New 15nm chips from Toshiba, SanDisk

Toshiba Starts Mass Production of World's First 15nm NAND Flash Memories | Business Wire

Toshiba, SanDisk start producing smaller, faster 15nm NAND flash memory | PCWorld

Currently, the 19nm chips Are what you see on our ssd drives that are top of the line. Supposedly cheaper and 1.3x faster, that's really exciting. I wonder who will be first to market with these chips in their drives?

 

Will really only be beneficial in PCI-e/m.2 format because current gen chips can easily saturate the SATA III bus.

 

That's mainly what i'm interested in. I think the theoretical maximum is 10gb/s on M.2, I'd like to see it at least exceed the Sata III performance sometime soon. Also, shouldn't the memory be slightly more dense now, offering the possibility of more storage in, say, the 42mm M.2 size?

 

I wonder how many P/E cycles will these 15nm NAND chips endure...

 

RAW writes? Less than 300. Enhanced with DSP firmware? Probably closer to 1K. On paper. Or, in a simulated nand write fest.

In real world use? I would be looking for 1TB and larger models with light use for this type of nand (along with OP'ing and at least 100GB free space at all times on the actual partitions used).

 

This doesn't sound very good... Or am I being just overly pessimistic and there's nothing to worry about?

 

Nothing to worry about. TLC NAND was supposed to be 1,000 write/erase cycles but it's shown to be many times higher than that. And even at 1000 w/e cycles with a 1TB model, writing 100GB/day with a worse case 2:1 write amplification you're looking at well over 12 years AT MINIMUM.

 

An SSD's longevity is not just about high erase cycles (as many have learned).

And TLC nand is still rated for ~1K write cycles. That hasn't changed. Why we're getting higher numbers is because of the DSP firmware employed to get the seemingly higher endurance.

Being pessimistic is a good thing if the value being evaluated would be catastrophic if used up.

And/or if you don't know your actual workload (as it applies to the storage subsystem) and there you definitely want to over estimate your usage.

Those 12 years minimum can quickly become 12 weeks maximum if the workload is not matched appropriately to the SSD chosen (the 'I didn't know I wasn't supposed to use it like that' excuse won't work for warranty purposes either).

 

I guess I am just a normal user. I don't do any storage intensive tasks on my PCs. SSDs in my systems are for the OS and most used applications. Total writes on my Samsung 840 SSD 120GB were less than 3TB in the last year. Apparently nothing to worry about for me then.

 

I don't disagree that you want to over compensate, but it's not "sky is falling" in any respects. I know failures aren't just NAND, it's primarily controller, which you can have the best NAND in the world and the controller can just give up the ghost at any time. But it's still not something to worry about a component that is likely to be replaced every 4-5 years for probably 95% of users. If you do a lot of data writes, you likely will be seeking a more robust solution, or it's live and learn the hard way. If you do lots of video processing you sure will investigate the most robust storage because the data and time is important to you, if you don't even if it isn't the NAND or the controller, the laptop subsystem could fail. Putting yourself in that position is the users fault. Not to mention ALWAYS ALWAYS ALWAYS having redundant backups.

 

All of the drive failures I've heard of are controller/firmware related - aside from the few people who were testing the 840's endurance back when it first came out (and even that took far longer than expected), I've never heard of anybody who's managed to use up all of their NAND's write cycles.

Has anybody here actually managed to send any SSD into read-only mode without running special-purpose software that writes garbage data to the SSD as fast as it can travel over the SATA bus?