SSD Thread (Benchmarks, Brands, News, and Advice)

yeah indeed. and the 80gb can be bought for around 250$ here right now.. a bit more, but not much.

 

Lucky you... UK is 17% I think.... (was 19% though and will rise again) and I think Germany is 19% too...

 

lol, even anad knows more about trim than you
http://www.anandtech.com/storage/showdoc.aspx?i=3531&p=10

as I said BEFORE...

when you delete data, ssd will remove pointer to memory location, ok ?
so, after rest of drive goes through same write cycle, a new data need's to be written on deleted space, problem is that, when you are writing, you can write a single block, but, if you wanna to delete single block, you first need to:
1. copy the rest of the blocks in the same page to cache
2. remove pointer from whole page cause page is smallest data controller can unmark form memory
3. write data from cache to new, least number times written cells

and, after that, all that remains is to write new data in least number times written cells, and that is step 4

what trim does is steps 1-3 right after you hit delete, so, after new write is not slowed by copying old data to new, least number of times used memory location...

and that, my ladies is what trim does




happy new ear to all !

 

no, it won't remove pointer to memory location

only the FILESYSTEM will do that. the ssd won't know anything. it just updates some random cell which happen to be the filesystem structure.


deletion, before trim, made NOTHING on the ssd. NOT A SINGLE BIT CHANGED.

only the filesystem gets updated. and that, is, for the ssd, just another file that it doesn't understand AT ALL.


i know trim perfectly. and i know the same about it than anand knows, and we are in 100% agreement.

 

it was 17.5% when i was in uk ten years ago...

 

Hm. So this would be the 34nm, new controller Intels then? A summer refresh would be on schedule with what they did last summer, though I wonder if I should wait for the price drop/refresh that's going to be coming after that. Plus, at that point, we're going to have to start worrying about whether the drives are SATA6, or Lightpeak, etc. etc. Hopefully there'll be Lightpeak to SATA adapters, because this next drive I plan on buying and sitting on for years and years.

Hopefully CES will provide some insight on what to expect.

 

yeah, can't wait for CES. now with the sandforce preview, intel knows what to show they need to show something that sandforce knows it can just drop it all and forget it

we'll see. CES and NAMM, both in january. for me, january is the real christmas

 

controller must know where to look for data when os ask him to
is he using pointer or whatever, that I don't know

that's true, flash memory itself would do nothing when you hit delete
I also agree with anand in terms of trim, don't get me wrong here

 

So if I'm understanding Wikipedia correctly, Light Peak -> SATA6 should be no problem?

 

control has an Adress translator, which tells which memory-address the os uses is which cell. this is for the wearleveling.

but even without that (see hdd), they just don't "get" deletes at all.

but on a hdd, that doesn't matter, as you can "just overwrite" at will. on an ssd, you can't. well, you can, but it's terribly slow.


all that happens to get updated is the file system index table. but that is just a non-readable file for the ssd. it does't understand that table AT ALL. so it just knows "uh that file got updated", and not "uh, there's an entry gone in the file table, so i could free some memory".

 

darQ96,

You're getting confused between data states on the nand itself (0,1,2,3) and the logical states on the controller of that nand.

Used=written to; Clean=ready for a write. Along with the other logical states of 'Used least', 'Never Used' and 'Don't Use Again'; a controller doesn't care about anything else for a particular location on a nand chip.

Sorry, but wiki is the last place I go for information. Ever. Doesn't matter if no other place has the info I'm looking for - wiki is non-info to me.

darQ96,

if individual cells are able to change their states independently, that is where error correction would kick in and/or the user would return the drive as faulty. Why would you want each bit to flip by itself? I'm sure you don't, but that is what you wrote above.

Continuing, that is exactly why writes slow down with a 'used' nand as the source/target - it is doing a read/modify/erase/write instead of just a write. You can see how doing 4 operations is longer than 1, right - especially when one of those four operations is extra long (in SSD time) like the Erase cycle?

The wear leveling algorithm is fairly simple and its execution is easily done in real time - real time of the SSD that is. I have not heard, nor read of one controller that was slow because of the wear leveling algorithm used. As an example, the SandForce controller I was just reading about can not only keep the wear leveling invisible to users - but it is also able to analyze and decide which data should be compressed or not - again, in real time - without a time penalty. This is not with a Core 2 Duo CPU either, but a lowly x86 based (I'm assuming) CPU that runs at the Mhz level.

Wear leveling and analysis and compression and Sandforce's ' secret sauce' - with no time penalty.

Read/modify/erase/write i s the biggest time waster on an SSD.

darQ96,

I have no idea what you're writing, but I do hope to learn something from you too (as I already have in the past, btw).

Again, wiki is not a source of information.

Do you have other sources you'd like to share? If there is a possibility of learning something new, I'm all for it, but the information and the way you're presenting it now is factually wrong from all the info I know. Prove me wrong!

SSD's are much more than 'dumb' flash chips strung together. I do not pretend to know all their intricacies, but I think I know about them at a level that is proven time and again that the controller is what makes an SSD an SSD - not the decades old 'flash' tech behind it.

Your knowledge of flash/nand may be useful; when you incorporate it with a fuller understanding of SSD controllers too.

Cheers!

 

Nope.

Incorrect. The drive does not know which memory locations are used by the filesystem. Only the filesystem knows that, and the drive controller does not understand the filesystem. You seem to be unable to grasp this simple fact. The drive is a simple bit-bucket; the data held in a block will be overwritten only when the operating system tells it to do so. And that will happen only if the operating system chooses to overwrite the block, which it will do only if the filesystem indicates that the block is unused. None of these decisions are made by -- or even visible to --- the drive controller. Thus the need for TRIM.

Your experience with flash access from embedded ASICs or processors is inapplicable here. The embedded ASIC/processor was likely handling the duties that would be handled by an SSD's controller, thus the disconnect between the controller's view (blocks) and the operating system's view (filesystem) did not exist. Apples and oranges.

 

Maybe Anand could learn something here too Seriously, this is an interesting point in the thread.

Not being highly technical, it sure makes sense to me that the controller IS the SSD. No maker has ever said, My flash chips are better than yours. (well, maybe a little) but many have said, my CONTROLLER is better than yours.

No matter what one's viewpoint on the tech, I think we must all take a moment to acknowledge the HUGE and by huge I mean GIGANTIC changes that have happened with SSD's in the past year. In size, number of players, the overall acceptance of MLC, the new players and the new generations of existing players.

Welcome Seagate! OCZ, you have changed for the better! Where is my affordable SLC SSD at say 128GB??? SSD's rule! I enjoy the further silence of my computer as much as I enjoy the speed and the fact I know there are NO moving parts!

Come one, come all to the 2010 SSD festival! What a year it is going to be!

 

it's not a bit bucket, it is well organized array
and where is filesystem info stored ?
all I'm trying to point is that ssd, as a one complete unit, must know what is used and what not, other than that, if you put filled ssd in another computer, it would not know what is on it

 

jeah, and trim ain't gonna to say it to ssd, trim will simply perform those 3 steps to speed up overwrite

 

well, no, sir, trim sais it to the ssd.

read up the spec for trim.

when you delete a file without trim, all the os does, is update the FAT.

if you delete a file with trim, the os updates the FAT AND REPORTS THE REGION THAT GOT FREED TO THE SSD.

THAT, sir, is the job of trim. NOT overwriting or anything. this wouldn't even work, as it would, then, AGAIN, have data on it. overwritten data, but still data.

 

no, the ssd has no clue what's on and what not. it just knows "uh, someone writes to me to that adress, let me save it somewhere".

the file system is just such data, the ssd does NOT understand that. for the ssd, the ntfs file structure and an mp3, and a movie, and a bootsector is ALL THE SAME.

 

well, I got all info I know from all over the web, so, posting all links would be pain in the ...
cell state and controller state, maybe
about other stuff, I've tried to explain that writing to davepermen...

I like to learn new stuff here to

 

ok. so lets clean up that whole mess completely. an ssd, like a hdd is a dump container for data. it has no interpretation of that data.

what EVER you write into such a container is, for the container, 100% gibberish. it does not know files, file systems, your photos, your texts, your music, the os, partitions, ANYTHING. it's just data that has to be stored at a certain location. in case of a hdd, that location is the actual physical location. in case of an ssd, it's a virtual location (that means, it keeps track of itself, where something really is).

the disk has 0 information about what that data is, and shouldn't have.

before trim, the disk had 0 information if certain of that gibberish is actually not needed anymore. why? because a hdd would never care about it anyways. if it's not needed, why care? just let it be where it is, it doesn't hurt.

BUT, for an ssd, it hurts. so now we extended the very complicated communication protocoll for disks from ONE command to TWO commands:

before, it was only this command:

WRITETO(STARTLOCATION,ENDLOCATION,THEDATA)

now it's

WRITETO(STARTLOCATION,ENDLOCATION,THEDATA)
and
IDONTNEEDTHISANYOMORE(STARTLOCATION,ENDLOCATION)

the second one is called trim.

that is ALL the os and the disk use to write anything on disk, or clean it away anymore.


anything else, you dream yourself up, dar. it doesn't matter how flash works, how it managed it all, if it's 1s or 0s (that's, btw, fully interpretation dependant), etc. all that matters is, reporting what gibberish-region is not needed anymore HELPS an ssd to do it's job better: to quickly allow accesses to the gibberish, and to quickly update the gibberish.


oh, i forgot another command, the most important one:

READFROM(START,END,TARGETADRESSINRAM)

 

looks true: http://arstechnica.com/gadgets/news/2009/06/high-nand-cost-delays-ssd-rampup-to-2010-2011.ars

 

actually our prices should right now be around 2$ per gb. from what intel planned at the launch of the g2. sadly, it's not yet happening. but i still hope they get <1$ per gb this year. then, no one would have a reason to not get it, except for storage-only parts.

 

i think i don't need a sandforce yet..



mine works quite fine..

 

true, all of that sandforce hype might just be a new algorithm with data compression trick, its controller hardware reliability is still unknown. performance wise, its 4k random read performance is even worse than intel.

 

jup. i stay with intel for the future.. they don't do tricks to perform well. they just deliver nice performance. i like

can't wait for 2.5 and 3.0 from them

 

Yep, found a deal in Oz with it about $500AUD.... which is good because here it is usually $800-$1000+!

Just wondering if I don't really need the storage at the moment if that extra speed would be better.... not sure if going to have any issues with it in my Envy 15 either.

 

for a laptop, get an x25-m, not an e. it's not tuned for desktop/laptop usage, but for server tasks. the x25-m has repeadly proven higher speeds in stuff like booting the os and similar, due to it's fine tuning towards consumer use. the E on the other hand dances around the M when you put in an sql database and do thousands of queries a second, or similar workloads that happen in server environments.

 

Happy New Years everyone

I must say I'm satisfied with the MSRP of the G2 and some people have gotten them in that neighborhood. I will refrain from more complaints about sellers. $2/GB is approaching. I got $414.99/160 GB= $2.59/GB

I want to talk about that Sandforce stuff, specfifcally in relation to OCZ. OCZ has made the decision to move forward with Sandforce for several good reasons, one of them being Indillix had to delay the successor to the Barefoot controller. Perhaps OCZ had already decided before the delay the Vertex2 wasn't going to use the new Indillix. My question is: what about the Agility? Would OCZ just kill it alltogether? Use the Sandforce 1200 in it? Use the run of the mill Indillix with cheaper NAND? Wait for the new Indillx and use that instead? OCZ did allude to maintaining several lines of SSDs.

 

ocz just takes anything that is a "fast ssd". i guess they will take the successor of the indillix, too. and sell it under yet another name..

we'll see what they do about cheaper ssds, but so far, the sandforce will be for higher prices again (but with better performance, obviously).

 

has anyone seen this thing(832G 2.5" SSD) on market after it was shown in last CES 2008?

 

i guess not..

 

i doesn't mark anything....
blah
I give up....
live happy with your new improved OS-ssd relation called trim

 

it does mark. the CONTROLLER has to know where what data is, and if it's used or not. for that, the controller has a map, and THERE it's marked to be empty or not.

you really don't WANT to get it, right?

everyone knows that ssds are BAD at writing to if it's allready overwritten. the degeneration part, and all that. this is ONLY because the cells are not "free" anymore.

and, get that, trim lets the ssd free them early again.

really, dude, how hard is it to accept what everyone including any manufacturer knows since years, and which led to that technology the way it is now?

 

marking or mapping has noting to do with trim
what trim does is moving one page to another memory location without block you erased, cause you can't delete blocks, only whole pages on ssd-s

but, as I said, i give up, trim marks staff etc etc, have it your way

 

Interesting, taken from rocket Disk web site:

"Mtron are poised to remain at the forefront of SSD technology. The recent announcement of the Neo Series has taken the indsutry by storm, with Read/Write Speeds of 260/240MB/s and 8,000 random write IOPS (using 4KB blocks). The Neo is scheduled for release at the end of Q1 2010."

 

no price no care.
lol....

 

Interesting how MemoRight and Mtron seem to have a lock, along with Intel, on the Enterprise SLC market and that this thread seems about 99% MLC. This technology in many ways seems so early that it is like the original 5 MEGABYTE hard drive, and then the Plus Hard Card came along at 20 MEGABYTES and set the world on fire

I used to sell RAM all day long at EggHead Software for $50 per MEGABYTE. That would make a GB currently $50,000 DOLLARS You really gotta laugh at that

 

darQ96,

Don't give up - listen to davepermen - he's trying, like I am, to help you understand.

But, that last sentence above is your proof that what you are saying is wrong.

Degrade an SSD so that there is no mistake that it is degraded. Make sure you do this in an O/S without TRIM and don't run secure erase or any other 'restoring' operation on it. Remove it from that computer and now, install it in another, again with no TRIM. Format the drive and test it.

That same degraded performance will still be there, because the controller DOES keep track of what is 'used' and what is not used or 'clean' nand chips.

If you're able to do this test - you'll see your version of reality blown sky high. And, you might finally see where we're coming from.

Really, in all your 'reading all over the net', the continuity of logic got broken somehow. Hope we can fix it for you.

Cheers!

LaptopGun,

Did you see my post here earlier:
http://forum.notebookreview.com/showpost.php?p=5681921&postcount=2576


Somehow, I get the feeling that OCZ would maintain several lines of SSD's; that is until the money stops coming in their direction.

 

darQ96 obviously doesn't want to understand. Or he's one of those people who won't even consider that he might actually be wrong about something.

At least anyone else reading this thread has been provided with sufficient accurate information from a few of us such that they can hopefully recognize darQ96's factually incorrect statements (or decide that they don't care).

 

jeah, well, I like to learn, but stuff you guys write are so wrong even I'm able to see that

there actually are some 100gb slc sammys, but they are very high priced, around 800 usd
MCCOE1HG5MXP-0VB
http://www.ssd-info.de/ssd/samsung-flashssd-server-25-100-gb

 

trim doesn't do that. the ssd might REACT LIKE THAT ON TRIM.

trim is ONLY about informing what data is not needed anymore. the ssd can then chose what it wants to do with that information.

this is what you just don't want to understand.

 

I didn't know Samsung had tried to crank their SLC drives up to 11... I mean 100GB.

Thanks for the responses guys about OCZ. Just some of my patent-pending random tech thoughts. WHo knows what the Hell OCZ is planning for their lines. I remember Anand characterized the Vertex in his original hands on as OCZ's midgrade and the then forthcomming Summit as the high end. All OCZ needs to do is higher the people who came up with Nvidia's naming scheme (8800 GTX GTS KO GT Infinity X2 CUDA...) and they'd be poised to take the world by storm. I'll stay for right now with the idea "we generally know what the Vertex 2 will be"

 

As has been stated numerous times, trim merely tells the drive that a page is no longer in use. The controller may choose to retain that information for later use -- e.g. to optimize how it moves data around. What happens at this point varies from controller to controller, but it is highly unlikely that the controller will wipe the data in that page (or move data stored in other pages in the same block) immediately upon receipt of the trim command. A controller that did so would perform poorly, negating most of the benefit of the trim command. A noticeable slowdown would potentially occur upon delete/trim, and the controller could not optimize its performance as fully as it could otherwise do in the presence of more information (e.g. it doesn't know what pages might be written or trimmed a few milliseconds later, etc).

So you could say that trim might result in the page being moved at some future point. Or it might not. Trim merely gives the controller the information it needs to make those decisions. It's up to the controller to decide what to do with that information -- and when/if to do so.

It's possible that the language barrier is preventing you from understanding what is being said here. Or not. Either way, enough on this topic.

 

has to be beautiful there, i hope. with fairies and unicorns and such

actually, i like the reality. it makes trim very simple, to understand and to use. and so i'm fine with not being in fairytale-land.

 

Here at last is my SSD Benchmark after using it for 2 weeks.

SSD: Intel G2 160GB SSD with latest TRIM Firmware
OS: Windows 7 Ultimate x64
CPU: Intel Core 2 Quad Q9650 @3.0GHz



I forgot to enable AHCI before installing win 7 in all of my excitement, so the middle bench is without AHCI enabled. The pictures on the left and right are with AHCI enabled, and the Rapid Storage Matrix Manager Drivers (whatever they're called now) installed.

How do those numbers fare compared to others?

 

darQ96,

And, I wish that you wanted to learn and share the knowledge you have with us. But you are trying very hard to do neither.

 

I actually agree with you here
those steps I've tried to describe was general situation, but, what and when controller will do depends on large number of different situations, and, ofc, controller type ( intel afaik has lowest write amplification controller in market, nice stuff )


but all I wanted to point out, at the begin of this mess, is that, in my opinion, ssd should be able to take care of its performance without OS (trim) or user (wiper and stuff) interference

 

ok, let's try sharing first
http://windowsitpro.com/article/art...state-disks-ssds-and-why-is-it-important.html

 

and we pointed out that you're right, the ssd should be able to handle all workload better. but you're not right that the os should not inform about deletions, as it informs about saves just as well. the os informs what it wans to be stored on the ssd, so it should inform if it doesn't need something anymore. that's just fair, isn't it? if i come to you and give you something weighty into your hands saying "please hold that for me", you would be happy if i would say, now you can put it somewhere on the floor, you don't need to hold it anymore.

trim is about synchonizing between os and ssd, something that actually should ahve been done from the start (on hdds, too).


and why you can't agree on that, while i agree with your "the ssd should take care of itself", is not explainable to me.

while any manufacturer, the os writers of all major operating systems, etc all agree.

 

Just wanted to say that I used to enjoy visiting this thread to read up on latest SSD developments and stuff.. until this thread started turning to this discussion about wear levelling, trim definition, etc etc and turned extremely dull and boring...

 

we're just annoyed that there are no real new ssd news to discuss

wait till CES and we will talk dreams of what will come soon enough