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

Heard that 20nm chips are made by Intel and Micron. They are under test producing. If all will be OK they will start mass producing in the 2-nd half of this year.

 

Previous page !

 

Yup. except (3) doesn't need to be written as it contains invalid data thus is free to use. The only reason it cannot be used is that it is used before. That is also why TRIM was introduced. During the course of operations, there are many such pages and without TRIM they needs to be kept as far as the controller is concerned(even though from the OS perspective, they are no longer in use).

 

For those people who don't like OCZ...

The OWC Mercury Extreme Pro 6G 120GB SATA 3 SSD Review – This 120GB 6G SSD Absolutely Flies! | The SSD Review

OWC Mercury Extreme Pro 6G 120GB Solid State Drive Review :: TweakTown USA Edition

 

I thought the controller had no clue about the data being valid (2) or invalid (3); that's why I said RE-Write (2) and ( 3) state pages...

However, if the controller DOES indeed know if the data is valid (2) or not (3), then, obviously it should not re-write (3) state page's data because if it does, it needs a serious spanking...

 

The controller can deduce whether a page is in (2) or (3) via different means. TRIM helps it a lot. How it deduce that distinguish a good controller from bad controller(early day SSD). With TRIM, they become more or less the same, in this aspect.

 

Question then is this:

Is a page containing TRIMmed data (so nothing but not erased) in a (1), a (2) or a (3) state ?


It can't be (1) because it's not erased, nor brand new...

It can't be (2) because there's nothing written (neither valid nor invalid)...

It can't be (3) because there's nothing written (data has been trimmed, deleted, but not erased for that matter)...



Unless TRIMmed data has been erased by the said TRIM...

 

TRIMed page is always in (3). It only tells the controller that this page is not in use anymore. But the controller cannot write to it(because it has been written to before) unless the containing block has been erased. TRIM is an 'information command', not an 'action'.

 

If this is true, then, there's no benefit trimming (speed-wise) because the controller still have to erase before writing; so, apart from setting the page from (2) to (3), what's the gain of trimming then ???

 

If you don't TRIM, either the controller has to find out a page is in (3) or it has to treat it as in (2). Therefore, an SSD without TRIM would carry lots of dead weight as time goes by which reduce performance and endurance. This can be migitated some what by over provisioing(such that the controller has much better chance of deducing a page is in 3).

A typical case. If you quick format an SSD that doesn't support TRIM(but has been written to once). The performance of the SSD would tank. But if you quick format with one that supports TRIM, the SSD performs like brand new(except SF which has its own logic but TRIM still help).

 

I agree with all you said, but again, I was talking SPEED-WISE:

if the TRIM command is only an information command, as opposed to an action command as you said, then, SPEED-WISE, there's no gain from trimming because (3) state pages still need to be erased before they can accept any new data, trimmed or not...


I've always thought that when a TRIM command hits the drive (controller), as part of a delete command from the O/S, the data is indeed eradicated, like if the page was back to square 1 (1) state.

If I get this well, TRIM only changes the state of the page from (2) to (3)...

Therefore, it's only benefit is for the controller to distinguish between (2) and (3) state pages...

That would mean any (2) page can't be changed to (3) on a NON-TRIM capable SSD...

That would mean no data could be deleted (erased) on a NON-TRIM capable SSD...

Scary indeed...


I got to go now, but I sincerely appreciate this discussion, as we are working (hard) to get the truth outta this, which is more than helpfull for the community.

I'll be more than happy to come back asap.

Rgds,

eYe

 

20 nm 8GB MLC NANDs from Intel and Micron going in to mass production second half this year


Intel And Micron Introduce Industry's Smallest, Most Advanced 20-Nanometer Process - HotHardware

 

Well with this, maybe they can fit 60-64GB packaged in an existing hard drive for best of both worlds for laptop use.

 

Noop. If that is the case, the WA will sky rocket. NAND can only be erased in big block. All controllers defer the TRIM some what in order to reduce the read-erase-write overhead i.e. WA.

 

Would it be smart to RAID 0 the new OCZ Vertex 3 120 gb's? People keep telling me it's stupid to RAID two ssd's because of the lost of TRIM o.o Thoughts? Suggestions?

 

I think they already can, I just don't know why they're not doing it.

With a single platter hard drive there should be more than enough space.

 

I can't tell about OCZ drives, cuz I never used them...

However, I've been using RAIDed SSDs (samsungs and intels) for over a year and a half now, so I can tell you a bit about it...

As a matter of fact, if you RAID 2 SSDs, you will loose TRIM.

I don't recall about the samsungs, but I cant tell you that with the intels, I just need to quick format them (upon windows's installation) about every six months or so, and they are like brand new. I think you would need to Secure Erase the OCZs to get to the same result; maybe my friend chimpanzee could provide you with more specific on this particular matter.

This being said, from my humble point of view, the benefit (speed) I get from RAIDing 2 SSDs by far overcomes the disadvantages, like having to quick format every six months or so; this is, and will always be a personal choice, it's a matter of preferences, there's no such straight line to follow and from which you cannot deviate.

So, to answer your question about being stupid or not to RAID SSDs, I think you might get as many different answers as there's people to give you their advice; thus, I guess you will have to find your own answer, as there's indeed no absolute truth about it.

However, to help yourself make a choice, you might wanna answer those few questions:

Why am I going to SSDs ? (Fun - gift - people's advice - speed - you-name-it !);
What am I expecting by switching to SSDs ? (same choices...)
Why am I targeting specifically the vertex 3 ?
Knowing SSDs are WAY faster than HDDs, why am I looking at RAIDing them on top of it ?
As a result of the lost of TRIM, am I willing to perform some kind of Secure Erase or quick format or so, in any case, am I willing to reinstall windows (cloned or not) every six months or so to recover the lost of performance resulting from not having TRIM ?


Your own answers to those questions might help you to fix your choice as to which SSD you'll be going with, and will you go with a RAID pile or not.
Again, there's no fit-to-all-answer on this, it all depends on your personal choices, goals and wishes !

If you decide to go in RAID, you might wanna take a look at the link in my sig about SSD degradation in a non TRIM environment.

For me, speed's gain overcomes the hassle of reinstalling windows every six months or so; what about you ?

 

Question is still unanswered:


If I get this well, TRIM only changes the state of the page from (2) to (3)...

Therefore, it's only benefit is for the controller to distinguish between (2) and (3) state pages...

That would mean any (2) page can't be changed to (3) on a NON-TRIM capable SSD...

That would mean no data could be deleted (erased) on a NON-TRIM capable SSD...


Am I missing something ???

 

Yup. TRIM only change page from (2) -> (3).

Without TRIM the controller has to determine it by itself. An example, we have LBA 1 and 2 which maps to NAND page 1 and 2, initially. When the OS write to 1, it becomes stage (2). When the OS rewrite it again(either it delete it at the OS level and later reuse it, or simply rewrite for the same file), the controller would change the mapping such that LBA 1 now maps to NAND PAGE 2. And NAND page 1 becomes stage (3).

That is why without TRIM, over provisioning is very important.

 

In a non-TRIM environment, how is the controller capable of changing a page from 2 to 3 if that's exactly the benefit the TRIM command provides ?

In your example, how is nand page 1 capable of becoming (3) from (2), in a non-TRIM environment ?

I am trying to understand how a page can become (3) from (2) with no TRIM, as TRIM is the command that tells the controller about it; how was it figuring it itself before TRIM was introduced ?

In a traditionnal HDD environment, when the O/S sends a delete command, the data is not deleted, only the index pointer, so the clusters (sectors, what you will) still hold the data, and this is not a problem as an HDD can overwrite data, which SSD can't do without erasing first.

The same goes with an SSD, data is not deleted, only the index pointer, but to the contrary of an HDD, the SSD has to clear the cell (erase the content) before it can write to an already written cell (page); it cannot overwrite data, it has to clear it first.

I guess my question is this: How was the controller any able to distinguish between valid (2) and invalid (3) data prior to TRIM, if TRIM only tells the controller this particular page is in a (3) or (2) state ? How does it find it by itself without TRIM ? Because if the controller is able to figure out by itself pages are either (1), (2) or (3), what the heck is TRIM for ?

 

That was what my examle tried to tell. Assuming there are only 2 pages available on the SSD but it expose as 1 page to the OS. The OS can write to 1 page only. After than it is FULL. To reuse it, it needs to overwrite on the same page. But internally, the SSD keeps on switching the mapping from NAND page 1 to 2 and back and repeat. No TRIM is involved. That was how SSD without TRIM functions. By over provisioning, it can have room to make this determination.

 

Hoping, hoping, hoping. Who's head do we have to bang?

 

it knows more early that the page is not in use so it can erase when he has time, instead of only the moment it gets overwritten by new data from the os.

 

I always thought that pages go from (2) to (3) when You delete file (pointer to file), and from (3) to (1) when TRIM command is sent or when GC kicks in.

 

I did try and put an order through but they only had 1 left when i ordered and they had that reserved for someone else. I was hoping they wouldn't turn up to collect it.
Anyway, that's gone and i'm now probably going to get the V64 Corsair Nova SSD.
Corsair CSSD-V64GB2-BRKT Nova: Amazon.co.uk: Computers & Accessories

 

no it stays at (2) if no trim is sent (if 2 means in use and written). once trim is sent it goes to (3) (if 3 means cell is marked as not in use). then, once GC kicks in (every ssd with trim has a gc. some have one with a timer, some have one instant after every trim, or some variation of that), the ones with 3 gets cleared so they're (1)

otherwise, a gc just checks which cells are partially written, and fuses them to fresh ones to free more of the others. a gc never knows which cell is not in use anymore without trim. never. that's the job of trim.

 

I read some reports that Indilinx drives have high failure rates. I don't know if it's true. Can anyone confirm or deny?

I've had very good experience with this drive, good performance and best battery life:
http://skinflint.co.uk/a453277.html

 

Thanks Phil
Finding one in stock for under £100 (incl shipping) is a problem.
I think i will raise my budget just a little bit more and get this-
Kingston SNVP325-S2B/64GB 64GB SSDNow V-Series V+ SATA 2 2.5 Drive Upgrade Kit: Amazon.co.uk: Computers & Accessories

I believe that is the drive that you recommended.

 

Same one. If you can find the V+ 100 cheaper that would be a bit better.

 

Thanks!

This sounds strange, because on my system TRIM is disabled for more then six months now and SSD is working fine. Actually it's working better without TRIM then with it.
I think that GC works by looking at MFT/FAT and labeling cells that aren't used to state (1). Just like GC in Java works.

 

Which SSD can understand MFT/FAT ? Most would only know LBA. Or else it would screwup data if it happens to be used for Linux ext3 rather than NTFS/FAT ?

drive working better without TRIM ? which model is that ?

 

Or LBA it is.

Super Talent MasterDrive SX, p/n: SAM64GM25S. It's same drive as Samsung PM800 or PB22-J and OCZ Summit.

 

so it has to know every filesystem in every version to know what is valid data and what not? believe me, filesystems like ntfs are NOT simple to parse.

no it just can't know all the file systems. it could even be dangerous if a revision would change something in a filesystem that would be valid for the os but invalid for the ssd interpretation.

no, the gc can't work like in java, as the filesystem is nothing fixed and well defined. the gc can just try to collect stuff that was written in a similar time period to flash-sections (like defragmentation), so it hopes it can then clear stuff faster the moment it needs to. mostly like when your room really isn't cleaned up but you're awaiting your new laptop, and you just push the stuff around until you have some space to then take the new laptop out the box. there's no real cleaning going on. just re-messing for faster emergency-cleaing when something new comes (and hopefully erases some of the mess at the same time for the next emergency cleaning).

and no, trim should NEVER harm your performance. as the gc can still work just as well. it now just gets MORE information MORE early so if it then works WORSE, it's a REALLY bad designed gc.

 

and btw, on an ssd that reads file systems, a windows home server v1 could not be installed as it cheats on the filesystem. the gc would completely crap out when the 64gb ssd suddenly has 5.5tb of valid data on it. it's all within controlled environment (called Drive Extender), but believe me, a gc would kill all your data randomly there.

 

I'm out from this discussion, just don't know too much to argue about it.

But one thing that I know is that my SSD works better without TRIM. Maybe it's because of southbridge or something is broken, but TRIM is no go for me right now.

 

how do you actually enable or disable it?

 

Code:

fsutil behavior set disabledeletenotify *

0 == enable
1 == disable

 

okay. i vaguely remember that. it's strange but surely a "speciality" of the masterdrive. anyways, as long as it works for you, all is fine.

 

Does anyone have any kind of prediction when intel's G4 could be released?

Planning to buy a new lappy sometime this/next year or something and just wondering.

 

G3's were just released 2 or 3 weeks ago.

nothing has been said about G4s.

 

Seems like Intel using their Tick-Tock model for SSDs too. Going by that the new generation will have a new controller and probably very good performance.

When will it appear? My guess would be Q2 2012.

There has also been some rumor about Intel using a Sandforce 2200 controller for a new SATA III top model SSD, the Intel 520.

 

I am not sure the tick tock model applies to Intel's SSD department at all. The tick tock model ties closely with the process(fab) but Intel no longer make NAND themselves. And out sourcing the controller to Marvell/SF means it is further away from the silicon. What has left is only their past experience in overall system design(which they can draw experience from CPU/chipset etc. as SSD is a computer by itself) and they brand power(representing reliability etc.).

 

This is a question for the SDD gurus out there.
General advice is to use your SSD as a system drive and program drive only. I read a couple post of people recommending to even move the pagefile over to the regular HDD to safe on the write cycles on the SSD.
Now here is my issue. I am using quite a bit of photoshop and the photoshop pagefile can get quite heavey (I easily get it up to 12GB). Wouldn't it be really beneficial for working support purposes to have this on the SSD, since the read of it and the write would be much faster than on the HDD? But it would obvioulsy shorten the livespan of the drive quite a bit.
Any thought?

 

Well, if You are buying SSD to speed up Your work then only logical conclusion is to leave pagefile on SSD.

 

Why else would you buy a SSD??? Isn't that the whole reason why you go SSD over HDD, to speed up? The difference in power consulmptoin can hardly be worth any of the preimum cost for a SSD.
I am only looking for the speed increase. Focusing on the Crucial C300 256 right now, and might just pull the trigger on it later. I know the Vertex 3 is much faster, but there is one major drawback. It is $170 more.
I saw some crazy speed comparisions though, where w7 e.g. loads twice as fast on the v3 than the c300.
Now if I take into consideration that I keep the pagefiles possibly on the SSD, I will shorten the lifespan to maybe 2 years (totally random number), than there is no reason to pay the premium for the latest drive (it's like buying a new car. The moment you drive off the lot, you loose about 15% value).

 

Don't worry about lifespan. If you can use it to that extend, it means it worth every dollar you have spent. SSD is here to serve you, not the other way round.

 

Your SSD should easily live >5 years unless you write extreme amounts


Source:
Q&A: Tom's Hardware And Kingston On SSD Technology : Kingston Up Close

 

Thanks guys. That really helps. I have never had a HDD for longer than 5 years mounted inside a laptop. Been replacing them on average every 3 years to increase storage capacity (probably have about 10 ATA drives laying around, anywhere from 30GB to 320GB ).
The mother of all questions for me still is, do I want to pay extra for the V3? If they had it as a 160GB drive I could probably compromise and go for the V3, but the 120GB is too small and the V3 is almost half as much as my brand new notebook. Seems weird to me to spend that much. The difference is only 47 cents a day over one year . You guys have been using SSDs for quite some time as I can see, and all those benchmark test tell you more than they tell me.
So what is your honest opninion? Price vs speed? I did opt out off the 2870 on my notebook, just becuase it was $350 more than the 2630, so in comparison, I should probably chose the C300... But I am also reading quite a bit about the C300 being an 'ancient' product already. (obviously the 2630 isn't ancient).
HARDCORE ADVICE NEEDED.
Thanks.

Btw. cheers out to everybody providing his great knowledge on NBR. It already saved me from sending back a perfectly working Envy 17 to HP.


EDIT. Prices for the SSDs (as being available and in stock):
Vertex 3 $553
Crucial C300 $374

 

I am not sure if tiller's case is similar to yours as SF(at least Vertex 2 etc.) doesn't seem to fit his work pattern.

In that case, I would say C300 has better bang for the bucks.

 

I say c300 because the real life differences will be small.