Micron To Begin Shipping Products With TLC NAND This Quarter

[Guest]

We are fresh out of the Micron earnings call and have some flash news to report.

Micron To Begin Shipping Products With TLC NAND This Quarter : Read more

 

[dgingeri]

Mistake, mistake, major mistake. Samsung has already proven that TLC is a really bad idea and doesn't work right. Now we'll have a whole new line of products that have disappointing performance without constant rewrites.

 

[TallestJon96]

honestly, this is really being overshadowed by xpoint. Everything until then feels like a placeholder, and pricing is largely more important than performance for many.

The mention of 8k gaming is very specific. Is it possible that Xpoint will be used in GPUs?

 

[gangrel]

The mention of 8k gaming is very specific. Is it possible that Xpoint will be used in GPUs?

That's the way I read it. Makes sense, too.

Also, I agree about the waiting game, to a point. Problem is that something like a new, mega GPU with 16G of XPoint is still a fairly long way off. They can't just sit for that long. The TLC is mostly ready to go. It does suggest that any improvements to prior tech, or even TLC, that aren't particularly mature, may wither, at least insofar as development all the way out to finished product. IF it made sense to do so, I could see proof-of-concept 2D NAND for, say, the 10 nm fabrication process.

 

[Sakkura]

Mistake, mistake, major mistake. Samsung has already proven that TLC is a really bad idea and doesn't work right. Now we'll have a whole new line of products that have disappointing performance without constant rewrites.

The Sandisk Ultra II and Samsung 850 Evo have already proved that the issues encountered by the 840/840 Evo are not something that will affect all TLC-based drives. It was the first mass-market SSD to rely on TLC, it's not surprising that there would be some teething difficulties. TLC in general is fine.

 

[spacev]

Using it for 8k gaming is just nuts! If that is where they want to use it, oh boy.
Also, last I heard, it was not as fast as DRAM.

4k gaming barely works with current high end cards, 8k is 4 times the pixels.

Plus, most people have 28 inch displays at best and 8k is overkill.

There will be no 8k gaming consoles in the foreseeable future and the PC ultra enthusiast market is minuscule! I dont see mainstream 8k gaming before 2020.

So good luck Micron.....

 

[norseman4]

Mistake, mistake, major mistake. Samsung has already proven that TLC is a really bad idea and doesn't work right. Now we'll have a whole new line of products that have disappointing performance without constant rewrites.

http://www.legitreviews.com/4tb-ssds-coming-in-samsung-850-pro-and-850-evo-series_172731

What's the difference between TLC and 3-bit MLC. I don't know of any, and Samsung is looking to release [strike]4GB[/strike](oops) 4TB 3b MLC next year.

 

[Sakkura]

http://www.legitreviews.com/4tb-ssds-coming-in-samsung-850-pro-and-850-evo-series_172731

What's the difference between TLC and 3-bit MLC. I don't know of any, and Samsung is looking to release 4GB 3b MLC next year.

TLC and 3-bit MLC are just two different names for the same thing.

The TLC Samsung is releasing next year is not 4GB, it's 256 Gbit = 32 GB.

 

[Rookie_MIB]

TLC is the wrong direction. 3D/V-Nand is the right way. MLC bit setup, large, stacked lithography gives you all the density and write endurance you could need. Yes, a little more complicated when going vertical, but dealing with the TLC issues is just too much trouble IMHO.

 

[norseman4]

http://www.legitreviews.com/4tb-ssds-coming-in-samsung-850-pro-and-850-evo-series_172731

What's the difference between TLC and 3-bit MLC. I don't know of any, and Samsung is looking to release 4GB 3b MLC next year.

TLC and 3-bit MLC are just two different names for the same thing.

The TLC Samsung is releasing next year is not 4GB, it's 256 Gbit = 32 GB.

I was pretty sure that it was the same thing, but not 100%.

As for Samsung's release, from my linked post:

Samsung announced that the 850 EVO and 950 PRO would be among the first drives to get the new 3bit V-NAND technology and that we’ll be seeing 4TB capacities of those drives in 2016.

 

[Sakkura]

TLC is the wrong direction. 3D/V-Nand is the right way. MLC bit setup, large, stacked lithography gives you all the density and write endurance you could need. Yes, a little more complicated when going vertical, but dealing with the TLC issues is just too much trouble IMHO.

You can combine 3D and TLC. That yields higher capacities than either 3D MLC or 2D TLC, and works better than 2D TLC. In any case, the TLC issues are manageable as proven by the Sandisk Ultra II.

As for Samsung's release, from my linked post:

Samsung announced that the 850 EVO and 950 PRO would be among the first drives to get the new 3bit V-NAND technology and that we’ll be seeing 4TB capacities of those drives in 2016.

Ah. You said GB instead of TB.

 

[CRamseyer]

3d TLC is in the 850 EVO already. Samsung' next move is going from 32 layers to 48. This increases die density.

 

[norseman4]

Ah. You said GB instead of TB.

So I did, I corrected my first post, though left the original error as an obvious error.

 

[Guest]

if xpoint really turns out to be "as fast as ram and as cheap as nand", we could think about "simply" putting it into the ram slots, replacing the ram, instead of wasting additional pcie slots.

according to a quick search ram has a bandwidth of around 30GB/s, while pcie3.0 x16 is only about 16GB/s.

we already have ramdisks, so even if hard- and software have to be adjusted for unified ram and storage, or even if we have to "emulate" a couple GB of ram, it might be possible.

 

[Sakkura]

if xpoint really turns out to be "as fast as ram and as cheap as nand", we could think about "simply" putting it into the ram slots, replacing the ram, instead of wasting additional pcie slots.

according to a quick search ram has a bandwidth of around 30GB/s, while pcie3.0 x16 is only about 16GB/s.

we already have ramdisks, so even if hard- and software have to be adjusted for unified ram and storage, or even if we have to "emulate" a couple GB of ram, it might be possible.

There's also already the concept of plugging an SSD based on NAND flash into a memory slot. 3DXpoint just makes that even more interesting since it should be better able to utilize the extra bandwidth.

Personally I was more expecting it to be used as a cache for traditional SSDs, instead of DRAM (DDR3/DDR4). That would bring the price down and also remove the problem of data integrity in the case of power loss, since 3DXpoint is non-volatile.

 

[gangrel]

If I recall the long XPoint article...it was, in fact, envisioned to plug XPoint straight into the RAM sockets.

Sakkura: I presume you meant SSHD, not SSD. I could easily see it as a cache for SSHDs for the reason you mention, but only if it's easy to implement this. XPoint threatens to supplant hard drives for many purposes, so the market window for an XPoint-based SSHD may be pretty narrow. I'm not saying the SSHDs will be disappearing all that soon; the issue is whether changing the implementation from DRAM to XPoint is simple enough to make it worthwhile.

 

[Sakkura]

Sakkura: I presume you meant SSHD, not SSD.

No, I mean SSDs based on NAND flash. They currently use a cache of DRAM. Replacing the DRAM with 3DXPoint would remove the vulnerability the DRAM introduces, because 3DXPoint is non-volatile. It would also reduce cost (or allow a bigger cache, if that helps performance). But it depends how much 3DXPoint will cost; if it's as cheap as NAND flash then you might as well make the entire SSD out of 3DXPoint.

 

[gangrel]

OK. I thought you might mean SSHD, because yes, even if XPoint is more but not a LOT more (say, not above about 2.5x the cost), using the NAND at all won't make that much sense. NAND prices will be under pressure, but there's a floor that can drop only so fast.

Meeting demand seems likely to be an issue with XPoint for a while.