Intel Optane (XPoint) First Gen Product Specifications Leaked

Subject: Storage | October 14, 2016 - 08:05 PM |
Tagged: XPoint, Optane, 8000p, Intel

Intel and Micron jointly launched XPoint technology over a year ago, and we've been waiting to see any additional info ever since. We saw Micron demo a prototype at FMS 2016, and we also saw the actual prototype. Intel's last demo was not so great, later demos were better), and we saw a roadmap leaked a few months ago. Thanks to another leak, we now have specs for one of Intel's first Optane products:

View Full Size

Now I know there is a bunch of rambling around the net already. "Why so small?!?!". What I think we are looking at is Stony Beach - Intel's 'Application Accelerator" seen here:

View Full Size

What further backs this theory is that you'll note the PCIe 3.0 x2 link of that product in the above roadmap, which couples nicely with the upper end limits seen in the 32GB product, which is clearly hitting a bandwidth limit at 1.6 GB/s, which is the typical max seen on a x2 PCIe 3.0 link.

View Full Size

Now with the capacity thing aside, there is another important thing to bring up. First gen XPoint dies are 128 Gbit, which works out to 16 GB. That means the product specs for the 16GB part are turning in those specs *WITH ONE DIE*. NAND based SSDs can only reach these sorts of figures by spreading the IO's across four, eight, or more dies operating in parallel. This is just one die, and it is nearly saturating two lanes of PCIe 3.0!

Another cool thing to note is that we don't typically get to know how well a single die of anything will perform. We always have to extrapolate backwards from the smaller capacities of SSDs, where the dies are the bottleneck instead of the interface to the host. Here we have the specs of one die of a product. Imagine what could be done with even wider interfaces and more dies!

View Full Size

XPoint fills the still relatively large performance gap between RAM and NAND, and does so while being non-volatile. There are good things on the horizon to be enabled by this technology, even if we first see it in smaller capacity products.


October 14, 2016 | 08:37 PM - Posted by Anonymous (not verified)

So~ I guess, I would like to see this in a 1/2 - 2 TB package on PCIE 16X. This would make a rather good cache drive for my NAS. :)

October 14, 2016 | 08:42 PM - Posted by Allyn Malventano

Yes, but remember that XPoint pricing will fall somewhere in between that of NAND and RAM, and RAM ain't cheap when you want 2TB of it!

October 14, 2016 | 08:52 PM - Posted by Anonymous (not verified)

So it's Intel's Optain versus Micron's QuantX! So the whole market will benifit form XPoint technology and not just Intel. I'd like to see AMD get some Radeon Pro Solid State Graphics (SSG) SKUs with maybe some QuantX instead of mostly NAND based Solid State Graphics storage in the current Radeon Pro offerings.

October 14, 2016 | 08:56 PM - Posted by Anonymous (not verified)

Edit: Optain
To: Optane

October 14, 2016 | 08:58 PM - Posted by CNote

How long until this ends up in hybrid drives?

October 14, 2016 | 09:08 PM - Posted by Allyn Malventano

If Intel does what I think they will with their RST software, I'd suspect one of these can make *any* primary storage into a faster hybrid solution. Remember, this is not the first time Intel has used a small fast drive to act as a cache for a larger, slower one.

October 14, 2016 | 11:14 PM - Posted by Paul A. Mitchell (not verified)

Allyn,

If I am reading one of those charts correctly,
the 32GB Optane only does sequential WRITEs at 500 MB/second?

Is that a typo, or am I missing something?

October 15, 2016 | 04:52 AM - Posted by Allyn Malventano

It's only two dies, and Intel may be limiting power draw on this product. It's something you'd do if it was meant for a mobile application. They may also be limiting speeds to increase endurance.

October 15, 2016 | 06:36 AM - Posted by John H (not verified)

This looks like a read only accelerator.. I don't see who would buy this being iops are held back to max 3x read of a mid range SSD like the 850 EVO due to pci-e x2 and only for 16-32gb. Seat of pants I don't think you'd notice any improvement. The cannonlake version of this might be a different story though.

Is there any roadmap for nvdimm yet?

October 15, 2016 | 06:37 AM - Posted by John H (not verified)

This looks like a read only accelerator.. I don't see who would buy this being iops are held back to max 3x read of a mid range SSD like the 850 EVO due to pci-e x2 and only for 16-32gb. Seat of pants I don't think you'd notice any improvement. The cannonlake version of this might be a different story though.

Is there any roadmap for nvdimm yet?

October 15, 2016 | 07:54 AM - Posted by So.... (not verified)

Umm.. doesn't anyone notice the multiple typo's in the fabricated product specs? "4KB2" "MB/s4" (twice) "Read3" "Write3." This is being reported as fact, but is obviously fiction. "we now have specs"

October 15, 2016 | 08:21 AM - Posted by Kraaketaer

An explanation for this might be that they're typos as in that they should have been superscript. There's a superscript "1" in the heading, which lines up with the numbers being 2-4, and it makes sense to clarify the circumstances of such specs on a slide - but these notes are obviously missing here. The placement of the numbers is also relatively consistent, which goes over well with this hypothesis. Might be a screenshot of an early presentation draft?

I'm not saying this is the case, just that it seems plausible.

October 15, 2016 | 11:48 AM - Posted by Paul A. Mitchell (not verified)

Yes, I thought those were footnote numbers,
but the footnotes were not visible.

October 15, 2016 | 08:27 AM - Posted by Kraaketaer

Am I right in reading that roadmap as Intel having practically abandoned MLC flash? This seems... weird. Sure, the 600p is a bit of an odd duck - better than SATA, at roughly the same price, but that's about it. Will there really be no MLC, mid-range version? Wow.

Also, what exactly is the difference between Brighton Beach and Stony Beach? Firmware? From the roadmap, the tech and interface is the same. If so, why does it take a full quarter to launch one after the other? What, if anything, differentiates "Optane Memory" from "Optane SSD" if they're both on PCIe NVME?

October 15, 2016 | 08:48 AM - Posted by Pexer (not verified)

How does Intel XPoint compare to current Samsung 960 Pro, Samsung PM961 and Samsung SM961?

October 15, 2016 | 10:07 AM - Posted by Anonymous (not verified)

More expensive, less performance, less storage than Samsung. No (X) point really.

October 15, 2016 | 10:08 AM - Posted by Anonymous (not verified)

BTW I am comparing products not theoretical limits of tech.

October 15, 2016 | 12:01 PM - Posted by Paul A. Mitchell (not verified)

A thousand times faster, a million times denser, and a zillion times longer (said Intel's PR department).

Next question, please.

p.s. More recently, Intel did publish measurements with a PCIe 3.0 NVMe add-in card populated with Optane, and I seem to recall that it saturated the bus during READs.

Allyn, if you're reading this, can you recall where that measurement was reported?

I'll see if I can find it, because I believe it comes close to answering your excellent question.

October 15, 2016 | 12:07 PM - Posted by Paul A. Mitchell (not verified)

I found it: Allyn wrote about that prototype here:

http://www.pcper.com/news/Storage/FMS-2016-Micron-QuantX-XPoint-Prototyp...

Note well in the graph how quickly this device ramped up to a steady flat line (the blue lines in that graph):

http://www.pcper.com/image/view/72918?return=node%2F65982

Quoting Allyn now:

"Thanks to the very low latency of XPoint, the QuantX solution sees very high IOPS performance at a very low queue depth, and the random performance very quickly scales to fully saturate PCIe 3.0 x4 with only four queued commands. Micron's own 9100 MAX SSD (reviewed here), requires QD=256 (64x increase) just to come close to this level of performance! At that same presentation, a PCIe 3.0 x8 QuantX device was able to double that throughput at QD=8, but what are these things going to look like?"

October 15, 2016 | 11:12 AM - Posted by Shambles (not verified)

At these capacities and the low cost of RAM wouldn't you be better off simply to expand your system memory by 16/32GB? Sure it would take a little while to fill your cache from a cold start but how often do computers get rebooted now anyways?

October 15, 2016 | 11:47 AM - Posted by Paul A. Mitchell (not verified)

> Sure it would take a little while to fill your cache from a cold start but how often do computers get rebooted now anyways?

We boot once per day, in the morning: our 13GB RamDisk Plus loads pretty fast by reading from a RAID-0 array of 4 SanDisk Extreme Pro SSDs.

One of the goals of non-volatile memory like Optane is to host the OS. Our version of Windows is also hosted by that RAID-0 array, and it's very fast now; but, it still goes thru a standard STARTUP sequence.

Properly configured, when an OS is hosted by an Optane-like memory subsystem, re-booting takes on a very different meaning -- something closer to INSTANT ON after a SHUTDOWN.

I can foresee a triple-channel RAM subsystem, like LGA1366, with 2 of those channels operating in dual-channel mode, and the third channel dedicated to a memory-resident OS.

Or, 6 DIMM slots total, with 4 operating in quad-channel mode, and the remaining 2 with Optane DIMMs operating in dual-channel mode and hosting the OS.

A while back, we proposed adding a "Format RAM" feature to BIOS and UEFI subsystems, to fresh-install an OS directly into a ramdisk.

A variation of that theme could cold start into a routine that restores a drive image to such a ramdisk, and then finishes booting normally by assigning the Windows C: drive letter to that ramdisk.

The latter concepts would work best with very large RAM subsystems e.g. current servers with 256GB to 1TB of RAM. But, as long as workstation-class machines have 32-to-64GB of RAM, these concepts would not work.

October 16, 2016 | 08:03 AM - Posted by Anonymous (not verified)

Good for mid range phones.

October 16, 2016 | 09:38 PM - Posted by Airwolf (not verified)

So it's significantly slower than a Samsung 960 Pro SSD? Am I missing the point?

October 17, 2016 | 04:48 AM - Posted by patrickjp93 (not verified)

This is a 1-die developer product. The best thing to compare it against is the PM761, not a full SSD with 16-32 dies.

October 20, 2016 | 09:06 PM - Posted by Anonymous (not verified)

You will see two packages in Stony Beach. So the math does not work here. One die is 16GB, but 2-package drive gives you only 16GB drive. You need to dig out the reason slowly by yourself.

October 24, 2016 | 03:13 PM - Posted by Paul A. Mitchell (not verified)

http://hexus.net/tech/news/industry/98392-intel-delays-introduction-3d-x...

October 30, 2016 | 09:40 PM - Posted by Paul A. Mitchell (not verified)

http://semiaccurate.com/2016/09/12/intels-xpoint-pretty-much-broken/