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

The two conclusions on the brackets contradict each other.

469 x 4K = 1.876MB/s

The Intel drive gets 30-40MB/s which makes it 7500-10,000 I/O per sec.

 

Hello everyone, got a few questions to ask before I purchase my own SSD drive. I'm looking at the Samsung MLC 128GB SSD and will buy if off eBay (those listed as New and pulled from systems).

I've read conflicting feedback that some SSDs run hotter than mechanical drives. Is there any truth to this? And if I partition my SSD, are there any ill-effects (wear leveling etc.)?

 

DSEO80 I hope you are right. I remember when i worked at Egghead Software and the software of the year was something called STACKER. It compressed your drive. Great when it worked, but man, when it did not work, for get about it. Burnt toast. I guess I cannot help thinking that MLC was done for price only with no regard for reliability. And by reliability, I do not mean amount of writes per se, I mean the ability to keep my data safe and happy on any given day. I was one of the first to refuse to sell STACKER even thought there was a 10 dollar SPIFF with it because of what i saw it do to some drives. I know that technically MLC is not compressing the bits, but it is doing something to double the capacity and since they have not been around that long, I cannot help but wonder. I guess in a way it is a moot point, as I will always have a good backup nearby. And I will probably go for cost benefit over SLC benefit.

Can you pint me to a 128GB Samsung that has no stuttering issues and great performance? Thanks, Dave

 

I have to admit MLC SSD drives have not been around long enough to judge their performance. However, MLC type NAND Flash chips have been around long enough that safety of your data shouldn't be too much of a concern. (Again chip level and hardware level failure rates are not the same, but you get the point..ALSO you don't write to flash drives as much as to a hard drive so SSDs will potentially be stressed much more than flash)
Several companies have tried to market MLC SSDs of poor quality, these companies were never NAND flash manufacturers, SSD manufacturers OR HDD manufacturers (OCZ, Patriot, etc..) and i think this is the reason that more complicated behavior of MLC drives caused issues. (ie. slapping together 10 nand flash chips does not make an ssd, as i guess we've learned).
The reason I push consumers towards Intel or Samsung drives for SSD are both companies have experience in actually manufacturing NAND flash chips. Also Samsung has been manufacturing HDDs for several years (although the buy most of the parts from US/Japanese companies). And Intel... is Intel..

The 128GB Samsung MLC i can vouch for because I have been using it for the past 4 months is the older model with 90/70 (read/write) performance. (I own two of the drives, and both have never caused performance issues)
The model number is MMDOE28G5MPP-OVA. I cannot 100% vouch for the newer 128GB model with 220/200 (read/write) as i do not own it. However that is the drive i am looking for as a replacement drive (although I may just wait another generation).

It seems that it would offer better performance than the Intel mainstream drives at lower cost. My experience has been that after a sufficient number for IO/s to have no stuttering (samsung, intel mlc drives) the read/write throughput of the drive becomes important in determining performance. (document loading, boot up times, file copies)

If anyone receives any of the newer gen samsung drives (256GB or 128GB models) a benchmark of 4k random writes would be much appreciated. My guess is that although they were able to substantially increase seq read/writes for large files, the gain is much smaller for smaller files since my understanding is they are using something similar to RAID.

Good luck on the search for your perfect SSD.

 

yes, sort of:

http://www.ocztechnologyforum.com/forum/showthread.php?t=48309

 

hm.. slc as well. still, the first bigger company that tried made it right from the start. else, the hype wouldn't have started at all. mtron and memoright both delivered..

it looks like, from the rest of your post, that we now have good stable mlc, too.[/QUOTE]

 

"ALSO you don't write to flash drives as much as to a hard drive so SSDs will potentially be stressed much more than flash)" Boy, that is for sure. I put stuff on my thumb drive and pretty much leave it there forever

Still, we think alike in that these bad MLC's have really taught us a lesson. I am thinking a 128GB of Samsung in maybe the next release. I would also be interested in more info about the most current release, referred to I think as the 220/200 Next question, WHY is SLC so expensive? I mean heck, EDO was expensive when it first came out and then went down to pennies per GB I also agre with daveperman that Mtron and Memoright did us all a big favor by getting the ball rolling. I would think the 64GB 3500 from Mtron would be getting more press, but I guess also that 128GB is trying hard to be the sweet spot.

 

slc is 4 times as expensive as mlc i thought (it has 4 times less storage with same transistor count, or something). or 2 times?

well, if it's 2 times, then you can compare current mlc and slc solutions and realise, that's more or less what we can get. 128gb mlc for around 400, as well as 64gb slc for around 400.

i'd love to see bigger slc solutions. i would pay 800 for 128gb of slc easily from what i've seen so far. but sadly, for notebooks, mtron doesn't deliver more than 64gb for the mobi versions (and no 64gb for the 1.8" version, which i'd love to have, too).

 

There is no MLC with 4 bits per cell at the moment that's in production. They all use 2 bits per cell. The problem with reliability and write performance might get worse as bits per cell increases in the future.

 

would still be cool 8 bits per cell, but slow to use (but huge in storage and quite cheap).



can't wait for memristors *dreaming*.

 

G.Skill Titan working fine on Vista with Boot Camp on Late 2008 Macbook Pro.

 

Yeah, there has definitely been a "shortage" of 128GB SLC, for sure It would be nice to have one as an option that was not $2500 or so. Sometimes all this fast change has that hurry up and wait feeling to it. Although I do understand that there are some, or maybe MANY out there who are happily using a SSD and enjoying it. I dipped my toe in trice. Once with a 16GB Mtron 3000 MOBI for a week or so till I needed to sell it on Ebay, and once with a $159 64GB G.Skill MLC early edition which simply was to erratic to use, so it got shipped back and I got the 15% restocking FEE. I did not use the 16GB as it was in my laptop and I just did not use it much. I did, however, learn that once could easily load VISTA Business AND Office 2007 and have plenty of room left over with a little bit of tweaking recycle bin, etc. THAT was an eye opener.

 

I'm astonished at the prices of 128GB SLC drives, considering there are now MLC drives with greater capacity and better performance than large capacity SLC drives, for a considerably lower price.

 

hey anyone know if this is any oood?
http://www.newegg.ca/Product/Product.aspx?Item=N82E16820231220&Tpk=FM-25S2S-64GB
G.Skill 64GB

I will use SteadyState and any other tweaks out there. by the way is this just a rebranded OCZ core?

let me know your thoughts

 

How you measure better performance? Read and write speeds? Those are only one metric. SLC drives still have better latency, proven reliability, and you won't find most if not all of the issues that plague MLC drives such as stuttering, higher temperature, higher power consumption, poor controller, or poor I/O performance, to name a few. SLC has been used for a long time now in mission critical settings, the only reason why performance is low is due to lack of competition in the past from the cost.

 

None of those are an issue with newer MLC drives, for instance the new Samsung ones that are popping up in Dells. 128GB and 256GB capacities, great performance (200MB read, 160MB write), low temps and power consumption, and no issues with stuttering. I'll grant you the latency, but I'll take the better read/write speeds over the better latency any day.

 

so are samsung 256gb at (200MB read, 160MB write) available by dell..?

 

Not in all models yet: http://www.engadget.com/2009/01/17/dell-adds-256gb-ssd-option-to-xps-m1330-and-m1730-laptops/

 

It's the latency that allows SSDs to be more responsive than HDDs. That's the reason Turbo Memory can be useful despite the low throughput of the flash used for them.

 

I was referring to the difference in latency between MLC and SLC drives. IE, a 128GB Memoright GT vs. a 128GB Samsung.

 

The price of semiconductor products is mostly dependent on how many masks are required during processing. (more masks means more steps, which means more time and resources involved in production)

MLC allows for double the density with the same number of masks as SLC. which is why SLC chips cost roughly twice as much for the same density. If 4 bit/cell chips are developed they will be cost about 1/4 of an SLC chip.

 

And yet HDDs kill them in small reads/writes which should benefit the most from low latency.
The latency FINDING the information is low compared to a HDD but reading and especially writing it is high.

 

Question. Say for example I wanted to install a 1.8 size drive in a laptop meant for 2.5. Would I need some sort of adapter to compensate for the difference in size? I would think it would be less difficult than trying to use a 1.8 ZIF drive, right?

 

Hey folks, I spent a little bit of time making a spreadsheet out of most of the SSD benchmarks from this thread and a few from the HDTune thread. Ultimately I don't think it's terribly useful due to the incompleteness and lack of breadth of the information, but for anyone who's interested:

http://colette.trianglesoft.net/2009/doc/090122a.xls

 

I thought in a few screenshots SLC drives had better read/writes in small files. MLC ones were abysmal though.

 

New G.Skill Titan 128 gb in unibody 13" macbook working incredibly well here...easy install, very fast, and no stuttering.

 

Latency for MLC drives is still about 0.2ms, double that of SLC drives, yet HDD's still have 75x the latency.

For me, a good MLC drive is where it is at. They are available now that are good enough to satisfy me. I can't see myself noticing the 0.1ms difference. The 16ms difference from my HDD to the MLC drive is another story.

Greg

 

No, reading latency is low. Writing is high. Reading is what determines "responsiveness". The Intel drives don't really excel at reading which is why those "tests" about loading time isn't good on the Intel drives. Sure their max transfer read rates are high but their small reads and average reads aren't.

Where writing is important as long as the minimum point is established(which is the "stutering" part) is with file transfers and installations.

 

Exactly!

Personally, all I need is a little more time to see the new OCZ MLCs get released and ascertain the new gen of G.Skill/OCZ MLCs don't suffer from stuttering or lagging, and then I'll likely be pulling the trigger (predicated on 128GB < $300 USD street).

BTW: Did anyone notice the prices went up 10 and 20 bux, respectively, on the G.Skill Titan 128s and 256s sold thru NewEgg? I warned about the Egg a page or two back! (simple supply and demand, really)

 

And that's why I think we have to get used to thinking in a different frame of reference for some of these parameters. We're working with a whole different medium and technology here, vs mechanical HDDs, and need to see the whole picture. Some parameters may not translate perfectly, nor accurately describe performance in terms we're used to using.

What good is a 100 or 200nS latency spec if we're limited to 50 or 100 IOPS in small random writes? According to my math, 50 IOPS = 20 mS, and 100 IOPS = 10 mS. IIRC, the early OCZ MLCs had some really horrid small random write IOPS of something like 6 or 7! Do the math on those numbers!

Someone correct me if I'm wrong, but doesn't a previously full used block need to be "wiped" or "prepped" in some manner before the new data can actually be written? If so, then there's spec we've not yet seen with HDDs that we now need to take into consideration.

I guess what I'm saying is: We may not easily be able to always easily translate our previously used HDD's parameters and metrics to SSDs, and still tell the whole story as far as real world SSD performance is concerned.

 

Stratcat:

Some of your slow write issues have to be balanced against the amount of writes done at the level that is relavent to finding the problem.

 

Well, I pulled the trigger on the G.Skill. Arrives next week. Hopefully, it'll run as advertised.

 

Same here. When it was on auto-notify it was $489.99. When I ordered it it was $499.99. When I last looked, it was $519.99.

Shipping to LA from New Jersey, this is the first time I've ordered something from them that didn't arrive the next day. Did they close their LA warehouse or something?

 

I ordered something from them not too long ago, it came from LA.

 

Does anyone here know how to check my computer SATA speed? I'm thinking of getting G.Skills but I'm not sure if I am running SATA 1 or SATA 2.

I want to check both my laptop. SZ360 and TZ91. Thanks

 

You can use PCwizard.

 

Hmm it says I have

Intel(R) 82801GBM/GHM (ICH7-M Family) Serial ATA Storage Controller - 27C4

I assume thats SATA1? Oh man if that is SATA1, I don't think G.Skills 128 is a good idea. :'(

 

If you have a program called Intel Matrix Storage Manager installed you can check the speeds.

Otherwise the only way is by checking with your manufacturer and/or Intel's datasheets.

ICH7-M is supposed to be SATA2-300 compatible. When it says its an "SATA controller" it doesn't mean its SATA only, its just a generic term meaning it could be SATA first gen, or second gen, but regardless its an SATA controller.

Yes you are absolutely right. OCZ Core drives will actually have reasonable IOPS until you fully write to the drive the first time. This is I assume why some people don't notice the so-called "stutter" while others do. After the writes are done and they need to "clean" up the written blocks for new data it slows down. You can't do a 5 min benchmark on a fresh SSD drive and expect that the performance will remain. SSD drives will take few weeks of usage until the performance drops to a value which will stay stable. All the reviews out on the web are really inflating the SSD drives more.

This is why on the X25-E datasheet it shows 4k random write IOPS as 3.3k(13.3MB/s) while people with the slower X25-M report 8,000-10,000 IOPS(30-40MB/s). The X25-E datasheet is showing steady state value while X25-M is reported with the drive fresh.

 

I'm pretty sure that the SZ3 is SATA1.

I wouldn't consider that as a reason not to buy an SSD. You can still get 150mbps. In most cirsumstances, the G.Skill Titan will not go higher than that.

SATA2 doesn't make the drive twice as fast, it just increases the maximum speed so long as the drive can go that fast.

Greg

 

Yes. Couldn't agree more.
But I think it makes the random read/write performance all the more important.

 

SATA-150 does not reach its theoretical 150MB/s. The real limit is 110-120MB/s. There is overhead which prevents it from reaching theoretical maximum. You can only reach that with RAID controllers that have dedicated processors on them. There is no point getting the new SSDs if the interface does not fully support your SSD. Just stick with your OCZ drives and optimize it.

 

I'm hearing reports of stuttering on the G.Skill Titans both 128GB and 256GB. Some users reporting issues with the drives stalling for several seconds when using Firefox.

 

heard that too.
jmicron simply is crap.

 

I can hardly stand it. No no Joe, say it ain't so! Not the TITANS too???? Can anyone say, "Samsung SLC" ???

 

I haven't experienced any stuttering on the Titan 256 GB that I own.

 

I'll be getting mine in the next couple of days. I'll report back on what I find.

 

"You get what you pay for"

The Vertex and the Titan drives are pretty cheap drives. Technology advances fast but advances predictably fast. The breakthroughs only exist to advance the technology that might otherwise not advance.

Now can people report the actual reported sizes of the drives please?? Does the 128GB actually show in My Computer or is it showing a little less than that like with most hard drives?

 

Ehh?

I thought the OCZ Vertex series did NOT use a JMicron controller... none have been seen or reviewed in the wild that I'm aware of, so I'm not sure why you lump it in with the G.Skill Titan... ???

 

Sigh. Read the couple of posts back.

Now can people with the new drives answer this?? Sure it says the size is 128GB. Does it actually report 128GB or is it "128GB" in standard hard drive terms(meaning its actually less)

I don't like the Anandtech review too much. Sure it shows the problems with the SSD drives. But that's it. It's too simple for an SSD review. Just like the file systems and the OSes are made for HDDs, the reviews are also tailored for HDDs.

The biggest problem is not the JMicron controller. It's that the JMicron controller is used for cheap SSD drives. What OCZ did was take bunch of flash chips(Samsung), some controller(JMicron) and assemble it. It's very likely the same JMicron controller used on the OCZ Core drives can be tailored for better performing SSD drive. OCZ/G.Skill didn't choose to do it to cut down the price.

Of course that doesn't mean that the particular JMicron controller is a fast controller. It probably isn't. But its made worse by the fact its used on a cheap SSD drive configuration.

 

Agreed.

RL general computing use involves varied and most likely more balanced stressing of the various operating parameters, and performance should be viewed in the terms of the totality of the overall 'big picture'. Agreed. Pure bench tests can only guide. But, at least in the case of the 1st gen MLC/Jmicron SSDs, quite a few early adopters have gotten stung by some fairly basic scenarios, such as loading the OS or doing file copies, or even using MS Outlook or IMing. Of course, we rarely hear from those that have had no issues.

It appears, at least as far as the previous gen MLCs, there's no free lunch.

Once again, exactly!

And thank you for your further clarification and expansion.

Your comments strike at the crux of my post: HDDs and SSDs are two different animals. They may have the same Form-Fit-Function, but are different technologies. Some parameters may translate well between the two devices, and others may not even exist! I'm sure there's a spec used by the SSD manufacturer's R&D and QA depts for this pre-write clean-up , as you described, even though we, as end users, may not know its name yet. We will likely add it to our lexicon at some point.

But it's a good example, IMHO, of how a purely HDD frame of reference is inadequate to describe or evaluate SSDs completely and thoroughly.

Links or references?

(Just trying to remain objective)