TRIM and GC don't seem to work on my Sandforce SSD - how can I get performance back?

So, I removed all my RAW data image files off the Inferno 100GB SSD I had because of this post by Anand:

Source:
A Quick Look at OCZ's RevoDrive x2: IBIS Performance without HSDL - AnandTech :: Your Source for Hardware Analysis and News


At this point I ran the CDM benchmark as a reference starting point.

I then left the system/drive idling for 9 hours at the log on screen.

Again, I ran the CDM benchmark to see what sitting on the log on screen would do to performance.

I also left the system/drive idling for 10 hours at the BIOS screen.

Finally, I ran CDM one more time to see if idling on the BIOS screen allowed the drive to finally 'recover'.

The CDM scores are included below.

Executive Summary: TRIM nor GC work on my SandForce Drive.

Worse, leaving such a drive idling kills the 4K Random Write performance by about 35%.

Anyone have any suggestions to get the performance that I paid for?

Needless to say this 'Inferno' was one expensive test (not to mention a waste of time) - no more SandForce based drives for me in the foreseeable future.

Unless someone has a magic setting I've missed?

BTW, these are run with the JJB's and stamatisx tweaks with IRST 10.0.0.1046 WHQL drivers at 38% free space on the SSD on Win7x64 and 8GB of RAM.


Note that this is not a recent phenomenon; I have never seen this drive 'recover' since it was new.

 

Well with the amount you've written to your SSD (> 30 TeraBytes?) I'm not surprised it's not performing as a fresh drive. But it might not be too late to restore it.

Try doing a secure erase with HDDErase and after that follow Anand's advice to avoid the bug: don't fill over 80%, especially if you use a lot of incompressible data.

I'd like to see how other SSDs handle the amount of data you write to SSDs. Chances are it won't be any better. Most MLC drives use the same quality NANDs. Other controllers don't have the protection mechanisms of the Sandforce controller.

Like me and other have said several times, with your usage scenario you'd be better off with an enterprise quality drive.

 

Need to write 0xFF (nulls) to empty blocks for GC to work. Best to do a Tony Trim.

 

Seems like the guide is meant for Indilinx based drives. Is that recommend for Sandforce drives too?

 

Your only remedy(other than secure erase) is to logout and let the system sitting there for a few nights.

This is one of the key issue I have with SandForce that I don't like(and would not use it). x25m is a much balanced drive.

 

While I agree that such amount of write warrant SLC(it is actually cheaper if you calculate write/dollar) but restoring performance is another issue.

For controller like Intel G2, TRIM would restore the factory performance(just backup, trim and restore). Sandforce for its so called protection mechanism cannot. Of course, no matter what controller it is, a generous OP is the only way to ensure long term performance and endurance(well prolong).

 

Thank you (everyone) for the suggestions.

Here is a little more information:

I have not written 30TB (yet!) closer to 3.5TB.

I have allowed the drive to 'rest' for probably half of its lifetime! Makes no difference. (rest days = over a month).

AFAIK, secure erase is not supported (easily) on this drive? I'm a little hesitant to make it worse (possibly).

I don't care if it gives this level of (non) performance - I'll simply use it as the most expensive ($400) 100GB USB 'key' when I finally throw it in an external USB enclosure. I just don't want to kill it needlessly - I've wanted to, but I would rather take a jack hammer, torch and gasoline to it rather than simply brick it with a non-compatible secure erase process.

Here is an SSDLife screenshot - it is 3.5TB right (and not 30TB?).


As to the Intel's being much more balanced - I know! But a G2 160GB performed at the same level of my highly optimized mechanical HD's (the Inferno was initially much faster).


Finally, as to enterprise level drives (SLC's), I'll only state again that current MLC based SSD's are faster than the insanely expensive (and much too small for me) SLC 'old tech' that is available.

Even if they (SLC's) operate at a consistent performance level (unlike the MLC drives), they are barely above the mechanicals for the things that are important to me (productivity, not simply more 'snap').

The worst part is that Patriot, while guaranteeing the Inferno for 10 years is not making any firmware updates, nor any SSD tools to handle these issues available.

Hope the next wave of ssd tech finally catches up to the dependability/usability of modern mechanical HD's.

Thanks again for all the responses. Hope more are coming.

 

Is that the estimate for where the SSD will read/write so poorly it will need to be replaced or ETA for it's death?

 

Not sure... but my ETA for it's 'death' is getting shorter and shorter as the new gen ssd's make their way around the world.

 

That means you have to adopt the SE procedure and use Sandforce.

Basically, sitting idle would make it recover about say 70-80% or something like that but never the 'wow' level as a new one.

Actually, SE may extend its life based on your usage pattern. Keeps on using it in a non-optimal state may end up to be more actual write to the NAND.

The only problem with SE is that it is not so easy task like TRIM as it requires hotplugging and booting using another HDD, not something suitable for most laptop.

Your usage pattern seems to treat it as a 'scratch disk'

BTW, you may consider an OP of say 30% which would be better.

 

What? All my research (LOL, not a lot) indicates that SandForce are the controller leader (at least at present).

Incidentally, if you were a test driver I'd hate to ride shotgun with you in a pre-production performance vehicle. I get the feeling it would be a real 'white knuckle' experience.

 

Until you know its characteristics. It is fast initially but for certain usage pattern would quickly come down.

Not to mentioned the stability(or should I say incompability with quite some devices).

 

Yes, I'm reading about that. The manufactures claim one thing, but user claims another. Still, you have to agree that tilleroftheearth applications rank in the special needs category. Wouldn't you say?

I haven't seen anything on that yet. But my interest in SSDs is still superficial since it's something I'd like to have, but not something that I really need. I'll keep reading the threads though, and gathering more information as the Holidays approach.

 

Yes. GC needs to see a block as containing no data (NULLs) to be able to work with the block. The filesystem doesn't write NULL data when it marks a space as empty. The AS-Cleaner component of Tony Trim does that for you.

 

Krane, no pre-production vehicles for me, but I have been told that on many occasions in the not too distant past. lol...




chimpanzee,

I think you are getting this backwards.

First, sitting idle gets me 0% recovery.

Second, SE resets the DuraWrite algorithm to 'zero' and this is what reduces the life expantancy greatly. Running with reduced performance is the 'optimal' state that DuraWrite dictates to the SSD so that it doesn't degrade faster (how I laugh at that marketing spin as a 'benefit' of SandForce controllers).

I have many systems available to do a SE on - just haven't seen a writeup with a SandForce controller how one would go about it though.

Funny that you mention Scratch Disk. With all the testing I've done, I feel/think that that is ultimately the best use of any SSD currently (for my specific use: photo-editing).

No O/S, no apps, no other files folders directed to the SSD - except PS's Scratch Disk and the similar functioning but differently named options in all the other RAW editing apps I use.

When the SSD dies, and it will eventually - just toss, replace and continue working.


As to over-provisioning it by 30% or so - I bought the drive because it had 28% OP - doesn't make a lick of difference.


***Dies to me means giving less performance than a mechanical HD (VRaptor, in my desktops for example).

 

Go over to OCZ and Corsair support forum and you would know what I am talking about.

My advice for any Sandforce buyer, make sure your machine(intended to be used with it) is ok with it(other people using exactly the same and have a 6 months long good experience).

As for the performance, again no one can tell if your usage pattern fits for it.

I go for the try and true x25m because I know what I need and it is basically hazzle free, unlike a sandforce product where firmware upgrade is not an optional excercise but more like a routine suggested procedure by the vendors.

And again whatever SSD you get, give it a 20% OP(and more if your usage pattern call for it).

Because of all the above, I would not buy another SSD until I can get one in the 128GB range for less than 150 USD. The small size one which is quite reasonable now(up to 60GB) is simply not enough for modern OS when factor in the OP requirement.

 

What do you mean by getting it backward ? sitting idle is the official recommendation by all sandforce based vendors.

SE is just a massive Erase. It does reset DuraWrite but that doesn't shorten the life(in certain usage scenario). A NAND life depends on how many erase you do. A once off erase is actually more efficient than lots of erase when you cannot find erasable block whenever you write, if your usage pattern lead to that situation.

Sandforce's DuraWrite is more like 'because you have written to the drive for so much stuff in so little time, I would slow you down'. But the problem is, in order to complete my task, I have to have that write, whether it is done in 1 seconds or 1 hour.

SE is a standard procedure that applies to all HDD/SSD. There are bootable linux which you can try(basically issuing ATA command to the drive).

It is up to you to decide if you feel comfortable with my explanation.

 

no. that is not how sandforce work. That Tony is I believe from OCZ and they don't recommend this method on their sandforce line.

 

And I'm stating that the 'official recommendation' is worth zilch in how much performance is 'recovered' by following that procedure for me.

SE is not just a masssive erase: it resets the wear-leveling algorithm to 'new' and allows heavily used nand chips to be heavily used again (allowing them to die sooner).

I'm not sure what you tried to explain (sorry) - but gambling on a linux 'solution' is not something I'm willing to do.

 

Recovered to what? What you are seeing is the proper variable speed of the drive. Theres nothing to recover to unless you empty all blocks which takes out the garbage collection map with it. And for what its worth from the pictures you attached in the OP are pretty good, though you are using zero fill test data which has high redundancy and is thus highly compressible. It therefore gives you a sort of idealized picture of the transfer rate of the drive. To get the worst case scenario, run CDM's random test data. Realworld speed will oscillate wildly in between depending on the workload you are subjecting the drive to.

I'd like to point out that even using incompressible datasets (like CDM's random data), the drive is roughly as fast as an Intel G2 and gets faster when the data is compressible. At 0 fill you are seeing the upper limit of what it can do with on the fly compression.

So heres what baffles me. You are the first person I have ever seen that currently requires a drive that is faster than what the industry can provide at this point in time. I've never seen anyone say that one of the fastest consumer drives available for less than £200 is simply 'not fast enough'.

You can secure erase and it will recover the speed of the drive (I did it 2 days ago after reinstalling Windows 7. New desktop PC) and CDM is posting scores like it was brand new again. But you have to rebuild the GC map so if you keep on secure erasing to maintain this artificial level of performance you will kill the drive quicker. How quick I don't know but I've written 516gb to my drive and its reading 100% drive life left. Apparently the Toolbox correctly reads the life left SMART value.

I still find it unfathomable how you can possibly be disappointed in the speed. What hard disk where you using before you got the Patriot? Whatever it was I'll bet the random writes were at least 20 times less than what you have now. But the true test is not in synthetic numbers. The real test is whether you think its too slow day to day in your work. If it is, then what else is out there thats better? If its not then what are you complaining about?

 

Tiller, read this: http://forums.storagereview.com/ind...64356__hl__gskill__fromsearch__1&#entry264356

Actually his last screen shot is showing 25 MB/sec 4K random write. That's very low for a Sandforce drive.



Time to secure erase it. Or whatever procedure brings it back to factory state.

 

May I ask you guys something? I'm new to the SSD, but OCZ recommends SandForce SSD owners not to use CDM and AS SSD because they use incompressible data. They recommend ATTO instead.

Guide OCZ Sandforce (Agility2/Vertex2) expected Benchmarks and General Info

So, why do SandForce users in this forum still use CDM or AS SSD?

 

Because real world data is not aligned. Most real world data that we add above the os and programs, music, video, pictures and the like are uncompressible. CMD and AS SSD best mimic this real world situation.

The problem with SF2's is lifetime write throttling being enabled by DuraWrite. SE disables this by resetting and you get a new drive for a while. When it kicks in you loose 60% or more of your write speed. You don't even have to write a full drive worth of data to the drive before this kicks in either.

At least with the Callisto the LWT enables much too easily and I have yet to be idle for days on end where it disabled itself. So this leaves SE's about every one too three weeks with no real work being done with the drive other than internet caching and updates.

If they made it so if the drive saw just 25GB of uncompressible data a day before it enabled LWT I'd be a happy man. I don't even touch with my current work flow 256MB a day and I can garrenty having to do a SE. too me this is plain rediculous but that does not seem to matter.

The drive with LWT is totally awesome, if I didn't have a fat pipe, if you consider 25 Mbps fat, I'd probably never notice when LWT enables. but I do and also because of this I can't do a real workflow with the drive. If there were not two bays on my system I would never go for at least an SF2 as a primary drive. This is just m experience with the Callisto.

If I do the full chipset mods and get the super high throughput to the drive one run of AS SSD will enable LWT, so I do only half mod to the registry.........

 

Great info! Thanks very much! I should have done more research before getting a Vertex 2.

 

But the OS and programs are compressible. In my opinion those are more important for real world performance of most people. As can be seen in real world benchmarks, where Sandforce drives usually do very well.

The reason I wouldn't use CDM on default settings is because it writes 15 to 20GB to the SSD. Whether I'm using a Sandforce drive or not, it's not good for it.

By changing the settings of CDM it is no problem for SSDs. This is what Tilleroftheearth did.

I doubt you could have found a faster drive for the money.

 

I have a suspicion, although I'm not sure as I don't use the programs that tilleroftheearth does, that his programs don't use random writes as much as they do sequential. Either that, or because of all the sequential writes from his workflow, while it may start out "snappy", as soon as he actually starts working (so, like, 15 minutes after he starts editing) enough writing has been done that Durawrite kicks in and all of his advantages are lost. So maybe it's not so much that it's slow (at least not initially), but that it's so quickly incosistent that if feels slow. I think he uses Velociraptors (maybe RAIDed? I don't remember) in his desktop machines, and that's what he tends to compare his SSDs to.

 

SE is masssive erase(no matter you ask Intel or Sandforce or other vendors). Whether you want to do it is up to you. It would use up 1 erase cycle for all blocks, nothing more, nothin less.

 

Secure Erase is just a single pass rewrite of the entire drive that is executed on the drive level, not by software on the drive or by software on another drive. Therefore it is not easy for stuff like malicious software to interfere with the erasure so we say it is a secure erase. The purpose of it is mainly data security although there are some ancillary uses for it.

 

You are never going to get those scores in a notebook so don't even bother trying. If you take it out and put it in a desktop PC then depending on the chipset it may or may not be possible. I put my Vertex 2E 120gb into my desktop rig which is using an Intel p55/3400 chipset.

In my Sony F series, CDM reported:
~20mb/s 4k read
~45mb/s 4k write

when the drive was brand new.

In my desktop PC using the same drive, CDM reported:
~30mb/s 4k read
~70mb/s 4k write

Different chipsets will get you more or less mileage but there is absolutely no way you can tell the difference without comparing synthetic benchmarks. I don't notice any difference at all nor can a human brain even keep track of the small reads/writes that just your operating system is going to do.

In fact, my initial response was 'my laptop was quicker' because it doesn't load all the RAID mode crap at bootup that my desktop PC does so the laptop beats it to the login screen.

If you want to you can spend countless hours benching, testing, tweaking and rebenching, trying to chase numbers that you are never going to get and which beyond a certain point don't mean anything. But why would you do that? What you are inconveniencing yourself over is minutia.

 

Here's my Agility 2 120GB in a HP DM3 Notebook:


Double the 4K random writes of Tiller's Sandforce drive. That's why I'm saying his 4K random performance is low.

 

What chipsets are the both of you using? What tweaks have you enabled? Are these speeds pre GC map or post? Are these benches taken when the computer is subjected to a heavy cpu load? My 4k reads and writes reported as almost doubling when comparing idle benching to benching with a prime95 torture test in the background.

Lastly, it helps to keep some perspective here. A year ago 99% of us were all happy with 4k reads and writes on the order of 1 mb/s. In normal use, without stopwatching application loading, tracking writes and synthetic benchmarking, I notice zero difference between my drive in my laptop and my desktop.

So even though your numbers are higher you have to ask yourself 'what does it mean?' and 'does it make a difference that matters?'

This reminds me of the audiophile crowd where people obsess to the point of irrationality over audio fidelity. At some point the meaningful difference is so small it would take a machine to measure it. At some point the increase in sampling rate becomes counter productive when nyquist is so far beyond the audible range of human hearing that half the data you are storing may as well be thrown away because only a bat would notice it. Not to mention the idea that firing sustained ultrasound at your speakers is a good way to blow the tweeters. And yet people don't consider the practical implications of all this, only what theoretically should be better.

 

I fully agree with the fact that it will be very hard to notice those differences in real life. Tilleroftheearth just wants to get all the performance he paid for. I can easily understand that.

And no matter what way we look at it, 25MB/sec 4K random write is still low for a Sandforce drive.

PS. If you get higher scores with CPU stress it's probably because of the Intelppm bug that Tiller is well aware of.

 

I wouldn't say its a bug, more like a design feature with some consequences that we don't fully understand yet. I've tried disabling intelppm and I believe it was you that suggested I try it. However after some time I noticed from the Intel turbo boost widget that my cpu wasn't turbo boosting so I consider this an example of a cure that is worse than the disease so to speak.

I understand wanting to get what you paid for but what is it that we paid for? We paid for a drive that is competitive with Intel G2 for alot less money than a G2 sells for right now. Under certain situations and depending on the data being read/written (specifically how compressible it is) then the drive can actually be significantly faster. What we got is still in many situations amongst the fastest consumer ssd available to date. I can't really complain about that. Sandforce's marketing is based on ATTO. Therefore, using CDM scores to express disappointment that you didn't get what was advertised is not right. You, me and Tiller got what was advertised but if you feel you didn't then sell up or return the drive and buy something else thats more to your liking.

I feel as if this constant sense of disappointment and negativity is becoming destructive. I prefer to just use my drive, safe in the knowledge I have a good warranty. I'm happy with it whether its in my laptop or my desktop and its better than what I was using before for what I feel was a reasonable amount of money.

I agree very much with this.

 

Right. Not that I need a faster drive though... In a real world experience, I'm pretty happy with the speed of x264 re-encoding I'm getting with my Vertex 2 considering using a dual core chip.

JFYI, here is my CDM result from two weeks ago.

 

Installing Office on a PM55 chipset takes twice as long as on a PM45 chipset. If I disable Intelppm it happens on normal speed. That's a bug in my book.

How to avoid issues with Turbo Boost and battery life are described in the specific threads.

 

That's what I totally agree of.

That's was one of the reasons I've decided that I don't want so blazingly fast SSD, but bigger SSD instead. After some point you just get enough speed.
Also I've noticed that most consumer apps and OS seem to be optimized for HDD anyway, and try to curb random accesess - making them unlikely to take full advantage of massively faster random access times. Linear throughput still very noticeable though, on install/copy/unpack and backup operations - it mostly translates to proportional time decrease on these operations. Randoms - not anywhere near as obvious.

 

Recovered to what is advertised/promised/implied; without needing to decode the DuraClass and DuraWrite mumbojumbo that essentially states that you can claim to have the fastest drive until we decide it will kill itself so we will throttle it to below 5400 RPM and even (sometimes) less than 4200 RPM performance levels.

Here, I'm not comparing to Intel's G2 (although I have tried that drive too), I'm comparing to what I use to get work done on my notebook(s); currently a 500GB XT.

Yeah, I'm impatient, but I'm hugely disappointed that the SSD industry cannot match 50 yr old mechanical tech - yet.

If secure erase 'fixed' the issue, I would have done it already. But, thanks to DuraWrite 'technology' [sic], this issue is a 'feature' that is not highlighted enough (in reviews/advertising/etc.), so, in essense, SE simply interrupts a normal workflow AND is guaranteed to kill your drive faster.

Yeah, I am disappointed in the (non) speed of a drive that cost me almost 6 times as much as the 'steal' 7K500 for $80 (to which it is at best equal to - when used in my 'normal' workflow). What replaced both the 7K500 and the Inferno for 'real work' is the XT hybrid.

Not only is the XT really faster than my VRaptors (desktop) in some (important) things (can't say the same for the Inferno), but it essentially matches the Inferno for 'feel/snappiness' too, while being able to keep working at a constant level of performance even at 'heavy' workloads - which is where the SandForce based Inferno falls very short of mechanical HD's.

Funny, but I can easily tell the difference in how the system feels.

With a sub-optimal SSD inside (so far, all that I have tried... ), right-clicking on the desktop should not be a 2 to 10 second ordeal (for the menu to come up). Nor should a download take longer to finish the last 10 bytes than it did for the whole 10MB download (not an ISP issue: this is the SSD at it's worst), or waiting 10 or more seconds for the UAC dialogue box to come up while the whole screen turns black (thought the SSD, along with the current work, was done for - the first time this happened about a year ago).

This is not minutia this is a seriously backwards step in storage medium solutions. (These examples are why/where I compare SSD's to being slower than 4200 RPM drives, btw).

I have the PM55 chipset with the tweaks as noted in post #1.

I don't care what the 4K RR/W are or were - the drive is simply slower or faster than what I am currently using - in actual (work) use - not benchmarks (benchmarks=statistics=lies, lies and more lies).

I can see/agree that you can't tell the difference between using your drive in your notebook or your desktop: in another thread I noted how I cannot see the difference between the XT which scored '162' vs. the Inferno which scored '274' benchmarking copy speeds. (Note: I can 'tell' of course the difference in copy speeds in these static tests - what I can't discern is how the two drives feel in normal use, except that far too many times, the Inferno is slower - by a lot).

See:
http://forum.notebookreview.com/har...le-copy-result-hdds-ssds-easy-comparison.html


Audiophile's pursuit of sonic 'perfection' are exact opposite of what you're describing - 'weight', 'air', 'feel', 'mood', 'space' and 'soundstage' are simply not measurable with any instrument except the ears/brain of a highly trained (not irrational, as you assume) individual.

Worse, even the metrics that can be 'measured' by machines just don't correlate into real world observations by a skilled listener.

Hey! Just like the SSD benchmarks here!

No, I paid for a top of the line SSD including the associated price premium too, that I thought would be faster than any previous mechanical drive I've used.

I really don't care if its the fastest of the SSD's - that just shows (me) how horrible the other SSD's are then.

What was advertised (as I've mentioned already) was very cleverly disguised as 'truth'. Not only by SandForce (and partners), but also all the reviewers too.

Even after buying and using the drive (lightly) for just after the return period, I too bought into the hype. However, after putting the drive through my normal usage scenario, wow, how the harsh morning light (real world use) makes the shadowy beauty of last night a thing of regret and horror (Phil, I know, I'm exaggerating again! ).

Hayte, I'm glad your usage scenario is able to give you a satisfactory experience; I wish I was in the same boat as you.

Please don't take this as simply negative and destructive for its own sake.

I just hope that the SSD manufacturers see that their drives need (much) improving and also hope to spare anyone else with the same demands of their systems as me the time and trouble of testing SSD's for themselves. At the very least, they'll know why they're experiencing the issues I have gone through.



Again, I thank everyone for their input but from my perspective there is no real solution for me at the current time.

I had a brief flicker of hope that the OCZ SSD Toolbox would help a little. But, as I don't have an OCZ Vertex 2 drive, it won't allow me to do anything on my Patriot Inferno.

Patriot, their 10 year warranty notwithstanding, is one of those 'suppliers' of technology but not capable of supporting/fixing/expanding on their shipped products in any way. At least it looks like that by browsing their support forums.

In the very near future (no real rush) this drive will be relegated to 'usb key' status when I put it in an external enclosure. There, the slow real world writes won't worry me so much.

More importantly, it will be a constant reminder to myself that the next SSD I do test - I'll complete testing on before the return period has expired, like I did for all the other SSD's I tested/tried and returned before.

 

You were given ATTO numbers in the marketing material so use ATTO and surprise surprise, you'll get pretty close or exactly the stated transfer rates.

Well it sounds like you would be happier with an XT. So whats the problem?

This simply isn't true and leads me to believe that you have other problems with your computer that have nothing to do with the storage disk. It doesn't even make sense given that seeking on an SSD is an electrical process that is vastly quicker than the mechanical process in a hard drive (gas powered actuator and spinning platter).

Secure erase doesn't fix anything. Secure erase is for data security, namely to prevent data recovery from a drive that contains (or used to contain) sensitive information. It just so happens that people use a full drive rewrite to exploit Duraclass by resetting the GC map. You should not do this and OCZ (amongst others) don't recommend you do it.

One of the defining features of Sandforce drives is Duraclass. Together with on the fly compression/encryption, this is Sandforce's claim to fame. These are not bad things and I believe that you fundamentally misunderstand the purpose of them.

Feel is what is important in the end but you must understand that you cannot assess the speed of a storage disk based on a gut feeling. There are so many variables involved which is why synthetic benchmarks exist to eliminate as many of them as possible by using uniform datasets to determine transfer rate. In one sense they only assess an SSD's ability to perform the test but on the other hand its scientific because its demonstrable and reproducible.

Slower than 4200rpm hdds? A backwards step in storage medium solutions? This kind of hyperbole is not necessary. The elimination of mechanical parts, noise, near zero seek times plus low power consumption and low heat production is not a backward step. 2nd generation SSDs are not slower than 4200rpm hdds. What boggles my mind is that you require something that doesn't even exist yet. It would be like me saying that I'm disappointed in the car industry for not producing a road legal vehicle that can fly and float on water. So instead of driving right now, I'll wait for the model that can fly and cross rivers.

But but but. Have you even tracked drive writes? Do you know what an unbelievable number of consecutive requests the thing is processing? It is impossible for you to keep a mental tally, even an approximate one.

I'm talking about the industry trend towards higher bit depth and sampling rates. For example the trend towards 192khz and DXD rate converters when these are not only impracticable and unnecessary but there are pioneers in the field such as Dan Lavry that consider them to be worse. The popular assumption is that higher is always better and this is not the case.

They are faster than the previous generation of mechanical drives. Have you ever considered that there may be other factors to explain the sluggishness? That human error may be responsible? One of the reasons why I brought up the audiophile thing is that many people I know over at other forums that are audio enthusiasts tend to find fault with the technology (i.e. vintage analogue synths sound warmer than modern analogue synths) when they ignore a huge part of what it is to listen to and interprete sound which is the perception of it (and it goes way beyond auditory stimulus).

At this point I feel like I should retreat from this thread because I'm finding it frustrating. I believe that if I continue this discussion it will only get worse because our points of view do not look as if they can be reconciled. Hope all works out for you and you are happy with your XT hybrid. In the end all that matters is that you are happy with what you have and everything else is just gravy.

I agree. One thing I really notice is the lack of storage capacity I currently have because its prohibitively expensive for me to buy more. In fact, it has changed the way I use my computer because instead of using 1 very big storage disk, I now use an ssd for the OS and frequently used applications and an array of hdds for mass storage.

At this point the speed I think is the least of my concerns. I was much more interested when Intel announced G3 and they said they could double capacity for the same cost due to the NAND shrink. That right there is something that alot of people are going to see and feel the benefits of every day.

 

This sounds very weird and some kind of software problem instead of SSD-related. I've never experienced delays like this even on slow 5400RPM HDD.

I would agree that Sandforce seem to use a lot of smoke and mirrors in their products though.

 

That's the issue I got with my Intel 80GB SSD. Some apps on the SSD would take a good 10-15 seconds to load. Right clicking would also cause a slight delay. A secure erase would correct it for a while, but then it'd be back. I'm moving this drive from my desktop to my netbook now though so we'll see what happens there.

 

htwingnut, like I've said many times - I've seen this in every single SSD I have tried - in spite of their 2 orders of magnitude faster R R/W's.

The problem with discussing these issues with SSD's is that it seems its an opinion, an impossibility or a 'you're doing it wrong' kind of argument in response.

Hayte, please understand that I understand the issues fairly well.

That doesn't help though with making the 'magic' of SSD's appear on my systems (nor my clients either, for that matter).

Let me answer you as clearly as possible (without quoting your post this time):

I don't care about benchmarks in any way shape or form - I merely provided it as an example of where the drive was at.

Yes, I'm the happiest with the XT - the problem is that a drive 4 times the cost cannot approach it with the work load I've thrown at them.

There is nothing wrong with my computer, nor it's setup. The other thread I provided you with before is ample proof of that, (I hope).

You state that DuraClass/DuraWrite tech is not 'bad' - I say it's the worst case of misrepresenting a component which I bought for 'Performance' that I've seen in a long time.

How can I 'feel' a storage subsystem's performance? Easy: identical O/S install, identical apps/utilities installed (in the identical order) and finally identical updates installed (in the same order too) between drive 'XT' and drive 'SSD'. Now, with the identical data (RAW image files) first work for a week or more with one system drive, then do the same with the other.

The differences and shortcomings of one drive or another jump up and hit you over the head like a clown with a baseball bat. Or, maybe its just that I'm just very sensitive to (very) small changes?

The point to keep in mind is that there are no differences between the test setups at all: identical hardware, identical OS/apps, identical data used and processed/converted the same way.

No hyperbole - I've witnessed this and have been stating the same for over a year here on NBR - meanwhile, you're starting on one track (of discussion) and then going off in a whole different direction (which I'll politely ignore).

The technology exists for what I need (7K500, Seagate XT and VRaptors) - but I am constantly searching for better. Current SSD's are not there yet. Don't think I don't appreciate what SSD bring to the table, but those 'benefits' are not my pressing issues: performance as it relates to productivity is.

Keeping track of writes? Hmmm, I don't think I have to. As stated above, in identical systems with identical installs/data used, the only difference being the HD/SSD in the system? No, I don't have to. I 'feel' it and see it (sometimes blatantly) by just doing what I've done for the last decade or so with the (too many) RAW images I need to process/convert.

Also, being the exact same usage, the 'number of accesses' is the same in both tests (HDD or SSD used) and, it is scientifically reproducible: put the HDD in the system and the response is normal - put the SSD in the system and the response is sub-optimal.

I can't see how 'user-error' can be involved here. With even more effort spent on the SSD install (JJB's and stamatisx tweaks, for example) than on the HDD install and the obvious differences seen between the two drives, what possible user error(s) can (is?) being introduced?

I'm ready to say I'm wrong - but I want to see a solution where the SSD (any SSD) is finally faster for my use than the HDD systems that I am comparing them to.

What makes me 'happy' is a system that is balanced in performance with everything I throw at it. With an SSD, that balance goes from 'insanely good' to (sometimes) 'insanely horrible' much too fast for my tastes (10 days for the SandForce, in this case).

I too do not wish to frustrate you further - but your response is frustrating me too: you offer no solution to my issues - only reasons/'logic' as to why I shouldn't be having them.

I'd be surprised if someone offered a solution now in this thread - hopefully I'll be seeing much better performance from SSD's in 2011.

 

If you don't care about the life of SSD and the hassle of doing secure erase, just SE it every ten days if you want the 'factory' performance of sandforce.

 

chimpanzee,

And who should I hire to do this (cheaply)?

And, isn't SE still the 'hassle of doing secure erase'?

 

Tilleroftheearth, I know you have a more extreme case of using HDD's/SSD's, and that does compound the issue. But I've also run into resistance even on other forums when mentioning issues and the performance isn't really worth the cost at the moment, and still have significant issues to work out. It's like you can't mention anything negative about SSD's without getting lambasted from all directions.

I love the idea of SSD's, but when those issues crop up it ruins the whole idea and concept of an SSD and makes you feel cheated from the significant investment you made in this newer technology. And just because others don't have the same issues as you do doesn't mean they don't exist. People turn a blind eye.

 

No idea, I am not advocating it, but you are asking for solution to fix the performance issue.

For me, any SSD(or HDD for that matter) requires me to reopen the box(be it firmware upgrade or SE) again after the first installation is out of my list

That is why I settled with x25m. Whatever I gain in productivity(in terms of performance assuming sandforce is faster), I lost to these processes(and could easily pay back more, there are already two firmware upgrade to fix bugs since it first came to market).

Even for the x25m, I only upgraded one of my 3 notebooks as like you I found that a well tuned 7200 HDD is not that far behind.

 

htwingnut and chimpanzee, thanks.

Just that (like I've stated... sorry for the repetition...) SE is not a permanent solution - with DuraClass/DuraWrite specifically in this SSD, it will simply happen again.

Curious that you have the same perception too chimpanzee - which 7200 HD's are you using? And what is your usage scenario?

 

Scorpio Black 500GB. Though I found that any 7200 rpm one is faste enough. My usage scenario actually is less demanding like yours(I assume photo editing which requires large amount of write).

I am a developer so I only use a handful of programs and are always opened(load speed thus is meaningless). I observed the disk activity comparing the X25M 80G vs a 7200 HDD. What I found is that while the X25M is definitely faster(both in usage and the process monitor number), it is not that much of a difference for my usage(basically I cannot use a stop watch to measure it)

 

SF2 performace from what I've seen is pretty good so far. You do have to live the SE route though and defer your workflow. I would never suggest one for a single drive system.

IF Sandforce would reset Durawrite to stop enabling LifeTimeWriteThrottling so easily and also get it disable it more readily it would be a first great step. As it is I SE so often I never get to see if TRIM actually ever realy does anything.

It took me a long time and alot of general BULL before I could get one of the OEM's to admit about this flaw of LTWT in these drives. Actually they call it an enhancement of Durawrite and it might have been if properly implimented.

Since not a single OEM or Sandforce will admit that LTWT is an issue that needs addressing anyone buying a SF2 based drive right now knowing this is IMHO being extreemly foolish. Go ahead and purchase one, just remember no one is looking to fix it as no one admits there is anything to FIX!!!!!!!!!!!!!!!!!!

 

tilleroftheearth, have you at least tried HDDErase?

 

Phil, no I haven't.

Is HDDErase different from a SE by any chance?