I could be misunderstanding, but it seems to may you may be leaning towards the view that TRIM may not be working? I really can't be sure either, especially since the reports I've read about TRIM on Sandforce are not that encouraging, but if you had to argue over one or the other possibility, what would you say?

We don't have the data so let's not have pointless arguments.

It's a shame but not a surprise you can't get support from your vendor. That (and their other business practices) is the main reason I decided against investing in SandForce drives.

And it's a shame we don't have some kind of international organization to share information about products.
If we would all put a dollar in, we'd have enough money to buy gear for volunteer reviewers. I don't know about you but I give a whole lot more to some of my information sources. Since you're so keen on international organizing, you wouldn't know how to get something like that off the ground, would you? Little investment (time or money) required, quick payoff, no political/religious barriers to participation, material conditions begging for mutual aid and of course no prospect of repression... it ought to be easy!

Yes, I have had much limited input on this issue at Corsair support


The information available in the web is much fragmented, certainly it's not easy to stumble upon the most helpful info when one might need it...
Review samples are generally available, however, doubts about website affiliations or shortcomings in tests could arise...
....of course knowledgeable communities and much involved users are not in shortage and its usually them that produce much of the valuable info...

I believe the functions and utility of such an "organization" are pretty much covered by the various technological forums and the interconnections established between them via the internet by the users themselves, taking part in multiple forums each...
...moreover, if reviewing would be its main function and if it was actually representative of such a wide body of users worldwide, I don't think there would be much need to fund it for review samples...I suppose there could be a danger of hand-picked samples offered by manufacturers, but then again, people would buy what they'd need and on available info at a certain time anyway, so the important thing would be to establish an as sound as possible testing methodology for reviewing for various usage patterns and to gather the available tests and compare them...
Then again, consider the volume of information and the many languages in which all that info would have to be translated to make it available around the world if it was to gather reports on all tech components, from all manufacturers and from great deal wider base of users...

Honestly, with the given scope of activity for the organization you're proposing, I don't think it's worth it...

You'd have to solve economic access issues to all those goods to make its scope and base meaningful

It was more a joke than a proposal but here you go:

That's a bit like saying the Wikipedia is not needed because of all the forums and newsgroups... and indeed all the unmanaged wikis.

What the stuff we have lacks, beside coverage and ease of use, is trustworthy information. There's lots of hype and misinformation around. Lots of useless noise. And precious little hard data on anything uncommon or non-trivial. An organization could fill the accountability and commitment gap between outfits like SPCR and anonymous posting/blogging.
Keep in mind that there is an adversary: the corporations which sell those products. If a source of information rises above obscurity, they will try to influence it. Advertisers have turned some well-known publications into fact-resistant shilling circuses. If it's a source of information based on public participation, they'll try to poison it with shills. If you let them, they'll end up writing shill-bots. You need some filtering and accountability to defeat that.
And don't discount how people are motivated. Some people will be happy investing time into thankless pursuits. But others crave social recognition. Anonymity can only provide so much of that (unless you have millions of readers or something of course at which point you can turn their eyeballs into revenue if nothing else). Look at how people care about something as silly as a distributed computing score.

I don't know about you but I prefer to take consuming advice from consumer organizations. But on some topics they don't have much so I turn to commercial publications, forums and such. Usually they concentrate on the important stuff (transportation, health care and so on) which tends to be fairly region-specific unlike tech stuff.

I started writing in response, but the text was getting bigger and the bottom line of my answer was there are more things around -but related with- technological production-circulation-consumption that I find more meaningful to deal with than establishing an international organization and review database for informing consumers on products' features and performance, so I wouldn't be interested in investing time in setting up such a project or discussing much about it...(when I do, I tend to go long, which is not always fruitful or met accordingly)..thus, I'm not going to feed the discussion myself.

I don't find the idea silly or unworthy of deliberation or implementation (that's why I bothered to take up the point, however briefly)...especially if it can find ways to "keep manufacturers 'honest'", to neutralize much advertising and maybe even force them to focus more on research for better products (much, much of this is, yet, dependent on regional factors) than trying by indirect ways to fuel consumption...that might be the ideal scenario for such an organization's scope...

Silentpcreview is an adequately popular website, it is very serious and dependable in its reviews, with devoted readers (much more than its frequenting forum members), so maybe you could give it a shot and put your thoughts and proposals into a text that could be displayed in the front-page and see what input and reactions it produces...
Maybe something started as a joke can go further than that..Good luck

I feel the same way. In a way that's a shame because not bothering to step up is what got us where we are.


Definitely.
But there's another, easier way to reduce consumption: freecycle.org and such. It's local however.


And the more immediate problem is that in spite of:

You can't see to get a SandForce user to chime in.
Maybe with a new thread title? Or maybe someone can suggest an appropriate forum/mailing-list/newsgroup?

I have no complaints, though, you've been quite helpful...

Maybe people will not think to look much in a silencing orientated website for SSD reviewing and commenting, given that SSDs pretty much take care of that silence factor on their own...

I for one would still like to know if you have TRIM working or not and what sequential write speeds others get. But it's not my drive so I'll get over it...

Well, about the speeds other are getting, there are some reviews worth reading, but I haven't kept the links to them...
If I get news from Corsair, if I stumble upon a helpful thread, if an answer comes up, I'll drop in and quote it...

Right now I don't see what else I can do with my drive to test further

Behardware website reports that they have tested the latest (december) version of the AMD drivers and they found TRIM to be supported both with Indilinx and Sandforce SSDs...
Same goes for the latest version of the Marvell drivers for the its sata chipsets...

This seems to contradict other reports about TRIM support in the latest AMD drivers...

Pierre, I have not been back to the site much lately, but I have read most of this thread. I have a Sandforce SSD, and if you can recommend a test that you would like me to run that would not unnecessarily wear my drive, I'd be happy to do it. I cannot agree with the posts to use the drive as little as possible or that 1/3 performance drops should be acceptable. I have not noticed any performance degradation with my drive in real-world use, but I have not run any performance benchmarks, other than Win 7's built in "Windows Experience Index," which gave me a 7.1 (on a scale of 7.9) primary hard disk, disk data transfer rate on Dec 12. I ran it again just now and got the same score.

@who1zep

Thank you for your willingness...I cannot really recommend a benchmark/test that would not have an effect on your drive (temporary at best)...
I don't like the idea of a sustained performance drop of 1/3 at all either, but it might well be the case...

To test for degradation one would need to run a test on a clean / secure erased drive, then perform all sorts of read-write tasks then retest...

I have read about cases of amazing performance restoration with TRIM, back right up with the initial top scores, and that's certainly great, at least from one point of view.

However, I am leaning towards the thought that Sandforce SSDs may actually behave that way, as HFat suggested, that is in non-aggressive data TRIM, for the purpose of longevity.
Maybe it could also be a matter of non-efficient win7 driver - sandforce combination...I will try an AMD driver as soon as TRIM is universally confirmed to be working (I get better performance with them anyway)

@whozep1: the thing we'd like to test can not be tested without wearing out the drive but perhaps you'll consider doing it anyway since it doesn't have to wear out the drive very much at all. How about copying a large compressed file such as a DVD rip to your drive and measuring how long it takes with a stopwatch? That would be useful. I you do not have large incompressible files on hand, you could create one with a free program such as Truecrypt. Of course the result would only be meaningful if you have two drives in your system or another way to do fast transfers (USB2 would not be fast enough).

Let me emphasize or clarify a few things:
I have no idea what is going on. I was merely stating what was possible, not what I think was particularly likely. What I suggested was that the controller might not take advantage of the information it gets through TRIM to defragment internally and create empty "flash blocks" as other drives seem to be doing. You need these empty "blocks" to get the fastest write performance possible.
But if you erase *everything* with TRIM, performance should be as good as new unless the drive deliberately slows it down.
I understand different SandForce drives can have different firmwares with different performance characteristics on the same hardware (market segmentation).
What Pierre observed is a performance drop of more than 1/3 actually... but only on a few specific benchmarks. The overall performance of the drive (whatever that means) did not degrade as much. It would be entierly possible for some users not to experience this performance degradation. Used in a laptop whose fastest connections to the outside world are its USB2 ports for isntance, the drive would probably never have an opportunity to reach the theoretical bottleneck caused by such a performance degradation.

Seriously 30% performance drop is too much for a storage drive? Have you never used rotating disks singular or in RAID?

Forgive me if this sounds like "get off my lawn" or "when I was a kid I had to walk through snow uphill both ways" but I've literally seen hard drives slow enough that defragmenting them took days. I'm not joking, I'm not exaggerating. Literally I had to disable screen saver and standby and had to literally check on the system and run the defrag multiple times before the defragment was sufficient.

Take a look a HDTune or HDTach graph and look at how much performance drops from the first 10% of the drive to the last 10% of the drive. I've seen drives that look like a ski slope they drop off so bad. Take this 640GB drive for example:



Just doing a quick bench you can see that it drops from 117 to 59 well more than a 30% loss and that is before you take fragmentation into account. It's just physics. And that isn't a bad example. I've done benchmarks on old rotating disk drives that the end of that slope was in the single digits. If you didn't short stroke it the last 30% or so was practically unusable from a performance standpoint.

And you do know that some SSDs become just completely unusable if you torture them properly. Not just a little slower but like don't reboot or you may not have the patience to wait the nearly indefinite time it'll take to ever access the data on the drive.

Anything a SSD manufacturer does to make a drive faster will either allow performance to drop after data is recorded or will reduce the accessible capacity of the drive (changing the price per usable GB). If you have plenty of money to burn you can fix these sorts of problems by massively over provisioning your SSD or if you don't have that much money just do it a slight amount (very similar to short stroking a hard drive in effect but not for the same reasons). Just google "SET MAX ADDRESS" and be prepared to back up and restore any data you have on the drive.

1. Backup data
2. Set max adress to make the drive appear smaller than the sticker on the drive says. Say 50% on the drastic side or a more reasonable 80% of advertised capacity or somewhere in between.
3. Secure erase the drive
4. Quick format
5. Install OS and/or restore data
6. Watch as your newly reduced capacity SSD is more resistant to slowing down when you benchmark writes. Not immune, just more resistant. (Disclaimer the improvement will vary by controller, This works well on Intel SSDs I don't know exactly how well it helps a Sandforce drive).

If you don't accept that some small drop in performance is unavoidable the only way to do that would be to set an artificial cap in performance and throttle the drive so you wouldn't notice the difference between perfect performance and the unavoidable drop. Of course then you'd complain that drive X that is throttled doesn't perform as well as Drive Y that isn't throttled.

The only real difference is with Hard Drives the synthetic benchmarks ignored fragmentation and with SSDs benchmarks highlight it (due to the differences behind the scenes). With either type of drive the real world performance drops with certain usage patterns.

I'm not saying you can't do anything to improve that performance. I'm not saying you can't complain. I'm just saying please don't wail if someone says hey that might be normal or I'm not sure if that's enough of a problem to complain about.

Pay no attention to the man behind the curtain. I've got to go walk uphill both ways some more. Now where did I put that monkey

My Mac mini has USB 2.0, FireWire 800 and an SD card slot. I don't think I have an external drive with Firewire, though. My external drive is USB 2.0, and wouldn't you know it, when it was in transit I learned of a USB 3.0 option for the drive! It was too hard to get an upgrade of Win 7 running on the Mac mini through OS X's boot camp feature to risk wiping the drive for your test, even though I may be able to get it back the way I have it now.

The reason I picked up an OWC (aka macsales) brand SSD is because I learned that Apple's OS X does not support TRIM, but that Apple users really like OWC because, even thought OWC does not support TRIM, the overprovisioning prevents performance degradation like other drives, which I first heard of through diglloyd's review when looking for info as a newbie to SSD tech. There have been many positive OWC comments on the macrumors forum.

http://eshop.macsales.com/shop/internal_storage/Mercury_Extreme_SSD_Sandforce/Solid_State_Pro

If you're willing to do something slightly more complicated, try typing these commands (assuming Macs have standard utilities now):
openssl rand 268435456 -out testfile1
dd oflag=dsync if=testfile1 of=testfile2 bs=268435456
Do *not* type this if you have named other files "testfile1" or "testfile2". The exact number is not important (I chose 256M). Don't forget to erase the test files afterwards.
This would not wear out the drive much because it only writes half a gig to it, and sequentially at that. It would have less impact than copying a DVD rip.
It's not as good a test as copying from a different drive because your OS may not cache testfile1 and read it from the drive you're trying to test. If you were willing to go an extra mile to give us a reliable test result, you could write testfile1 to a RAM disk instead: http://osxdaily.com/2007/03/23/create-a ... -mac-os-x/
Running the test a couple of times to see if there's a difference between runs would be helpful as well.

Am I supposed to use a stopwatch or will the file generate the info you are looking for? If a stopwatch, how will I know when it is done? I use the Win7 side exclusively, so anything that can run on Win7 should run on my Mac mini.

dd gives you the information. It's more accurate than a stopwatch which allows the use of smaller files. The parameter I used tells it to wait until the data is commited to the drive (and not just sent to the cache). If I was confident the drive didn't lie about completing writes we could use a much smaller file actually.
But dd and openssl are not available on Windows. Nor is time. So I don't know how to do this in Windows. But you should be able to rig something together if you care. You could first create a test file with Truecrypt's GUI which is probably easier to install than openssl. You could then make a batch file with "time < nul" commands (they replace a stopwatch), copy and this utility: http://technet.microsoft.com/en-us/sysi ... s/bb897438
Tell me if you want to give it a try but you don't want to write the batch file yourself.

Are you hinting at this (btw, it's on the eight SERP for "SET MAX ADDRESS"!)?

The first reason I've taken up this hypothesis to be most probably true is that it fits well with what has been said about Sandforce SSDs by other users in the forum threads-links I have provided above.
The second is my experience with the drive...it reached very slow writes at one point (2/3 degradation) but it has recovered and between tests, given a day to "relax" with everyday tasks, the write performance consistently measures at around 59-65mb/sec for sequential writes...

Are those the speeds you're getting now, as opposed to the ~35MB/s you used to get? So it looks like the controller *does* defragment after all, just slowly. That or it cripples the drive when it notices you're writing too much.

I posted this back in the first page, that bench results at least seemed to get better (the real performance test - copying a large archive were not as good but confirmed the improved performance)...

I ran 3 benches since, they were a little lower than the 1/3 numbers mostly. Between benches and average usage it was pretty stable at those numbers.

That is an OK document to read and documents some of the performance gains/losses depending on how you compare but it basically tells you to go elsewhere to actually find out how to send the command.

http://www.hdat2.com/hdat2_faq.html is on the first page of results (I actually had to google SERP because I didn't recognize the acronym). I would hope that anyone willing to do something this advanced to their drive would be willing to truncate that address to end up at http://www.hdat2.com/ and poke around enough to learn what they needed to know.

http://en.wikipedia.org/wiki/Host_protected_area is the first result (the I'm feeling lucky result) and has links to several tools and discusses the methods. Which would again point you to hdat2 among other tools.

http://www.samsung.com/global/business/ ... _Tool.html is on the first page of results and would probably work in a pinch for just doing the Set Max Address but I haven't tried it (the most likely down side is it would just complain about not seeing a samsung drive in the system and may not let you do anything until it sees one). Even if this tool didn't work for you seeing their instructions and screen shots might give you some clues.

I'd expect linux users to bump into using HDparm.

I'd expect those familiar with the Ultimate Boot CD project to think of using one of the provided boot images.

In short I wasn't pointing to a single reference. I was saying this is advanced stuff. You need to be sure you know what you are doing before you try it. And yet I am suggesting it as a good way to make your SSD more resilient to slow downs.

You can get the same affect on Intel drives by never formatting them to full capacity but this leaves the possibility of the placebo affect. If your drive was a return or you wrote to it doing a benchmark that tests the full drive without you realizing it you could assume that you were getting the benefit when you really weren't. Using Set Max Address keeps HDTune and friends from testing the space you intended to reserve. It may also be true that other controllers benefit from Set Max Address but don't take advantage of unformatted space like Intel G2 drives do. I suppose if I was going to buy a large quantity of SSDs I'd probably buy one of each controller I was considering and do some in depth testing of partial formats vs full formats vs set max address.

I've never done Set Max Adress on a SSD yet but then all my SSD use has been on my systems and I've specifically avoided write benchmarks. I'll eventually be putting a SSD in my girlfriends laptop and I might do a Set Max Address on that SSD before I put it in her system and I might do Set Max Adress on mine before I migrate to Win 7 (I'm not a big fan of the partition UI during the Win 7 install). If I don't I'll just use a Linux partition manager to set up the partition before I install Win 7 (Assuming the Win 7 installer won't be fussy and overwrite my prepared blank partition that is sized just the way I want it.)

I've just misunderstood you, I was thinking it should be a strictly SSD-related question.
Thanks for sharing.

Yes, that is the thing about SSDs. For all intents and purposes they have to pretend to be hard drives when queried by legacy hardware/software. Not everything is compatible but the manufacturers spend huge amounts of money to make it as seamless as they can.

Everything I said about set max address I meant only for SSD users but the same processes would work for hard drive users (I just can't think of a good technical reason to do it on a hard drive).

Aren't so called short-stroked disks supposed to be faster?
Or have I misunderstood again?

Short stroking a traditional hard drive doesn't require using set max address because hard drives don't use complex behind the scenes stuff like wear leveling. There is no real world drawback to short stroking with easier implemented procedures.

If I want my traditional hard drive short stroked I can just take my windows XP CD and

1. Remove any existing partition
2. Create a partition as small as I want say 25-50GB on a 640GB HD
3. format (Full or Quick)
4. Install OS or restore backed up data

Any software that benchmarks the entire drive like HDTune/HDTach and such will ignore the gains of short stroking in this manner but in real world use the OS will ignore all the unformatted area on a traditional hard drive. And because of that there is no real world penalty for benchmarking a traditional hard drive that is short stroked. When the test is done my short stroked drive behaves the same as if the benchmark never ran.

I absolutely see an advantage to short stroking traditional hard drives and have done it more times than I've counted but I don't see a technical advantage to using Set Max Address on traditional hard drives.

The only advantage I see is from a marketing standpoint when you want to say produce 2 physical capacities but want to sell 4 or 6 capacities to create artificial price segmentation. For easy math lets say a HD is worth $0.1/GB and we have 500GB platters and don't want to change the number of heads.

without set max address I might sell

500GB at $50
1TB at $100

But say my marketing department wants to sell a $75 drive and a $25 drive as well

with set max address I can sell

250GB at $25
500GB at $50
750GB at $75
1TB at $100

the 750GB drive in this imagery scenario is really a relabeled 1TB drive.

In the real world it is more complicated than that (costs aren't linear per GB, factories continue to spit out older platters at the same time as newer platters are made elsewhere, other technical tricks can be used to reduce capacity like selling a 500GB platter with less heads so only one side of the platter is used, you can sell a 750GB drive made two different ways at two different prices due to the variation in performance, and on and on)

But since most consumers would never grab a program that will send the set max address command such a strategy could be used and probably has been used in the past.

Ironically in the SSD world that same scenario changes the price the other way

Say you have a SSD with 128GB user area flash. (in reality it may have more than 128GB flash but the amount of extra flash doesn't affect this imaginary comparison).

If you Sell it as a 128GB drive it might have a price of say $95
If you set max address to 120GB with the same controller and flash it might be $105
If you set max address to 100GB with the same controller and flash it might be $140

Totally fictional prices but if you want to see a real world example take a look at viewtopic.php?f=7&t=58811 and search for the phrase



The lower the usable capacity on a SSD (with the same flash and controller) the less write amplification (which means the drive can last longer and has faster writes in a degraded state) the higher the IOPS, so they raise the price.

But if you are willing to do the work yourself you can buy the cheaper drive and set max address to the lower capacity point of the more expensive drive and get the benefits. (You won't get the warranty increase but you'll get technical benefits)

According to Intel, it's about the same for overprovisioning an SSD: "Define a partition that is less than the maximum available capacity of the SSD. This option can be found in the OS drive configuration tools." So which is the real advantage of setting a lower max address?

However, just as a side note, I've just bought a 160GB G2, set the max address with HDAT2 to 120GB (several tries to obtain the right number), and then restored a Windows fresh (recovery) install on it: to be honest, I'm not impressed at all by the new overall performance, but being a laptop drive I cannot easily swap the previous mechanical drive back to see how slow it was before the SSD.

http://www.techpowerup.com/articles/other/157

Check it out and see if your "restored OS" is properly aligned. If not you aren't getting the full performance of the drive.

As to the SSD and not using set max address as I said before you, a program, or another user could write to an area on the drive that you intended to reserve. Set Max Address hides the extra capacity from the rest of the PC so that can't happen. It's not absolutely required to get the benefit but it negates a possibility of thinking you have the benefit when you, someone, or something else has negated that benefit.

Intel says you don't have to do anything else other than partition less than maximum capacity but you do have to NOT do certain other things for that to be true.

Intels suggested method is to take a factory fresh never before used drive and partition to desired capacity.

It isn't factory fresh if it's ever been formatted to full capacity even for a second and then reformatted to a lower capacity

It isn't factory fresh if it's ever had whole drive write benchmarks run. A single pass of HDtune/HDtach/IOmeter and such in write mode will invalidate the drives optional spare area (the controller won't let it take all the spare area just the optional spare up to the advertised capacity).