Any news on the release date for the Crucial M500 1TB SSD?

[Kippa]

Any news when thee Crucial M500 1TB SSD will be released? I am thinking of upgrading to an SSD, just want to know does anyone have any idea when it is due out?

 

[ThePiston]

I've checked and I can't tell a specific date. I do see that they are in production though:
http://www.micron.com/products/solid-state-storage/client-ssd#fullPart&306=M500&306=M500

edit* the 960MB is "sampling" though

 

[ThePiston]

this page says first week of April:
http://www.upgradeable.com.au/blog/2013/03/14/960gb-crucial-m500-ssd-now-available-for-pre-order/

Australian preorder site: http://www.upgradeable.com.au/parts/UCA1639.html

 

[sub.mesa]

Should be petty soon now. Beware that availability may still be fluctuating during the launch. This SSD I've been having an eye on for many months now. It has all the potential to become the best SSD of 2013.

The 1TB version (actually 960GB) seems to be vastly cheaper per GB than the 120GB and 240GB offerings. Those wanting to switch to solid state storage completely and not use mechanical storage will find this SSD to be the first serious option that is reasonably affordable.

The Crucial M500 is a huge upgrade over its predecessor. It sports a brand new controller with good performance, RAID5 redundancy on the NAND level, power-safe capacitor to prevent corruption on sudden power loss as well as general good performance. Assuming the M500 will continue the trend of competitive pricing, this could very well make it the best SSD of this year. Together with the Intel 320, this will be the second consumer-grade SSD that is inherently reliable.

 

[_Rick_]

To me that sounds a lot like something with enterprise class features - meaning that it won't be cheap.

The RAID 5 redundancy is something that's actually a bit worrying, all things considered, because it means they expect their flash to start dying during the warranty period, and hide that behind some 1:10 or so redundancy factor. It also remains to be seen whether that is any more effective than just using spare area, as RAID 5 amplifies the number of writes, and thus leads to faster NAND-exhaustion. As flash is supposedly immune to read errors, RAID 5 would gain very little advantage (some performance advantage when NAND dies during a write, as you can queue up writing the parity information, retains data if NAND dies in such a way that it is rendered unreadable), and only add disadvantages (higher complexity, more writes, calculation of parity, more power consumption).

Oh well, with the launch only a few weeks away, we'll see what they ended up doing.

 

[ThePiston]

I expect we'll see some reviews posted before they are available for purchase

 

[Hellhammer]

I expect we'll see some reviews posted before they are available for purchase

We don't have review samples yet, so I'm very doubtful that the shipping will start next week.

 

[sub.mesa]

To me that sounds a lot like something with enterprise class features - meaning that it won't be cheap.

As always, the actual cost is negligible. Just like ECC and other 'enterprise' features, these are only expensive because the companies use political pricing. They know their enterprise users can pay much more, so they need to find a trick justifying that. This usually means separating product lines and forcing those who can pay more, to actually do pay more.

The array of capacitors would add a dollar perhaps to the total price. I think it's well worth it. Just like ECC would be worth paying 2 dollars more as consumer. But if they gave everyone ECC, there would be no more opportunity to rip-off enterprise customers. They still want their wealthy customers to pay thousands of dollars for a few dollar additional costs. That is how the system works, mate.

The RAID 5 redundancy is something that's actually a bit worrying, all things considered, because it means they expect their flash to start dying during the warranty period, and hide that behind some 1:10 or so redundancy factor. It also remains to be seen whether that is any more effective than just using spare area, as RAID 5 amplifies the number of writes, and thus leads to faster NAND-exhaustion.

I'm not sure you understand the whole meaning of bit-level redundancy? The whole idea here is to avoid 'bad sectors'. I also don't see how spare space is relevant in this context. The actual overhead is 1:16 (both in available space and in amplified writes).

The redundancy is crucial (pun intended) to protect the Flash Translation Layer (FTL) that stores the difference between logical LBA and physical NAND addressing. Corruption on an SSD is much worse than corruption on a HDD. The corruption can cause the entire SSD to fail. Once the FTL is corrupted, your problems start. Many people who are having problems with their SSD, encounter exactly this issue.

Now you are telling me this protection is irrelevant? We consumers should be happy with the crappy SSDs that we got today?

As flash is supposedly immune to read errors

It is? Did I miss something? I always thought modern flash had to spend more than half of their raw capacity on error correction, otherwise no sector could be retrieved without corruption. If you believe this is incorrect, I highly anticipate your understanding of it.

 

[AdamK47]

Hmmm.... 4 of these or get two more 512GB Vertex 4s. Running out of space on my 3TB array. Only have 103GiB left.

 

[ThePiston]

Hmmm.... 4 of these or get two more 512GB Vertex 4s. Running out of space on my 3TB array. Only have 103GiB left.

only bad thing about M500 is that they are new and untested in the wild. firmware will probably need tweaking once it's out too

 

[_Rick_]

As always, the actual cost is negligible. Just like ECC and other 'enterprise' features, these are only expensive because the companies use political pricing. They know their enterprise users can pay much more, so they need to find a trick justifying that. This usually means separating product lines and forcing those who can pay more, to actually do pay more.

The array of capacitors would add a dollar perhaps to the total price. I think it's well worth it. Just like ECC would be worth paying 2 dollars more as consumer. But if they gave everyone ECC, there would be no more opportunity to rip-off enterprise customers. They still want their wealthy customers to pay thousands of dollars for a few dollar additional costs. That is how the system works, mate.

I understand - and that's why they'd be cutting off their own leg, by offering these features cheaply. Unless they see themselves setting those losses off by massively increased sales and profit int he consumer sector.

I'm not sure you understand the whole meaning of bit-level redundancy? The whole idea here is to avoid 'bad sectors'. I also don't see how spare space is relevant in this context. The actual overhead is 1:16 (both in available space and in amplified writes).

The redundancy is crucial (pun intended) to protect the Flash Translation Layer (FTL) that stores the difference between logical LBA and physical NAND addressing.

So it's actually redundancy in the "file-system"-like layer of the SSD, and not the actual data? Much like a real file system, the organizational information and the data are likely stored separately? Or are they actually using 16/8 dies and spread parity across dies/half-dies? Spare space is relevant, as it is what you use, when flash starts dying. With parity spread throughout, you lose spare space, but you can continue using existing data (slowly) even if one cell fails.

Corruption on an SSD is much worse than corruption on a HDD. The corruption can cause the entire SSD to fail. Once the FTL is corrupted, your problems start. Many people who are having problems with their SSD, encounter exactly this issue.

Again, the FTL is a tiny subsection (similar to a FAT, if I had to guess) of the SSD's memory. Corruption in the organisational layer is always worse than actual data corruption, but usually this corruption comes from firmware bugs, memory and transmission errors...rarely from flash failure. Also, it's just as important to protect the FS-section of the data, as it si important to protect the FTL. Either of those failing means all your data becomes close to impossible to recover.

Now you are telling me this protection is irrelevant? We consumers should be happy with the crappy SSDs that we got today?

I'm saying that the problems are elsewhere, than in the flash layer. And that 1:16 redundancy isn't a whole lot, if indeed your flash is suddenly dying.

It is? Did I miss something? I always thought modern flash had to spend more than half of their raw capacity on error correction, otherwise no sector could be retrieved without corruption. If you believe this is incorrect, I highly anticipate your understanding of it.

According to Wikipedia, ECC is about 1/32 of the raw capacity, for NAND flash. It could be possible, that by using TLC, they are forced to use more ECC than that, and thus increase to 1/16 (and use that as base for their "RAID-5" claims). That will probably allow them to offer a competitive warranty and lifetime.
They surely won't be using more ECC than absolutely necessary, as this means less capacity per die, and thus makes the device more expensive per GB.

More than half would be pretty disastrous.

All in all, I suspect that what they're actually doing is spread ECC across dies, instead of keeping it in-die, and increase the ECC ratio to 1:16, to deal with cheaper (per GB) TLC dies. It does improve wear leveling, and it will cover more bit errors, but won't help with an entire page dying at once.
The problem with using actual RAID 5 is that write amplification goes way higher, because your minimal writable units grow massively (factor 15). This means a read-erase-write now takes 16 times the writes/erases it used to take. This doesn't sound good for longevity.

 

[Ao1]

Is it going to use Enhanced ClearNAND Flash?

http://www.micron.com/products/managed-nand/clearnand

 

[Hellhammer]

Is it going to use Enhanced ClearNAND Flash?

http://www.micron.com/products/managed-nand/clearnand

Nope.

That FBGA code corresponds to MT29F256G08CECABH6-10ES:A, whereas ClearNAND is MT29FCA... or MT29FEN...

 

[corkyg]

Thank you Ao1 and Hellhammer for getting this thread back on track for the original question. So, is the release date sill Q2 2013?

 

[ThePiston]

word on the street is April 11. I'm waiting for these to build so I'm anxious.

 

[Hellhammer]

word on the street is April 11. I'm waiting for these to build so I'm anxious.

I'm still skeptic about that because we have absolutely no info regarding the review samples. April 11th is only 13 days away, so the deadline for reviews would end up being very, very short (shipping alone takes at least a few days). Of course, this wouldn't be the first time a manufacturer only gives us a day or two to come up with an in-depth review...

 

[ThePiston]

all of the 7mm versions (except the 960GB version) have been in production since beginning of March at least. The 960GB must have gone from "sampling" to "production" only in the last week. So, if they're in production you could get one of them soon - theoretically anyway.

There's nothing worse then waiting on one part for a build.

I think the SSD price drops over this past week are because of this rollout. It'll be the SSD to beat for a lot of reasons if it lives up to the hype.

 

[Cerb]

I'm saying that the problems are elsewhere, than in the flash layer. And that 1:16 redundancy isn't a whole lot, if indeed your flash is suddenly dying.

OTOH, it could a life-saver, if any data got corrupted. The nice thing about ECC above the lowest storage layer, is that you don't have to care what the source of the error was, or will be.

They surely won't be using more ECC than absolutely necessary, as this means less capacity per die, and thus makes the device more expensive per GB.

ECC at a higher level, for your data, not trusting the storage device, is always preferable. Hardware bugs happen. Power loss, ripple, voltage spikes, etc., happen, and might cause corruption.

This means a read-erase-write now takes 16 times the writes/erases it used to take. This doesn't sound good for longevity.

No, it means they'll take 1/15th more, plus an additional write somewhere, per standby command received, or other similar events. A custom RAID-like parity scheme need not suffer the negative consequences of traditional RAID 4/5/6, which assume that a disk will report all errors correctly, drop out, or work perfectly, and that all writes are good writes.

Also, you can safely bet that similar schemes are used by other makers, too, if not straight-out data duplication, but only for the mapping data itself.

Finally, do we have any official word on whether they are using a RAID-like scheme for the M500, like they did the P400m? It wouldn't be a bad thing, but I haven't found anything substantiating it, as of yet, just rumor.

Relevant: https://www.usenix.org/system/files/conference/fast13/fast13-final80.pdf

 

[Lifted]

This doesn't look so hot...

www.micron.com/~/media/Documents/Products/Data%20Sheet/SSD/m500_2_5_ssd.pdf

Table 7: Drive Lifetime

Capacity | Drive Lifetime (Total Bytes Written)
120GB | 72TB
240GB | 72TB
480GB | 72TB
960GB | 72TB

That's 75 writes across the 960GB drive (or per cell). :hmm:

Doesn't make sense considering the 120GB drive has the same 72TB figure.

 

[Emulex]

Well if its just 72TB, buy two samsung 840 non-pro and rock those, $500 (or less) with recent deals for 500gb x 2 .

$375 * 2 for 840 pro's (newegg recent deal) if you need many more times that drive lifetime.

 

[Cerb]

*SOMETHING* is definitely up, there, that's for sure. The same total write endurance rating minimum across the capacity spectrum makes absolutely no sense, to me. Even if they used lower-endurance NAND for the larger drive(s), the stats should at least vary some. 72TB for 120GB is about expected (600 fully random overwrites), but that should make for nearly 600TB for the 960GB drive.

 

[Hellhammer]

This doesn't look so hot...

www.micron.com/~/media/Documents/Products/Data%20Sheet/SSD/m500_2_5_ssd.pdf

That's 75 writes across the 960GB drive (or per cell). :hmm:

Doesn't make sense considering the 120GB drive has the same 72TB figure.

Don't forget the small print:

1. Total bytes written calculated with the drive 90% full.
2. Access patterns used during reliability testing are 25% sequential and 75% random and
consist of the following: 50% are 4KB; 40% are 64KB; and 10% are 128KB.
3. GB/day can be calculated by dividing the total bytes written value by (365 × number of
years). For example: 72TB/5 years/365 days = 40 GB/day for 5 years.

Given the workload, that's on par with other drives from what I have seen. E.g. OCZ Vector is rated at 36.5TB (100% 4KB random writes).

 

[Cerb]

The small print explains the 72TB figure for the 120GB, which is quite reasonable, but not at all why that figure is 1/8th that, in relative terms, for the 960GB, much less why it's no higher at all for any other capacities.

 

[SSBrain]

It's their way of saying that these drives are intended for (and warranty valid for) consumer workloads, I guess.

By the way, the 960 GB model should generally have a much lower write amplification than the 120 GB model, so should last even more than its capacity difference from this one. I would expect 2 PB of writes at the very least.

 

[Hellhammer]

The small print explains the 72TB figure for the 120GB, which is quite reasonable, but not at all why that figure is 1/8th that, in relative terms, for the 960GB, much less why it's no higher at all for any other capacities.

Manufacturers are often just conservative with these figures.