benchmark to see if a perceivable drop in speed occurs based on real world use? If so that would be concerning, if not does a in synthetic benchmark really matter?

I've seen relatively small drops in synthetic benchmarks on the X25-E's but absolutely no perceivable drop in real world use.

[One_Hertz]

When the SSD is brand new from the factory or has been secure erased, all the cells contain no data in them. All the drive has to do is write data to them to get the job done. Once all the cells have been filled even once, what the drive will have to do in order to make a new write is to delete the previous contents of the cell first, and only then make that write operation. Naturally this takes about double the time than writing to a blank cell.

When you fill up the drive with IOMeter test file, all the cells become "filled" and even after you erase the test file, they are still filled because the drive does NOT know which cells contain valid user data and which do not. So it is not about how much space you are currently using, its more about how much space you've used so far overall.

The above applies to most of the performance SSDs...

I have no idea why Vertex drives are seeing a drop in the reads... That logically should not happen and doesn't happen to all the X-25 drives. The degradation should only apply to the writes.

A possible solution to performance degradation is to make a program that would be aware of the internal wear leveling algorithms of the SSD in question and would be able to tell the drive which cells do not contain valid user data so it could go ahead and do the erasing whenever you want it to and not right before doing writing operations. I believe Intel is working on something like this, but it seems like they aren't having too much luck. Programs like this would be SSD specific for obvious reasons... There won't be one program that will support all SSDs.

[lowfat]

Hi Lowfat
Thanks, I will be sure not to run Iometer then!
What method do you use to get your drive back up to speed?

I used HDDerase 3.30 as mentioned above.

[Tamerlane]

I have no idea why Vertex drives are seeing a drop in the reads... That logically should not happen and doesn't happen to all the X-25 drives. The degradation should only apply to the writes.

I am trying to follow the comparison here but I may be misreading, or not have seen, some of the data.

Doesn't lowfat's CrystalDiskMark benchmark on the X-25M show the exact same or worse degradation of reads and writes as the the same bench performed on the Vertex by "Bob Dobalina" in the OCZ forum?

Or is there otherwise a pattern of evidence that Vertex degrades more?

[XS Janus]

A possible solution to performance degradation is to make a program that would be aware of the internal wear leveling algorithms of the SSD in question and would be able to tell the drive which cells do not contain valid user data so it could go ahead and do the erasing whenever you want it to and not right before doing writing operations. I believe Intel is working on something like this, but it seems like they aren't having too much luck. Programs like this would be SSD specific for obvious reasons... There won't be one program that will support all SSDs.

Maybe M$ could do this on OS level? I don't think it is anything an end user should be concerned with. In Intel is probably trying to this on firmware level...
But it does sound logical... cells marked for deletion should at least be deleted when the drive is idle... Something like reverse of write back caching - cache the info on what cells need to be emptied... and empty them on idle time

[One_Hertz]

I am trying to follow the comparison here but I may be misreading, or not have seen, some of the data.

Doesn't lowfat's CrystalDiskMark benchmark on the X-25M show the exact same or worse degradation of reads and writes as the the same bench performed on the Vertex by "Bob Dobalina" in the OCZ forum?

Or is there otherwise a pattern of evidence that Vertex degrades more?

I think Bob's reads degraded a lot too? Lowfat's X25-M is showing 244MB/s reads; 250 is normal so that is fine, no degradation of reads while the vertex had quite a lot. Writes are both bad percentage wise. Vertex 195 to 75 for two drives, X25-M 70 to 27 for one drive. Actually, both drives degraded to exactly 38.5% of their original write performance. What an odd coincidence.

The degradation isn't really a big issue, as other have stated, you just need to set up a system where you take an image of ur boot array, then wipe it, then put the image back on. The defrag of HDDs takes a similar amount of time so I don't see what the problem is for enthusiasts.

[Griff805]

Someone might want to tell this guy about the performance degradation before he spends tons of time and money...

Who is OCZ letting run the place over there?

http://www.ocztechnologyforum.com/fo...ad.php?t=52566

[One_Hertz]

I am sure they'll make a sticky soon to explain everything to people. There are lots of posts about it now all over the place.

[WRC]

Until microsoft fully supports ssd's, performance may always drop. Supposedly windows 7 will offer better support for ssd users but I think we need a totally different OS

[Rhys]

Until microsoft fully supports ssd's, performance may always drop. Supposedly windows 7 will offer better support for ssd users but I think we need a totally different OS

From what I hear Linux may be a good choice, it already has Trim commands ect! seems to evolve faster than windows.
But of course the drives need to be able to support those commands.

Edit: Microsoft might improve W7 support with a service pack or in a next RC.

[Rhys]

I am trying to follow the comparison here but I may be misreading, or not have seen, some of the data.

Doesn't lowfat's CrystalDiskMark benchmark on the X-25M show the exact same or worse degradation of reads and writes as the the same bench performed on the Vertex by "Bob Dobalina" in the OCZ forum?

I don't know which degrades worse, but sure did look like reads were beginning to suffer badly in places in the atto. Also PCper results show a pretty bad reduction in read speed also.
It seems convenient that because the x25 'Learns' usage patterns, the drive was wiped after each benchmark. This would have also prevented the reviewer from discovering the degradation problem.
Intel say they can't replicate PCper results! Not sure I believe that.

[Griff805]

Are all the review sites still hung over from CEBIT or something? Where are the Vertex reviews?

[jcool]

Ordered a 60GB and 120GB Vertex today... should arrive tomorrow, what do you guys wanna see for testing? I was thinking HDTach R/W, ATTO, Crystaldisk and maybe a few IOmeter runs on the ICH10R. Didn't know you had to fully erase the disk afterwards to get it back up tp speed yet, thanks for that - any other tools out there? Cause I can't find that one anywhere for download.

[F@32]

Until microsoft fully supports ssd's, performance may always drop. Supposedly windows 7 will offer better support for ssd users but I think we need a totally different OS

This whole degrading performance/defrag issue is another dropped ball by SSD manufacturers and MS. Plugged one hole-alignment in Vista and W7, here goes another one... This should be handled at FW or OS level, not HyperFast/PerfectDisk...

[Brahmzy]

This whole degrading performance/defrag issue is another dropped ball by SSD manufacturers and MS. Plugged one hole-alignment in Vista and W7, here goes another one... This should be handled at FW or OS level, not HyperFast/PerfectDisk...

Agreed. Obviously we're not the first to suddenly find out about this. I'm sure engineers/devs at M$ and the SSD manu's know about these limitations. Will this continuie toplague any future SSDs? Or is it perhaps and technology issue that will be addressed at the hardware level? I think it needs to be addressed at the hardware level. I don't want my OS having to worry about anything more to do with starge than is absolutely necessary. It should be able to read and write at will. I think the controllers on these things need to be more intelligent.

Again, I vote that the OS should remain storage - independant for the most part until there is one hardware standard. Unfortunately, that will never be the case, so it needs top happen below the OS.

[XS Janus]

Ordered a 60GB and 120GB Vertex today... should arrive tomorrow, what do you guys wanna see for testing? I was thinking HDTach R/W, ATTO, Crystaldisk and maybe a few IOmeter runs on the ICH10R. Didn't know you had to fully erase the disk afterwards to get it back up tp speed yet, thanks for that - any other tools out there? Cause I can't find that one anywhere for download.

No offense but all those benches are done all over the place 100x already
Try and be creative and give us some real world results, large and small file copies, load times and compare that to other storage setups you can get your hands on.
Any real world experience you can share will be appreciated

[jcool]

So people have shown that all over the place, including difference between 32mb and 64mb cache models of Vertex? I must have missed that

Sure I'll try and do a few real world benches as well, unfortunately I won't have time to do a proper review/comparison because these drives are way overdue already, and I got customers waiting for their builds

[Ao1]

No offense but all those benches are done all over the place 100x already
Try and be creative and give us some real world results, large and small file copies, load times and compare that to other storage setups you can get your hands on.
Any real world experience you can share will be appreciated

Exactly. Why do people buy a SDD drive? To run benchmarks made for HDD? Surely people want something that gives snappy OS and program performance and if so the question is not how benchmarks are affected but how does real usage get affected, which is something no-one seems to have answered.

That said, I'm intrigued as to what is happening to here.

An SDD that shows 50% of written data is not showing how may cells have written data. In addition technology is being used to limit wear by distributing data across cells to limit the times they are written......if you try to shrink a SDD partition you can't do it.

Wear technology and the lack of ability to flush the cells of unwanted data. I'm still trying to get my head round this so please bear with me.......

If you have a 32GB SDD drive and during the Win 7/ Vista install process temporary files create 24/26 GB of written data the cells are technically 80% full as the files that were temporarily written are still kept on the SDD cells.

Does that not mean that the SDD is already 80% degraded by just installing the OS. Add a few basic programs and take into account temporary files being written in the process and the drive's cells are likely to be 100% full, so why would a 32GB drive benchmark well on a fresh install and then see a significant drop in performance after a few days?

From what I understand it does not matter how much data is in the cell, it takes the same time to overwrite it. So once the cells are full the drop in performance should be definable based on how many cells are being overwritten.... but is it?

The other thing that is puzzling me...in the scenario of a 32GB drive OS install leaving all cells with written data, how does the drive know where to distribute data to prevent wear? Assuming it does know how to overwrite data on cells that are not required why can't a process be developed to clear data that is not wanted? If wear distribution can only work on cell with no data does the drive wear out quicker as a result?


I guess what I am getting at is this.

1. Does degradation affect real use performance? (I can answer this at least for the X25-E. No).

2. If the time taken to overwrite a cell is quantifiable the drop in performance should match depending on how much data is being overwritten. (If it isn't that would seem to raise more questions).

3. Does degradation occur quicker on smaller drivers in comparison to larger drives? (If it doesn't that would seem to raise more questions).

I'm just an end user of SDD technology trying to work this out. Thanks for hearing me out.

[Ao1]

Deleted

[Halk]

Audience are you trying to spread a pack of lies?

Go to any hardware manufacturers support forum and you'll see plenty of problems with the hardware.

You don't have to disable prefetch or indexing, those are just tweaks that OCZ suggest may improve performance.

You have to format the drive every couple of days? That's just rubbish. The guy that had severe performance issues had used a piece of software not designed for SSDs, and had to use another piece of software to fix it.

What hardware compatibility issues? Again you're talking rubbish. Any compatibility issues from the OCZ forums seem to be with very old hardware or with weird hardware.

It wasn't delayed for months. Another lie.

[Ao1]

Deleted

[Halk]

Of course it was delayed. It wasn't delayed for months though. You need to realise that people see XS as a source of information, and posting a pack of lies is irresponsible. There's perhaps a grain of truth behind each of them. Yes some people have had problems, yes it was delayed, yes eventually performance can become degraded. But why the enormous exaggerations?

[Ao1]

Halk, I don't think I exaggerated anything that I did not read from the words of OCZ's own moderator, but fair enough I've deleted the posts as I don't really want to detract from the questions in my earlier post, which is not OCZ specific btw.

[Ao1]

Didn't know you had to fully erase the disk afterwards to get it back up tp speed yet, thanks for that - any other tools out there? Cause I can't find that one anywhere for download.

This version works for the Intel drives, later ones don't.