Samsung Begins 3D XPoint Assualt With Z-NAND SSDs and HybriDIMMs

[Guest]

Samsung unveiled its dual-pronged attack on 3D XPoint with both Z-NAND SSDs and Netlist HyrbiDIMMs.

Samsung Begins 3D XPoint Assualt With Z-NAND SSDs and HybriDIMMs : Read more

 

[memadmax]

I like samsungs' approach with the HybriDIMM alot better than intel's, because intel is in cahoots with microsoft to lock down the system via tech corraling: "You want to run this hardware? Well then you have to run this microsoft operating system"...

Get it? And you guys that are cheerleading microsoft and winblows 10 are might as well be the sheep in that corral...

 

[hannibal]

The support for X-point will arrive to Linux and iOS platforms too, so no worry in there. But hybriDimm works now, so it has merits on its own.

 

[epobirs]

Considering who the biggest customers for 3D Xpoint as memory are likely to be and that those same customer aren't likely to be building their systems for Windows Server, I very much doubt this is going to be locked into Wintel offerings. The use case for really big enterprise systems is far easier to find than that for a desktop PC. I would expect a lot of work remains to be done before a desktop OS really knows how to much real use of this. It can be treated as a traditional RAM disk but that is just scratching the surface of the potential.

 

[costeakai]

nice dog fight

 

[jasonf2]

In the immediate term Hybridimm will only have server applications. Most (if not all) consumer motherboards and chipsets will top out on addressable memory long before 256 gb. Intels magic here will be that the 200 series chipset coming with kabilake will support xpoint and handle the increased headroom necessary to put nonvoltile on the memory bus. Samsung's move here could lead to some workstation class boards becoming popular in the high end consumer class depending on how expensive xpoint really is (in the short term). At the very least we should see some competition here starting the race to the bottom on the emerging techs. I hope AMD and Zens chip set are taking all of this into account and at least getting ready for the need for a terabyte+ of memory on the Ram bus.

 

[serendipiti]

I don't agree the assumption of endurance coming from using TLC NAND as SLC.
I suppose the difference between 3D NAND and Z-NAND is that 3D NAND is stacking 2D NAND in multiple layers, while Z NAND takes the 3D concept to the cell itself by changing the doping areas and materials. This way the cell may have less leaks, resulting in a higher endurance, etc...

 

[josejones]

So, when will these new Samsung Z-SSD's come available and for which chipset? The Z-270 or the Z-370 motherboards? Kaby Lake or Cannon Lake? 2017 or 2018?

 

[hannibal]

Well, the point of hybriDimm is that it use old production techniques, so it can be produced with old factories. That seems to prove the point that this is indeed MLC acting in SLC mode. That makes it relatively cheap to produce them, compared X-point that requires own production capasity and technic.

 

[Tykkopoles]

I like samsungs' approach with the HybriDIMM alot better than intel's, because intel is in cahoots with microsoft to lock down the system via tech corralling...

So you approve a technology of one company not because the technology has any merit, but simply because you are throwing a temper-tantrum about that company's competitor. GGWP, mate.

In adult news, Z-NAND and HybriDIMM are both very cost-effective approach for enterprises to massively overhaul storage and memory solutions. I'm definitely interested to see some non-emulated tests on performance under heavy server workloads. I, for one, am happy that there are multiple companies innovating on volatile and non-volatile options.

 

[computerguy72]

Funny how Samsung had to 'massage' the numbers using the xpoint results adapted to a pcie x4 with standard drivers and then using a emulated 'simulation' to get their own numbers in there. Intel/Micron seem to have a solution that can dramatically outperform anything else. Samsung no doubt hopes to dull the enthusiasm but they will need to do better than this that's for sure.

 

[MRFS]

The use case for really big enterprise systems is far easier to find than that for a desktop PC. I would expect a lot of work remains to be done before a desktop OS really knows how to much real use of this. It can be treated as a traditional RAM disk but that is just scratching the surface of the potential.

Several years ago, we filed a Provisional Patent Application for adding a
"Format RAM" feature to BIOS subsystems, easily adapted to UEFI subsystems:

http://supremelaw.org/patents/bios.enhancements/provisional.application.1.htm
http://supremelaw.org/patents/bios.enhancements/provisional.application.2.htm

Installing an OS into a non-volatile ramdisk seems like a natural extension of Optane's capabilities.

A more ambitious approach would dedicate one-third of a triple-channel
memory subsystem to Optane, and adding the logic required to
BOOT from that subsystem, either during COLD starts or during WARM starts:

4 x DIMM slots w/ DDR4 operating in quad-channel mode
2 x DIMM slots w/ Optane operating in dual-channel mode, hosting NV-OS


Compare what we are doing right now with our workstations:

(a) first, we "Migrate OS" to a fast RAID-0, and maintain the source OS
as a backup, in case we need to boot from that original partition;
cf. PartitionWizard freeware's "Migrate OS" feature;

(b) if/when we need to restore a drive image to the fast RAID-0,
we boot from the original partition, and then run GHOST restore
(this works much faster than running GHOST from a CD-ROM).

Because the logic required to create and maintain a transparent NTFS ramdisk
has already been developed e.g. see RamDisk Plus from www.superspeed.com
it doesn't seem to me like a LOT of new R&D would be required
to format a ramdisk as part of a fresh OS install. See our Provisional Patent
Applications above, for one straightforward approach.

(I'm open to hearing why the latter statement is NOT entirely correct.)

Hosting an OS like Windows in a non-volatile ramdisk
is something that prosumers would enjoy using
on their primary workstation.

With that capability working and perfected, such a workstation
cannot be too far from an "INSTANT ON" ability too.

And, we already host all of our Internet browser caches
in a 13GB ramdisk, and the performance of Firefox
is noticeably accelerated by this one change.

... just thinking out loud here, OK?

MRFS

 

[MRFS]

p.s. Some of the features which I just described above
can also be approximated / demonstrated with a modern workstation
that is powered by a reliable UPS / battery backup:

(a) assume that it runs 24/7 and only powers OFF
for periodic maintenance: this renders conventional DDR4
effectively "non-volatile" (because the power stays ON
without any interruptions between scheduled maintenance
events);

(b) enhance the BIOS/UEFI subsystem to "Format RAM"
and host an OS in a ramdisk formatted using the
uppermost 50GB of DDR4;

(c) with a Windows OS, that uppermost 50GB is
recognized as the C: system partition and operates
exactly the same as C: on any other modern storage
subsystem;

(d) the remaining DDR4, ending at one byte address
less that the starting byte address of C:,
operates just like the DRAM remaining
after configuring a ramdisk with RamDisk Plus.

All of the above can be done withOUT needing Optane necessarily; and,
all of the above can be done WITH Optane too.

Thus, the question for today is this:
what does Optane bring to a workstation which
a workstation cannot already do withOUT Optane?

One answer: it eliminates the time-consuming
SAVE and RESTORE steps which are required
at SHUTDOWN and STARTUP of systems
using a utility like RamDisk Plus. That, in turn,
means that browser caches remain intact
even after power is switched OFF.

As correctly observed above, YES, this is
"only scratching the surface" of what
Optane brings to the table.

 

[xyriin]

The important take out of all this is that Samsung has a solution that looks like it could very well hit the market before XPoint and that's really the point right? (no pun intended)

Intel and Micron have been teasing XPoint for over a year now with no product to market. Furthermore, each new data point that comes along makes it a little less shiny on the performance and cost side. Seeing as Samsung wouldn't need an infrastructure change or build out to release Z-SSD it seems like XPoint may end up going nowhere.

 

[MRFS]

Seeing as Samsung wouldn't need an infrastructure change or build out to release Z-SSD it seems like XPoint may end up going nowhere.

i had a good chuckle over the weekend, when I began assembling
an NVMe workstation in my mind.

I went to Newegg and looked for "U.2 cables", and
then just poked around to see what else came up e.g.:

http://www.newegg.com/Product/Product.aspx?Item=N82E16817801139&Tpk=N82E16817801139
http://www.newegg.com/Product/Product.aspx?Item=9SIAA6W3YY8665&Tpk=9SIAA6W3YY8665

So, with a Highpoint 3840A, I can assemble a RAID-0 with 4 x Samsung M.2 NVMe SSDs,
and I won't even need a backplane: the latter U.2 cable plugs right
into that Syba M.2 enclosure:

http://www.newegg.com/Product/Product.aspx?Item=N82E16820147466&Tpk=N82E16820147466

The 2.5" form factor has become rather universal,
and a modern chassis should already come with
proper cooling for 3.5" drive bays.

The wiring topology is almost identical to standard 2.5" SSDs.

With a tower chassis that has 2 empty 5.25" bays, this next enclosure should
cool 4 x 2.5" SSDs nicely, and it's not too expensive either:

http://www.newegg.com/Product/Product.aspx?Item=9SIA9PV4374597&Tpk=9SIA9PV4374597