960GB SSD for "under $600"

[Kippa]

I am more weary about ssds since my old SSD died on me. The warranty on it was for 2 years and I had it 2 years and 10 months. It was only a 120gb drive and was used a lot though. A 1TB ssd should last quite a while though if not filled up.

 

[Coup27]

I am more weary about ssds since my old SSD died on me. The warranty on it was for 2 years and I had it 2 years and 10 months. It was only a 120gb drive and was used a lot though. A 1TB ssd should last quite a while though if not filled up.

It's very unlikely that your SSD died because the NAND was exhausted. More likely a controller or firmware failure.

 

[SLM Crew Chief]

. Remember an SSD will only only data for a year or so while mechanical disks are good for many, many years.
.

Really? I thought that was all worked out?! I have a few thumb drives that have held data for 4 or 5 years.

I have a Samsung 840 250g coming for my gaming system and it only gets used every great once in a while when I feel like some Left For Dead 2. Until two days ago it hadn't been booted for well over a year!

 

[Hellhammer]

True. I know they sold some of their stake in IMFT early last year. I'm not too familiar with their current arrangement with Micron. Maybe Kristian can explain?

http://www.anandtech.com/show/5603/intel-sells-some-of-its-imft-stake-to-micron

I don't think anything has changed since that, at least I haven't heard anything. Intel sold their stake at the two fabs about a year ago, but they still own 49% of the Utah plant.

Samsung has estimated 13 year lifespan at 40 GB/day of writes but they could be biased, and so I still don't trust TLC for smaller capacities.

It's not biased, it's simple math. 40GB of writes per day equals to 43,800GB of writes in three years (I think that's the lifespan Samsung was quoting?). Even the 120GB model can withstand at least 128,000GB of writes, so that figure is accurate if we assume a write amplification of 3x (which is quite accurate for client workloads).

 

[Hulk]

Really? I thought that was all worked out?! I have a few thumb drives that have held data for 4 or 5 years.

I have a Samsung 840 250g coming for my gaming system and it only gets used every great once in a while when I feel like some Left For Dead 2. Until two days ago it hadn't been booted for well over a year!

I can't find any good links now but I remember flash drives being "tuned" to hold data longer while and SSD un-powered would only hold data for a year to 5 years?
I'm having a hard time finding good data on this.

 

[Koing]

Bets on when we can get a 3TB SSD for $300?! 5yrs?

Koing

 

[Kippa]

I wouldn't mind an SSD using the the same form factor of a 3.5" of a hard drive if it needs to be bigger to fit more chips instead of it having to be 2.5" that it is at the moment. I'd certaintley consider buying a 3.5" 2TB SSD if it didn't cost that much. Can anyone tell me why SSDs have to be 2.5"? Surely then can do 3.5" and fit much more chips in.

 

[Hulk]

I wouldn't mind an SSD using the the same form factor of a 3.5" of a hard drive if it needs to be bigger to fit more chips instead of it having to be 2.5" that it is at the moment. I'd certaintley consider buying a 3.5" 2TB SSD if it didn't cost that much. Can anyone tell me why SSDs have to be 2.5"? Surely then can do 3.5" and fit much more chips in.

Controllers seem to have "settled" on 8 channels. Right now the NAND package size is 64GB, 8x64 gives 512GB, which is why this is the largest size we've seen so far. The smaller 2.5" form factor is cheaper and fits in both laptops and desktops. Extra room isn't need since they only have 8 channels. We'll see 1TB drives with 128GB NAND. For the record the Intel 320 series used a 10 channel controller.

 

[blastingcap]

It's not biased, it's simple math. 40GB of writes per day equals to 43,800GB of writes in three years (I think that's the lifespan Samsung was quoting?). Even the 120GB model can withstand at least 128,000GB of writes, so that figure is accurate if we assume a write amplification of 3x (which is quite accurate for client workloads).

The part you didn't quote me on was where I wrote that TLC needs to be lower priced because it's less proven technology than MLC. Samsung's estimates could be wrong, despite what they claim, is what I'm getting at. If TLC drives cost a little more than half of what MLC drives cost, I would get a TLC drive twice as big as I would get if it were MLC. But the actual price difference isn't that big. Until TLC falls in price more, I don't think it's worth getting over MLC drives.

 

[groberts101]

As the nand industry transitions over to smaller die and increase TLC production.. eventually you guys won't have much of a choice in the matter.

So, if you want a newer SSD in the next year or two?.. you'll have to take what they package it with. Although I would hope we can keep some higher-K MLC or toggle in production just to appease the enthusiast segment.

Remember when everyone was getting upset and overly concerned when they moved from 10k to 5k nand?

And did that public concern cause the production transition to cease? Nope.

Also keep ijn mind that as the controller firmware becomes more complex and new algorithms are developed.. having 500 write cycles on a 1 - 2 TB SSD will become less of a concern to end users if the price is lowered enough to sell units.

When boiled right down.. we basically just live in a society where speed is easily.. and quite often traded for lifespan when it comes to cars and electronics. We tend to live for the moment, is all.

 

[SLM Crew Chief]

Quote:
Originally Posted by SLM Crew Chief View Post
Really? I thought that was all worked out?! I have a few thumb drives that have held data for 4 or 5 years.

I have a Samsung 840 250g coming for my gaming system and it only gets used every great once in a while when I feel like some Left For Dead 2. Until two days ago it hadn't been booted for well over a year!

I can't find any good links now but I remember flash drives being "tuned" to hold data longer while and SSD un-powered would only hold data for a year to 5 years?
I'm having a hard time finding good data on this.

If you find that can you please post a link? I'd like to read it before I install the new drive. This might be a deal breaker on SSD's for me.

 

[Hulk]

If you find that can you please post a link? I'd like to read it before I install the new drive. This might be a deal breaker on SSD's for me.

Here is one Dell white paper that claims MLC is good for 3 months after the P/E cycles are expired. And at max 10 years. So depending on the P/E cycles the drive had undergone and the temperature, somewhere between 3 months and 10 years.

http://www.dell.com/downloads/global/products/pvaul/en/Solid-State-Drive-FAQ-us.pdf

 

[SLM Crew Chief]

Thanks Hulk. That's pretty scary.

 

[taltamir]

20nm IMFT is rated for 3k P/E cycles, right?

At least Intel's is, can't see how Micron's would be any different since it's physically from the same fab.

Its more then just from the same fab
IMFT = Intel Micron Flash Technology
A jointly owned subsidiary firm of intel and micron. Its a single company that manufactures flash for both intel and micron.
So, basically IMFT flash... that being said, there is binning for different quality levels.

 

[Hellhammer]

The part you didn't quote me on was where I wrote that TLC needs to be lower priced because it's less proven technology than MLC. Samsung's estimates could be wrong, despite what they claim, is what I'm getting at. If TLC drives cost a little more than half of what MLC drives cost, I would get a TLC drive twice as big as I would get if it were MLC. But the actual price difference isn't that big. Until TLC falls in price more, I don't think it's worth getting over MLC drives.

TLC isn't exactly unproven technology. While it hasn't been used in SSDs until now, it's been used in USB sticks and various other devices for years. Samsung's estimations are not off, as I've tested before. There is also a SSD 840 at XtremeSystems that did over 3,500 P/E cycles, although it's good to keep in mind that retention time goes down as well (rated at 1 year after 100% of P/E cycles have been used), so in real world the usable amount of P/E cycles is less (I would say ~2000 to guarantee decent retention.

For comparison, the 120GB SSD 840 managed 432.9TiB of writes, whereas a 128GB Vertex 4 with 25nm IMFT MLC NAND managed only 393,9TiB. The fear towards TLC has no proof behind it.

Its more then just from the same fab
IMFT = Intel Micron Flash Technology
A jointly owned subsidiary firm of intel and micron. Its a single company that manufactures flash for both intel and micron.
So, basically IMFT flash... that being said, there is binning for different quality levels.

I'm well aware of that, but Intel and Micron validate the NAND on their own. For example, Intel offers 25nm MLC with 5,000 P/E cycles, whereas all Micron 25nm MLC I've seen is rated at 3,000 cycles. Hence Intel and Micron NAND are not exactly the same, Micron could be for example using lower binned chips in the M500 (and saving the better ones for enterprise), making the endurance less than 3,000 P/E cycles.

 

[taltamir]

I'm well aware of that, but Intel and Micron validate the NAND on their own. For example, Intel offers 25nm MLC with 5,000 P/E cycles, whereas all Micron 25nm MLC I've seen is rated at 3,000 cycles. Hence Intel and Micron NAND are not exactly the same, Micron could be for example using lower binned chips in the M500 (and saving the better ones for enterprise), making the endurance less than 3,000 P/E cycles.

You said you can't see them being different because they are made in the same fab.
I clarified they are more closely related then merely sharing a fab.
But also I am the one who brought up binning as an explanation why they WOULDN'T be exactly the same as you have stated.

 

[blastingcap]

TLC isn't exactly unproven technology.

Why compare OCZ's drive which is among the worst in terms of MLC testing at XS? Why not compare TLC to other MLC drives like Samsung's 830 drive which lasted for well over 20k writes IIRC?

Even if TLC were just as good as MLC, there's just no way I'm paying anywhere near what the asking price is for a 840 when it's barely any cheaper than 830's were. TLC should cost way less to produce, so I know I can just wait it out before prices fall. And if I needed a good MLC drive, they are still around, and some cost not much more than 840s; on sale, they may even cost less!

But reasonable minds can disagree and I wouldn't hold it against anyone if they want to entrust their data to a TLC drive that isn't that much cheaper than a non-OCZ MLC drive.

 

[AdamK47]

Nice! A 1TB SSD for under $600. This is another significant step forward in mechanical drive obsolescence.

I can get 4 of these 1TB drives for the same price as the six 512GB drive I purchased last month.

 

[WhoBeDaPlaya]

But reasonable minds can disagree and I wouldn't hold it against anyone if they want to entrust their data to a TLC drive that isn't that much cheaper than a non-OCZ MLC drive.

Well, given that my SSD is only used for holding the OS, games and VMs (which are backed up anyway), I don't particularly care.

Don't get me wrong, I wouldn't be happy at the inconvenience, but it's a hit I can take. Whether a SSD is SLC, MLC or TLC doesn't make a lick of difference to me - I treat all of them as fairly volatile storage given that a controller herp derp could render the entire drive inaccessible at any time (as compared to mechanical drives, which tend to die gradually).

 

[yefi]

With enough TLC it won't matter.

Why do you think it matters at current capacities? If you were to write 30GB/day (which is a damn lot) to even a 120GB 840, it would last you 40 years according to xtremetech. Even if you normalize the number to 1000 P/E cycles, it would still last you 11.4 years.

 

[jwilliams4200]

Even if you normalize the number to 1000 P/E cycles, it would still last you 11.4 years.

You seem to be assuming a very low write amplification, around 1.1.

Some workloads may have considerably higher WA, in the 2 - 5 range.

 

[Hellhammer]

Why compare OCZ's drive which is among the worst in terms of MLC testing at XS? Why not compare TLC to other MLC drives like Samsung's 830 drive which lasted for well over 20k writes IIRC?

My point was to show that it's not necessarily the NAND type that matters, but the whole package (NAND, controller & firmware). It's obvious that the SSD 830 is more durable because it's been designed by the same company and uses higher endurance NAND, so I don't think it's a fair comparison to begin with (there is no doubt that TLC is less durable than MLC).

Even if TLC were just as good as MLC, there's just no way I'm paying anywhere near what the asking price is for a 840 when it's barely any cheaper than 830's were.

SSD 830 was cheap only because Samsung was getting rid of their inventory, it's not even available anymore. It wasn't cheap at the time of its release and the hot sales started only a couple of months before the introduction of SSD 840.

TLC should cost way less to produce, so I know I can just wait it out before prices fall. And if I needed a good MLC drive, they are still around, and some cost not much more than 840s; on sale, they may even cost less!

At least in my opinion, the SSD 840 has been priced competitively, see for example the tables here. It's often the cheapest name-brand SSD around and if it's not the cheapest, it's at least among the top three. Of course if you want to include every small-brand SandForce drives, then you'll probably find an MLC NAND based SSD that's always cheaper than the SSD 840. However, I would rather pick the SSD 840 than buy a SandForce based SSD from a small, unproven manufacturer.

The big picture in the SSD 840 is pretty good IMO.

 

[yefi]

You seem to be assuming a very low write amplification, around 1.1.

Some workloads may have considerably higher WA, in the 2 - 5 range.

It's a justified assumption, given that his isn't an enterprise workload and the drive has demonstrated a WA of around 1 in the Anvil Storage bench. The drive I have seems to be adhering to the rule of WA of around 1 as well.

 

[blastingcap]

At least in my opinion, the SSD 840 has been priced competitively, see for example the tables here. It's often the cheapest name-brand SSD around and if it's not the cheapest, it's at least among the top three. Of course if you want to include every small-brand SandForce drives, then you'll probably find an MLC NAND based SSD that's always cheaper than the SSD 840. However, I would rather pick the SSD 840 than buy a SandForce based SSD from a small, unproven manufacturer.

The big picture in the SSD 840 is pretty good IMO.

I agree that I was spoiled by the deep discounts last year. There was even a brief moment in time where I could have picked up 830 (256GB) drives for about $128 after all cashback and incentives. So you are right as far as current 830 availability and pricing. But reputable drives like the M4 are still around so I'm not counting OCZ Agility 2's on clearance or something... I am saying that if you wait for a sale on something like an M4 you won't necessarily have to pay much more than you would for a Sammy 840 of the same/similar capacity.

For those who must have a decent SSD NOW, and can't wait for a sale, I suppose the 840 is priced fairly, but given the hiccups we've seen with SSD controllers and firmware in the past, it is probably still less risky to buy an older SSD with a proven controller-NAND-firmware combo like Plextor, Crucial M4, Sammy 830, Intel X-25M G2, etc. This is the case even if we ignored the TLC/MLC distinction and used non-sale pricing.

That said, Sammy has a good reputation overall, and buying a 840 is a reasonable purchase... and definitely isn't the worst SSD risk one could take!

 

[jwilliams4200]

It's a justified assumption, given that his isn't an enterprise workload and the drive has demonstrated a WA of around 1 in the Anvil Storage bench. The drive I have seems to be adhering to the rule of WA of around 1 as well.

That may be the case. I'd even go so far as to say it is more likely than not to be the case.

But if there is even a 10% chance that the WA of his workload will be 2 or higher, then I think it is bad advice to assume a WA of 1.1 in estimating the endurance. In the absence of solid evidence, it is better to be conservative about these sorts of things.