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

but you don't need to buy it as OEM afaik
check this out:
http://discountechnology.com/Samsung-SSD-IDE-Hard-Drive-MCCOE64GEMPP-01A
is this oem or what

 

so now i got your missconception!!!

trim does NOT do zero fills. trim is telling the ssd or hdd that something got deleted. so far, NO HDD EVER GOT THAT INFORMATION!!
NEVER

that is what trim is for: to inform a disk that something is unused anymore.

this, till now, NEVER got reported. a disk could only find out the moment you overwrote some old data.

no, a delete would NEVER propagate to a hdd or ssd before trim. NEVER.

now you get it?

 

Who in their right mind would p i s s away that kind a cheddar on a Samsung SLC drive when for the same money you can get an Intel X25 - E?

 

well, people want to buy ssds that they can trust in, even in case of failure. samsung is nobody to trust much, so it's one reason to suggest people to look for another ssd.

and yes, you can buy samsung ssds directly. it's just, they don't actually want that. and they stated that. they want to sell to vendors that resell. so it's then their fault, if something goes wrong. and yes, they stated it about like that. i've seen it on the web, and i've heard it on the phone when asking for an ssd that was not available in europe back then.

 

YES, it says brand new OEM, means no warranty from Samsung. I was told that Samsung never sell samsung branded SSD in the U.S market.
In addition, this 64G SLC IDE SSD is ty pricy, I wont buy it unless i am stupid. I can buy OCZ vertex EX, Agility EX and Intel-E for the same price, and get better performance.

 

Level One:
Doesn't matter what data is inside the actual nand (a '0' or a '1') - it's whether the nand controller sees it as 'data filled' or 'empty' that determines if an SSD is 'degraded' or not.

Level Two:
TRIM does not 'fill deleted 1's into 0's'. TRIM performs an erase command to the nand chips where the deleted file(s) resided. This Erase command cannot access individual nand chips - it Erases whole blocks (128KB, 256KB or 512KB blocks) at a single time. To do this, it may need to move needed (non-deleted) data to another location so as to free up the nand cells that do contain non-needed (deleted) data.

Level Three:
GC only sees nand as used (data filled) or clean (not data filled). It cannot discern whether any used nand chips are data we currently need (data we haven't yet deleted) or whether is is data that we don't need (because we have deleted/moved it). There is no way for it to know this.

Level Four:
Mechanical HD's do not need trim because they do not incur a 'write' penalty when writing over 'used' areas. They simply need to write what is needed - not erase first and then write the required data. How this is accomplished though is that 'deleting' on a mechanical HD just erases the directory entry - not the data. This keeps deleting 'fast' and also with no downside because the next write does not slow down with regards to a 'clean' or 'used' area of the drive - it is simply related to the mechanical prowess of the HD in question. Actually, as a bonus, this is why 'undeleting' software works with a mechanical HD - this software can 'see' the un-erased data even without the directory entries pointing to it.

Level Five:
Because SSD's cannot erase individual cells, but must erase in blocks, this makes the SSD incur a write penalty. When all the cells are used an SSD must erase first, before it can write new data. This is what makes the SSD 'degraded' because it is doing a 'read/erase/write' for each new piece of data it needs to write... instead of a single/simple 'write'.


It is easy for me to see why you can degrade an SSD further; the controller can't handle it! When a controller is using cache to reduce stuttering by caching the DATA it is trying to handle - it is easy to see that unless you have unlimited cache - you can eventually throw enough data at it to make it cry.

With TRIM - an O/S integration is not only needed - it is unthinkable how you can do without it.

I make a file. I delete a file. I move a file. The only thing that an SSD 'knows' for a fact is that the file was 'touched' or modified in each example. It has no way to know what I did with the file, nor whether it should keep it, clean the nand pages it was on or anything else except to regard it as data the user possibly wants. Once created, every file is considered in the same light - keep it!

With TRIM, with the second and third examples of 'delete' and 'move'. The SSD also gets the 'clean the nand cells associated with this file' command. Very unlike GC which does nothing of the sort - because it can't.

GC, on the other hand, goes something like this:

The clean cells I have are not arranged optimally. What is optimum? Contiguous cells. What can I do to make them optimum? Take all the partially full cells and arrange them so that they are contiguous.

By doing this, not only is it wasting time/power and cell life needlessly. It is also simply 'defragging' itself so that hopefully the next writes are sequential (therefore faster) than if it hadn't performed GC at all.

Hope this helps you understand this better?

Really, if you have more questions, please put some effort into reading Anand's articles critically - even if you think he slighted Samsung users - he has the best current information I've read so far on SSD's. Especially on the points you raise.

I find it really funny that you don't want O/S initiated TRIM - what's the problem? GC has one up side and many potential downsides, TRIM enabled SSD's, once it's prerequisite for an O/S that supports it is given, have no downsides at all.

Your stance, is therefore illogical - as is your attacking Anand too.

 

actually, all trim does, is saying region A - B is not used anymore.

what the ssd then does (erasing cell blocks, remapping cells, what ever it wants) is on it's own interpretation.

a non-trim-supporting ssd just doesn't care about the message. like hdds. they don't need to care, as they never "free" regions of their disk. they just overwrite the garbage.

 

what do you guys think about checking the compress this drive to save space box? what do you think tiller?

 

The tests I've done before on compressed system drives did not offer a satisfying experience. However, that was long before Core 2 Duo's and SSD's too.

Just thinking out loud, it should offer a benefit with a high-end Quad Core CPU and a fast SSD - but then again, when has theory ever caught up with practice?

The advantages should be possibly faster boot up times (less MB/s to read), but the disadvantages are not too clear to me - least of all the possibility of an unwanted interaction with the compression algorithm, a specific SSD firmware and/or a specific SATA driver interaction that takes these SSD's over the edge (again).

I'm sure there are others more qualified to speak on this issue, but those are my thoughts just before I sign off!

BTW, Thanks for wanting to know my thoughts on this.

 

Do not compress. There is no need. I have NEVER yet seen good things come of a compressed drive.

 

Okay:

http://www.tomshardware.com/reviews/windows-ssd-performance,2518.html


I know, Tom's again. But, this shows me that SSD's are not here yet.

My standard config desktops I run 4 Raptors (VRaptor +3 Raptors) so this was a pretty good read for me.

I would love to have this level of performance in my notebook, but for my desktops - I'll skip this generation of SSD storage 'upgrades'.

Why? As every other test fails to mention - this 'performance increase' is only for the first startup after a cold boot. After that - we are running at RAM speeds thanks to SuperFetch.

I wish that Intel had some real competition so that we can buy the G5's tomorrow, instead of sometime in 2013.

And, let's not start on Tom's testing procedures!

They're expecting TRIM to work with ancient drivers (IMSM 8.9).

Sigh...

 

not really. performance increase is just 100% all day long. unlike any form of cache including superfetch. which helps much, but not always.

and i would happily replace all your raptors, as they're big, noisy, and still slow compared to ssds big way.


about compression: on a desktop, one might consider it. on a notebook, never. why? as every access now needs cpu cycles to read/write, it increase the chance that the cpu can't idle, draining battery.

but it's all, most likely, minimal.

 

I tested compression for a few minutes ran the usual suspects benchmarks. Sequential reads were showing 1300mb/s and sequential writes 38mb/s. I think it might be how the compression is screwing up the tests. Winsat disk was the same. Took about the same time to uncompress as to compress and the drive was at 23% fragmented after uncompression.

 

Is there anything really special about intel's storage drivers over the default Microsoft window's driver?

 

Under Vista potentially better performance

In Win7 - at the Moment the MS drivers give you Trim, the Intel ones don't.

 

ermmm
NO
when you hit delete, windows delete pointer to certain memory space allocated on hdd, or ssd...
data on memory space still remains where it was before, only you don't have any pointers to that place, so, windows will see it as a blank space...
but, memory space can't be considered as a glass for water, like, when memory is used, glass is full, and when memory is empty, glass is empty
ALL memory, nevermind if it is magnetic plate or nand WILL contain any kind of data, regardless to what windows are point to that specific memory space, empty or used space
so, you see, you can't actually delete data, it won't vanish, it always need to contain some logical state, low (zeros) or high (ones)
spiners will have 0's since zeros, and ssd, will be filled with high state, since nand can't be rewritten if it uses regular states, it needs to use opposite form what magnetic plates are using cause of the way flash memory works...

and this is not the way I think stuff works, it is the way I actually KNOW stuff works cause I have certain experience in low level micro-controller and processor programing...you know, z80, atmel stuff and so on....

same goes for tileroftheearth
data cant go away, it only can be overwritten, so, all your levels, built from level one, and level 1 is wrong, thus, all following is wrong too...

 

darQ96, I suggest that you read up on how SSDs work. They don't work like Z80 procssors or Atmel ASICs. davepermen is quite correct that the trim command itself does not write anything. He is also correct in that a simple (non-trimmed) delete is insufficient for the SSD to know that those blocks are unused. A delete merely updates the filesystem's MFT (for NTFS, or equivalent for other file systems) so that the filesystem knows that the block is free. The drive doesn't know how to interpret the filesystem -- it works at a much lower level -- so it does not know that those block(s) are now unused. Thus, the trim command was created as a filesystem-agnostic way to tell the drive when a block no longer holds filesystem-relevant data. This will work any filesystem -- NTFS, FAT, EXT3, HFS, etc -- without requiring that the drive include a dedicated CPU and custom software to understand every known filesystem type.

 

Not due out till March but Anand says...
OCZ's Vertex 2 Pro Preview: The Fastest MLC SSD We've Ever Tested

http://www.anandtech.com/storage/showdoc.aspx?i=3702

 

skriefal,

Thanks for your input, hope darQ96 listens to you, if not to davepermen and I.


darQ96,

Unless you have the schematics from Intel, Samsung, Indilinx and every other SSD manufacturer's controllers, you do not know what the 'clean' state is of their nand. You only know what each controller is reporting to the O/S, when queried.

There is such a thing as a 'clean' state: this is the state where the nand and the SSD perform at their fastest possible. This is because the SSD is not wasting precious CPU cycles (it's own CPU) keeping track of the used and non-used states of each nand cell.

Furthermore, my Level 1/2/3/4/5 statements do not rely on each other. They are mutually exclusive. Except for 2 and 5.

You may not agree that the controller determines the role the nand chips play (my Level 1), but we cannot agree that the nand chips are write-able individually on a bit by bit level. This stance you've taken is disproved by any attempt at researching this area. Facts: We can write to a 4KB cluster, but we can only erase much larger blocks (512KB being a 'norm' right now). This is the very basis of where the 'write-penalty' comes from.


I don't think we're adding anything to this topic; except showing your need of further educating yourself on this topic. Hope you do and come back here to educate us further too.


davepermen,

just want to acknowledge that your correction to my original post was correct (of course!), just thought I mentioned it earlier, but I didn't.

 

man, I feel like I am in english class again, too much reading here.

 

Young man! You spoke out of turn, now put your gum on your nose and stand in the corner until you're called again!

 

And write on the blackboard 100 times... I love SSD's I love SSD's ....... Happy New Year to all!

 

"Intel tells me it will be a "mid-year" refresh...we'll see 160GB down at $225, 600GB at $500+ and the 300GB drive somewhere in between. "

happy new year and happy new price.

 

that's just a speculation. i would be shocked to see 600gb at 500. sounds too soon when 256gb drives are still selling for 700.

 

Thanks sgilmore62!

Reading this article I see that my hope (a few posts back) seem to be answered. At least a little.

I asked for a 'real' competitor to Intel and SandForce seems to be stepping up to the plate.

However, after I wiped the silly grin from my face having finished the article...

Intel is still the 'preferred' choice for many reasons. Not the least is the trust it has from me, at least.

What really struck me is that Intel's 'E' series SSD - over a year old now, still gives this new controller a run for it's money. And at a fraction of the capacity too (we know that the bigger the SSD the faster it can perform).

For a whole 3% improvement in PCMark Vantage - Overall for the SandForce controller, aka Vertex 2, this is non-news to me right now. The silly grin is entirely erased and forgotten, by now.

The hidden news that I see is that Intel is still flying high above anything available from anyone else in the world - with 'old' tech. I just hope they get prodded to introduce their 'new' tech soon.

On another note, this ties in nicely with sgilmore62's comments on compression. The SandForce controller is doing some crazy things with compression and that is what I'm really scared of, with this new controller.

Talk about proprietary - with no responsibility to go with it! (SandForce, like Indilinx just make the controller - the SSD makers using this controller are supposed to investigate and totally understand for us that our data is safe...

Oh oh! This is an OCZ Vertex 2... guess I'll be considering this line, if they're still available - in 2015 or so (btw, how many SSD 'models' does OCZ have claim to?).

 

If they don't do something 'shocking' (like they did with the G2's) then they have to consider the possibility of stalling the adaption of SSD's on the scale that they dream of.

 

OCZ has discontinued the Samsung controller Summit series SSD's. The rest of the OCZ lineup is all Indilinx based until the Sandforce drives are released.

 

sgilmore62,

That is actually more of a problem than it seems to solve.

Samsung distances itself from its products and forces resellers like OCZ to assume responsibility. Now, OCZ is distancing itself from these products that are still available to be bought.

Both sellers are 'in the clear' as far as they are concerned, but who gets left out in the cold? Consumers, of course.

And that, is the 'gotcha' in offering everything under the sun just to put some coin in your pocket, instead of offering and pursuing products that offer real and long term value to your customers.

Not picking on OCZ specifically here; many companies follow the same path. Just that my needs are not served when price is the only or major consideration (which many companies, including OCZ, seem to think is our only determining criteria, when considering 'new tech').

 

Dont worry about SanFroce,
Marvell new controller will be out soon, and intel gonna make its new product line.

 

corsair seems to be in conservative side, its tech guru ramguy discourages buyers in corsair forum to take its indilinx based x-series as this line is more problematic based on failure rate.

 

this is only true for hdds. and this is why the region the memory pointer pointed to will not get reported as "free space" to the disk.

but flash cells have two states: containing data, or not containing data. a free cell can easily be written to (filled with data). a non-free cell first have to be freed (be it 0es, be it 1es.. doesn't matter), to be able to fill it with fresh data.

and that freeing is the slowest part of flash. so if you can do it BEFORE writing to it, that's a gain.


second point:

by knowing which regions are free for the ssd, it can use those free cells for new data more quickly, as i thas more free cells to spread writes out to.



trim should have been implemented in any os. due to the fact that hdds never 'cared', as they don't have to, os optimized this away and forgot to inform drives about that.

now they are in sync again: both ssd/hdd and os always have the same amount of useful and random/free data.



happy new year.

 

i'm so glad i got my second intel 80gb gen2 ssd today (i had to get another one as this one, the order, couldn't get delivered till christmas, and i needed it by then).

so it got all my love before i knew there's something better out there.. happy new year, it'll be fun. is this the year of ssds? for me, that was 2008.. and i'm still in love

 

and that from you.. perfect new years gift happy new year! may ssds fullfill all your dreams this year, as they failed so hard the last one.

wanna try an intel gen2 80gb? i have one too much right now

 

with current great value of CHF, it's much cheaper to buy computer stuff in switzerland, i remember i was ripped off over four grands of CHF for a inspiron 3800 nine years ago when i was there. Prosit Neujahr!

 

davepermen,

I again graciously decline your offer, first because the 80GB G2 would be too small for me to do a real test with and second, because I could not repay your generosity to my satisfaction, ever.

Happy New Year to One and All!

Best Wishes and good health to each and everyone here that I've made such good friends with. Tirelessly exchanging our knowledge freely and so easily helping each other like good friends should.

May each of your families and all your friends near and far prosper and achieve all their goals and dreams in the coming year.

Cheers!

And, Good Night... Peace on Earth... Goodwill to All Mankind.

 

So I have an option, for about the same price, for one of the following:
- Intel x25-m G2 160GB; or
- Kingston SSDNow 64GB Intel x25-e

Capacity is not an issue. I have external drives and would only use for operating system, programs and a couple of games.

I understand that the x25-e uses SLC rather than MLC memory, so should be a lot faster. However, I am not sure how TRIM affects this (whether applies to the x25-e also) by comparison, and which one would be the better choice. I can get both for about the same amount (the x25-e actually a little cheaper), so was just wanting some opinions...

Are there any whispers of new Intel SLC drives soon?

 

I think TRIM is mainly for MLC drives, but don't quote me on that.

If it were me with that choice I would go for the X25-M without a second thought, but that's because I load the OS and some games (mostly the multiplayer ones) on the SSD. If storage isn't an issue with you that stick with the SLC.

(Very happy with my Gen2 X25-M 160GB, by the way. $400 isn't too shabby, either)

 

are you sure it's x25-e? it should be over $700...
http://www.dealtime.com/xPO-Kingsto...ries-Drive-Intel-X25-E-SATA-Solid-State-Drive

 

of course ssd-s don't work like micro-controllers, I didn't said that, I've said that I have programed those to use flash memory
and again, you are wrong about deleting, if drive don't know what memory location is not used by the user or OS, then data would be written on random memory locations and non deleted data would be overwritten, and ssd could not work correctly without trim
but, as you know, ssd-s can work without trim, they only get slower

 

yeah, pricing worldwide can be quite messy right now. if you want to have fun with strange prices, check out the thing i want to buy currently: a jazzmutant lemur.

it's available from around 1500$ to 3500$. i currently try to find out how to get it for cheap to me in switzerland you're from where? germany?

and why does it work? as the ssd gets informed that data is deleted at the moment the os writes to the same location again. at this moment, it can free the cell. the problem is, freeing cells is slow, so informing the ssd more early that they could be erased again in preparation helps for stable performance.

is it really that hard for you to understand it?

how sweet well, good morning into the new decade

well, i guess i'll take the 80gb into my main desktop, then, and offer my old mtron 64gb ones to my dad, both in raid0 give him 128gb, slc, 200MB/s read/write. that should help him to work with hd-movies when ever a disk is bottleneck.

 

again, there is no clean state, on slc-s there is low and high state, ONLY 0 and 1, NOTHING ELSE...
on mlc there are 4 states, 0,1,2&3, and none of those 4 is used for marking cell as empty
I actually know something how flash memory tranzistors work, and, as I said before, nand memory couldn't change it state more than once if first state would be low

no there is not, try to google out some wiki stuff about slc or mlc, as I said above, slc uses only 2 states and mlc usually 4

2 and 5 are levels that you are actually write on most things

individual cells in nand are able to change their states independently, but, controller is the one that is limited to minimum write size and minimum delete size ( cluster for write, page for delete )
so, that's true, and that, well, basically is what slows writes, but it is not because the cell is used and it contains data, it is because of wear leveling algoritam that needs to use all cells evenly

well, I kind a know what I'm typing here, and, believe it or not, flash nand was used much-much more before ssd era, and, I kind a know some stuff about flash, and, just to back up my writing, here are some basic stuff form wiki about slc & mlc
http://en.wikipedia.org/wiki/Single-level_cell
http://en.wikipedia.org/wiki/Multi-level_cell

 

you should STOP thinking about the bits. they don't matter. flash disks work in cells. clean state is if all cells are marked free to simply write to them. fully dirty would be each cell has to first be cleared.

really, dude.. that stuff isn't hard.


and in the end it all doesn't matter. what we want is the os to report exactly which region is used (by sending the actual data to disk) and which isn't.

it's then the ssds job to handle that, how ever it wants. how it does, doesn't matter. all that matters is that the os gives that additional info, so the ssd can do it's job better.

TRIM is just that info. trim doesn't force any deletion or anything. it just informs "these memory regions are free", and the ssd can then do what it likes.


what is 1 and what is 0 is a question on definition. there's just one thing important: "filling" and "emptying" a cell works differently (and in different sizes), and thus has a meaning. on a hdd, "filling" and "emptying" works identical, and thus has NO meaning.

 

and again, there is no no-containing-data state
remember that

rest of the stuff is ok, but you must not forget wear leveling, cause, when you have write that is smaller than a page, and page is the smallest amount of memory that controller can delete form flash, then few blocks form that page need to be written on new location + rest of the new data...

 

bla bla. my stuff is right, you just don't "interpret it right". i'm sorry that you have such a hard time understanding it.


and yes, there is a "containing data" state. not on the cell, but the controler knows which cells are written to (the one you wrote to from your os). and with trim, you inform the ssd which do "not contain data anymore", so it can drop those cells out of the "don't touch" pool.

really hard. i wait for the coin to drop for you...

 

looks it had a promotion just ended yesterday...nein, ich bin nicht aus Deutschland, sondern aus Nordamerika.

 

yeah, promotion wasn't available for me

oh, fun. but you still learned german? or have german roots?

now i would say buy it for me, and send it over, but i know that it most likely still cost more than buying much more around here.. stupid duty taxes stuff..

 

ja, i'm paying 13% which is even higher than your 7.6% VAT...

 

yeah, i would be able to get a bit of money back. but the problem is, getting it from america to here would cost me around 1000$ or more on the borderline.

we get warned everytime when stuff is cheap in america. people go over there and can buy laptops for half the price it is here. but don't try to go on the plane back.. you will pay more taxes and stuff to get that laptop onto the plain, exporting it from the states, than it would have cost here, actually.

i don't get the why part, but i don't care. i just have now a complicated life to find out how to get that thing reasonably cheap...

 

for ssds, btw, the fun is, nobody cares. they think they're "just some usb stick style thing", and believe if it's written on it that it costs 10$ so no taxes nothing. they don't get that the whole box was, back then, around 1000$ worth of money

 

the price for 160G G2 here is quite reasonable: http://www.ssdshop.ch/cat120.htm