Friday, October 14th 2016

Intel 8000p - The First Consumer-Grade 3D Xpoint Products

The Intel-Micron joint collaboration in the development of what is promised to be the next step in storage technology is inching ever closer to reality. According to Bench Life, which published a leaked specifications list for the upcoming Intel Optane Memory products, the first application for consumer-grade 3D XPoint technology straddles the line between an SSD and system RAM. Intel calls it a "System Accelerator" solution, and it's meant to operate as an intermediate caching solution between a system's RAM and storage. Codenamed "Stony Beach", Intel's 8000p (and entire 3D XPoint-based products) support is still up in the air, but it's expected that only Kaby Lake and subsequent platforms will be compatible with the technology - which, if true, is sure to limit the product's market penetration.

The consumer products will initially sport capacities of either 16GB or 32GB, leveraging the NVMe protocol at PCIe Gen 3.0 x2 bandwidth in the M.2 form-factor. Mirroring NAND technology, the greater capacity solution will sport the highest performance: with the 16GB part coming in at 1400 MB/s read and 300 MB/s write speeds, against the 32 GB's 1600 MB/s and 400 MB/s, respectively. We see similar results in regards to IOPS, with the 16GB solution offering up to 285,000 read and 70,000 write operations per second, against the 32 GB's solution respective 300,000 read and 120,000 write. As usual with new technologies, expect all these metrics to only go up in time.
This in itself isn't much to talk about - the latest consumer-grade NAND-based SSD solutions already sport read and write speeds comparable to these Intel Optane drives (Samsung's 960 EVO springs to mind). That said, a product's performance can't be solely appraised by the way it looks on paper. As it is, the exact testing conditions that originated these performance ratings are unknown. More specifically, the metrics where 3D XPoint is supposed to excel versus NAND: latency access times, and speed at lower queue depths. And of course, one should not forget the lack of information on what amounts to the leading metric towards general consumer adoption: pricing.

True SSD-class products based on 3D XPoint are expected slightly after Intel's 8000p series, at the very end of 4Q 2016, and 1Q 2017, under the codenames "Mansion Beach" and "Brighton Beach" respectively. Details are close to nonexistent on these products at this time, save for the usage of the NVMe protocol and M.2 form-factor.

Source: Bench Life
Add your own comment

21 Comments on Intel 8000p - The First Consumer-Grade 3D Xpoint Products

#1
FR@NK
I dont see the point of this...it still uses the same old PCIe interface which isnt faster then current NVMe drives. It would have to be super cheap or else your money is better spent on more ram or a faster SSD.
Posted on Reply
#2
cdawall
where the hell are my stars
Very confused what this is supposed to offer over a normal NVMe SSD. Lower capacity and similar/lower IOP's to things on the market a year ago.
Posted on Reply
#3
Brusfantomet
So this is SSD cashes take 2 from Intel?

My aging XP941 is comparable to those sequential numbers, guess the kicker is the latency here (L4 memory?).
Posted on Reply
#4
FR@NK
Raevenlord said:

These are old slides from IDF15 back in Aug 2015 comparing Optane to intel's older 1st gen NVMe P3700 drives.
Posted on Reply
#5
Raevenlord
News Editor
FR@NK said:
These are old slides from IDF15 back in Aug 2015 comparing Optane to intel's older 1st gen NVMe P3700 drives.
Indeed they are. Unfortunately, for now, these are the best metrics we have with which to assess 3D Xpoint's supposed strenghts.
Posted on Reply
#6
FR@NK
Raevenlord said:
best metrics
They are so vague its hard to conclude its contrasting a two year old drive. It still a great article. Thank You.

I also notice they have the System Accelerator Gen 2 listed as BGA. I wonder if this will come already soldered on mainboards? That could get the price down enough where it will be useful. We will need more PCIe lanes!
Posted on Reply
#7
Steevo
I'm waiting to see what HBM2 on a motherboard will do for us, 16/32GB cache between RAM and SSD.
Posted on Reply
#8
slozomby
or you could just buy an ssd backed raid controller. this is not new tech its just adapted to use the m2 slot.
the concept makes a lot of sense when dealing with huge io, like fusio-io (now part of western digital after they bought sandisk) was doing 5 yrs ago with sans, or when you have spinning disk(s)/raid. personally I don't see much point in caching ssd's with a slightly faster ssd.
Posted on Reply
#9
Hood
Steevo said:
I'm waiting to see what HBM2 on a motherboard will do for us, 16/32GB cache between RAM and SSD.
This looks like about the same thing. As I understand the concept, the point of this is to accelerate system response when your main (or only) drive is a cheap SATA SSD or even a hard drive(?). So this should make any crappy system feel more like it has a fast NVMe boot drive, I guess. To make sense, this has to be fairly cheap, like $30 and $60 for the larger one. This could be revolutionary for the average user, even though it sounds boring to those of us with 950 Pros or Intel 750 drives. I was really hoping that XPoint drives would run from unused RAM slots, so it would be more universally compatible. Requiring an M.2 slot severely limits the adoption of this technology.
Posted on Reply
#10
slozomby
Hood said:
Requiring an M.2 slot severely limits the adoption of this technology.
not just an m2 slot. but also kaby lake+

"but it's expected that only Kaby Lake and subsequent platforms will be compatible with the technology"
Posted on Reply
#11
alucasa
Me thinks Intel misses the early SSD era where they could charge a fortune for a 32gb SSD.
Posted on Reply
#12
SAL9000
Everyone here needs to read more about what Intel has in mind for this technology. Intel actually suggested to get the most out of 3D XPoint, it needs a completely new storage interface and they weren't talking about PCIe. If anyone here has been reading tech news that doesn't concern the PC, there is a new standard that is in development. It is a memory based fabric. The Gen-Z interface which is designed for storage class memories which basically is designed to communicate through a systems memory instead of PCIe or a processor bus. There is also Open CAPI which is a processor based interface but can be used for storage. It is a 150-160GB interface. That kind of speed would be amazing.
PC's no longer have the mind share at Intel, data centers and supercomputers are higher on their list of priorities.
Posted on Reply
#13
slozomby
SAL9000 said:
Everyone here needs to read more about what Intel has in mind for this technology. Intel actually suggested to get the most out of 3D XPoint, it needs a completely new storage interface and they weren't talking about PCIe. If anyone here has been reading tech news that doesn't concern the PC, there is a new standard that is in development. It is a memory based fabric. The Gen-Z interface which is designed for storage class memories which basically is designed to communicate through a systems memory instead of PCIe or a processor bus. There is also Open CAPI which is a processor based interface but can be used for storage. It is a 150-160GB interface. That kind of speed would be amazing.
PC's no longer have the mind share at Intel, data centers and supercomputers are higher on their list of priorities.
intel isn't part of the gen-z consortium
Posted on Reply
#14
SAL9000
You are correct but on EET.com there is more about what I am referring to http://www.eetimes.com/document.asp?doc_id=1330609

This is what I should have said earlier (Quoted from EET website)
Similarly, work on Gen-Z was fueled by Intel’s suggestion at its annual developer conference last year that its 3D Xpoint memories will need a new systems interface. Intel has yet to detail its plans for a 3D Xpoint interface. So far it is saying it will use a combination of its proprietary Quick Path Interconnect and PCIe to link its Xeon processors and Altera FPGAs

Obviously these new types of memories can saturate PCIe quite easily.
Posted on Reply
#15
esrever
SAL9000 said:

Obviously these new types of memories can saturate PCIe quite easily.
If they can, they aren't even close yet.
Posted on Reply
#16
SAL9000
They are close, Seagate came out with a card that has 4 - M.2 cards on it, the 8TB Nytro XP7200 NVMe SSD. and it has a 16x PCIe 3.0 connector. So imagine future storage technologies which are even faster and the need for a higher bandwidth link is easily understood.
Posted on Reply
#17
bug
Steevo said:
I'm waiting to see what HBM2 on a motherboard will do for us, 16/32GB cache between RAM and SSD.
Even as a cache, there are still notable differences, since XPoint is non volatile.
Either way, this is so early in the game, speculation is pointless.
Posted on Reply
#18
Frick
Fishfaced Nincompoop
Hood said:
This looks like about the same thing. As I understand the concept, the point of this is to accelerate system response when your main (or only) drive is a cheap SATA SSD or even a hard drive(?). So this should make any crappy system feel more like it has a fast NVMe boot drive, I guess. To make sense, this has to be fairly cheap, like $30 and $60 for the larger one. This could be revolutionary for the average user, even though it sounds boring to those of us with 950 Pros or Intel 750 drives. I was really hoping that XPoint drives would run from unused RAM slots, so it would be more universally compatible. Requiring an M.2 slot severely limits the adoption of this technology.
Wait, is there a tangible difference in general response time between NVMe amd ACHI drives? I cannot fathom that being useful, unless for very, very specific tasks.

But as already pointed out, XPoint is supposed to be more. I remember when they announced it, I just didn't expect it to be a first gen product like this. Waiting for them busses.
Posted on Reply
#19
DeathtoGnomes
between this and Gen-Z, or whatever, maybe we will get PCi interface replacement for video cards and then maybe they will build cards right side up finally! :laugh:

Oh OH and wireless PSU!!:pimp:
Posted on Reply
#20
Hood
Frick said:
Wait, is there a tangible difference in general response time between NVMe amd ACHI drives?
Yes there is, my 400 GB Intel 750 NVMe boot drive does most tasks noticeably faster, and some of them much faster (things like Windows Update, Windows Disk Cleanup, virus/malware scans). I think it's the best $300 I spent on my system. As for XPoint, I'll wait for the reviews - this early implementation may not help much...
Posted on Reply
#21
theeldest
The difference between Optane and other NVMe drives:

Optane hits full performance at low queue depths (as is commonly seen in client workloads).

NVMe drives hit full performance at QD64+

Now we just need Microsoft to add Storage Spaces tiering to Win 10 without needing powershell and there's an effective way to use this technology right away.
Posted on Reply
Add your own comment