Can't wait for the sales on last gen SSD! :P I could use another 830 or two if they suddenly start to get cleared out cheaper than they've already been...Reply
TLC does follow the same 2^n math because the die is cut down in size. So an 8GB MLC die has 4 billion cells (actually more since NAND size is actually measured in Gibibytes) but with TLC that number is reduced to ~2.67 billion cells (multiply that by three and you get ~8GB). The die is physically smaller, which leads to more dies per wafer and hence cost savings.
Well my point was that 8GiB has exactly 2^n bits, which is not divisible by 3, so you can't even reach exactly 8GiB with TLC unless you treat the last transistor as SLC (8, 32 GiB) or 2bpc MLC (16 GiB). So maybe they skipped the whole 2^n thing and went with something that adds up to approx. 120GB total. Anyway, more spare area makes more sense, I was just thinking out loud.Reply
The first chart of this article says 840 is MLC(2bits/cell) and 840 Pro is TLC. But it contradicts the performance nature. Perhaps their values are swapped?Reply
You're right, sorry about that, it's fixed now! The schedule in Korea has been pretty strict, so I didn't have too much time to write this article :-)Reply
I hope the retail price will be a little lower. Given it is using TLC, its price is actually the same as current MLC 120GB SSD. Assuming 840 Pro will be faster then 840, then there are lots of other options that provide equal or slightly less performance with MLC for the same price.Reply
How will this drive's controller handle different sized drives? Will the 512 have better performance? Will smaller drives have poorer performance?Reply
Hard to say at this moment, but 256 and 512 has same sized cache. 128 GB model has smaller cache, so at least there should be some difference. Interesting to see normal 840 model. It is slower in writing than 830 but in other aspect guite near or even better if specks are correct. But at this moment 830 offer better performance for the money.Reply
Thanks for the info and definitely looking forward to a review. Could you also post the MSRP at which the 830 launched? That would give a little hint towards street pricing (maybe). :-)Reply
everything that drives down cost of Flash reduces write cycles. Smaller processes, now TLC. Sure controllers spread it around and are getting better, but there's a limit somewhereReply
Yes, the individual chips will have less write cycles, but each chip can store more information. If you kept writing at 100mb/s on a drive till it dies, the drive with the larger capacity will generally last longer.Reply
If most of your writes go to just a little 'hot' data, there are sneaky ways to use a mix of expensive and cheap Flash to store it at minimum cost.
For instance, if you're the human optimizing a database's storage system, you could put active tables on 'pro' drives and some rarely-written ones on 'consumer' drives. Or a single SSD could contain 3/4 MLC and a 1/4 SLC, and its controller could send all writes to SLC at first and evict the SLC's 'coolest', least-likely-to-be-rewritten pages to MLC as required.
I don't think we'll see much of that soon, because complexity sucks and nobody is clamoring for a compromise between 'pro' and 'consumer'. But someday, it may take tricks like that to make ever-less-durable Flash useful at all.Reply
29 Comments
Back to Article
sabaramo - Monday, September 24, 2012 - link
looks like an awesome drive, now the wait for the newegg reliability reviews :DCant wait for next gen ssd's! Reply
Impulses - Monday, September 24, 2012 - link
Can't wait for the sales on last gen SSD! :P I could use another 830 or two if they suddenly start to get cleared out cheaper than they've already been... Replyajp_anton - Monday, September 24, 2012 - link
Do TLC chips really add up to 2^n bits? Maybe that's the reason they are 120/250/500GB. Or what you said about spare area, but just checking. ReplyKristian Vättö - Monday, September 24, 2012 - link
TLC does follow the same 2^n math because the die is cut down in size. So an 8GB MLC die has 4 billion cells (actually more since NAND size is actually measured in Gibibytes) but with TLC that number is reduced to ~2.67 billion cells (multiply that by three and you get ~8GB). The die is physically smaller, which leads to more dies per wafer and hence cost savings.This is all explained in the TLC article as well: http://www.anandtech.com/show/5067/understanding-t... Reply
ajp_anton - Monday, September 24, 2012 - link
Well my point was that 8GiB has exactly 2^n bits, which is not divisible by 3, so you can't even reach exactly 8GiB with TLC unless you treat the last transistor as SLC (8, 32 GiB) or 2bpc MLC (16 GiB).So maybe they skipped the whole 2^n thing and went with something that adds up to approx. 120GB total.
Anyway, more spare area makes more sense, I was just thinking out loud. Reply
A5 - Monday, September 24, 2012 - link
There are plenty of numbers along the 2^n chain that are also divisible by 3. ReplyUrizane - Monday, September 24, 2012 - link
Math fail. Replyajp_anton - Monday, September 24, 2012 - link
Please name one, and I'll give you that many $. ReplySkarn - Monday, September 24, 2012 - link
2^50 = 1,125,899,906,842,620.1,125,899,906,842,620 / 3 = 375,299,968,947,541
Every power of two beyond 50 will likewise be divisible by 3. Reply
Skarn - Monday, September 24, 2012 - link
Ignore that. I just realized my machine is introducing rounding errors at this magnitude. ReplyDeath666Angel - Monday, September 24, 2012 - link
Sounds like http://xkcd.com/217/ Replysurt - Wednesday, September 26, 2012 - link
2^N is never divisible by 3. This is due to the principle of unique prime factorization. Replysweetie peach - Tuesday, September 25, 2012 - link
zeroscrew irony Reply
ajp_anton - Tuesday, September 25, 2012 - link
2^0 = 1, not divisible by 3.2^n = 0 doesn't exist, unless n -> -∞, in which case I'll gladly give you -∞ $. Reply
Old_Fogie_Late_Bloomer - Monday, October 01, 2012 - link
/thread Replyk2_8191 - Monday, September 24, 2012 - link
The first chart of this article says 840 is MLC(2bits/cell) and 840 Pro is TLC.But it contradicts the performance nature.
Perhaps their values are swapped? Reply
Kristian Vättö - Monday, September 24, 2012 - link
You're right, sorry about that, it's fixed now! The schedule in Korea has been pretty strict, so I didn't have too much time to write this article :-) Replyk2_8191 - Thursday, September 27, 2012 - link
Thanks for the fix! =) Replyiwod - Monday, September 24, 2012 - link
I hope the retail price will be a little lower. Given it is using TLC, its price is actually the same as current MLC 120GB SSD. Assuming 840 Pro will be faster then 840, then there are lots of other options that provide equal or slightly less performance with MLC for the same price. ReplyChas1 - Monday, September 24, 2012 - link
Wouldn't that equate to 256 Terabytes? ReplyChas1 - Monday, September 24, 2012 - link
Well, 255, I guess. Point being that's a lot of space that I don't think it has. ReplyNathan Leopold - Monday, September 24, 2012 - link
How will this drive's controller handle different sized drives? Will the 512 have better performance? Will smaller drives have poorer performance? Replyhaukionkannel - Monday, September 24, 2012 - link
Hard to say at this moment, but 256 and 512 has same sized cache. 128 GB model has smaller cache, so at least there should be some difference.Interesting to see normal 840 model. It is slower in writing than 830 but in other aspect guite near or even better if specks are correct. But at this moment 830 offer better performance for the money. Reply
Death666Angel - Monday, September 24, 2012 - link
Thanks for the info and definitely looking forward to a review.Could you also post the MSRP at which the 830 launched? That would give a little hint towards street pricing (maybe). :-) Reply
Zoomer - Monday, September 24, 2012 - link
Please test latencies! We'll see how much of a difference there is, and perhaps for different drives as well. ReplyHacp - Wednesday, September 26, 2012 - link
There might be a difference in latency, but the dram cash should mask all of it. ReplyMamiyaOtaru - Wednesday, September 26, 2012 - link
everything that drives down cost of Flash reduces write cycles. Smaller processes, now TLC. Sure controllers spread it around and are getting better, but there's a limit somewhere ReplyHacp - Wednesday, September 26, 2012 - link
Yes, the individual chips will have less write cycles, but each chip can store more information. If you kept writing at 100mb/s on a drive till it dies, the drive with the larger capacity will generally last longer. Replytwotwotwo - Monday, October 01, 2012 - link
If most of your writes go to just a little 'hot' data, there are sneaky ways to use a mix of expensive and cheap Flash to store it at minimum cost.For instance, if you're the human optimizing a database's storage system, you could put active tables on 'pro' drives and some rarely-written ones on 'consumer' drives. Or a single SSD could contain 3/4 MLC and a 1/4 SLC, and its controller could send all writes to SLC at first and evict the SLC's 'coolest', least-likely-to-be-rewritten pages to MLC as required.
I don't think we'll see much of that soon, because complexity sucks and nobody is clamoring for a compromise between 'pro' and 'consumer'. But someday, it may take tricks like that to make ever-less-durable Flash useful at all. Reply