Discussion Optane Client product current and future

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Dayman1225

Golden Member
Aug 14, 2017
1,152
973
146
Intel Optane SSD 800P looks to be Brighton Beach. My current guess is that some of the Gen 2 features outlined in the older roadmap are going to be fit to be used with Brighton Beach. Also, M10 doesn't look like it'll be a BGA device, but M.2 based one. Presumably we'll see BGA variants, but won't be the only ones for M10.

The performance figures also seem suspiciously similar to the early leaked Optane Memory ones. Like this one: https://eteknix-eteknixltd.netdna-s...16/10/Intel-Optane-Memory-8000p-3D-XPoint.jpg



Dayman1225: How is that relevant to this thread?

Omg.... I thought I posted in the QLC SSD thread MY BAD!
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
120GB 800P should be $139. 64GB M10 should be $80.

The preliminary pricing for 800P seems to be:

$109 for 60GB
$199 for 120GB

Pricing on the current 32GB Optane has dropped to $59.99 free shipping at B&H:

https://www.bhphotovideo.com/c/prod...empek1w032gaxt_32gb_optane_memory_module.html

And $60.89 free shipping at Antonline.com:

http://www.antonline.com/Intel/Comp...Drives/Hard_Drives+Solid_State_Drives/1292690

So maybe $80 for a 60GB 800p SSD isn't really that far away.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
^^^^^ If it is true that the 32GB Optane Memory (basically) becomes the 120GB Intel 800p SSD via better dies and four bits per cell.....then it will be interesting to compare to 128GB NAND SSDs (Including Intel 760p) on various parameters. (I am guessing latency will be better, but Sequential write will be less (not sure how much though in each of these two categories. Then there is idle and active power consumption to consider (very important for laptops) as well as endurance and data retention)
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
With Optane part numbers, "MEM" implies SLC and "SSD" implies QLC.

I'm sorry. But he's wrong. He writes articles that reads like fiction. Some would say "fan" fiction. Fictions(and fan fictions) are pleasant to read. However, they are very, very far from reality.

Refer back to what I said on one of my original posts. The second one said:

That means 32GB packages are available. 3D XPoint dies are rather large for a 128Gbit(16GB) version so that's going to remain the largest one. 32GB means a multi-die package. The 900P will be available in capacities of 960GB, and 1.5TB in the future.

How do you think they achieve capacities on the larger 900P drives? By using QLC? That's ridiculous. The same method 900P uses to achieve higher capacities are what 800P and M10 will use to achieve its high capacities.

This is one package, but multiple dies inside. It doesn't increase bandwidth or production costs per die, its meant for more compact configurations. This is how they achieve 128GB(1Tbit) in one DRAM module with 8Gbit DDR dies. This is how they achieve BGA single chip SSDs that have 256GB+ sizes.
 
Last edited:

cbn

Lifer
Mar 27, 2009
12,968
221
106
How do you think they achieve capacities on the larger 900P drives? By using QLC? That's ridiculous. The same method 900P uses to achieve higher capacities are what 800P and M10 will use to achieve its high capacities.

Yep, The 480GB version of the 900P has some packages with more than one die.....and the upcoming 1.5TB version would have four dies per package (assuming the layout is the same).

However, when I look at the picture (in the article linked in post #28) It looks to me that the packages on the 32GB Optane Memory and 120GB 800p SSD are the same height:

4649571-15192426129687078.jpg



With that noted, the preliminary pricing you mentioned of $199 for 120GB 800p in post #22 seems indicative of a drive with four 128gb die per package, rather than a drive with four bits per cell, but one die per package.
 
Last edited:

18436572

Junior Member
Mar 2, 2017
9
2
41
How do you think they achieve capacities on the larger 900P drives? By using QLC? That's ridiculous.

So, you're saying that it is impossible for anyone to increase capacity by stuffing more than one bit into a cell? Because that's what you're saying. When the first MLC drive came along, was your argument that "because we've always done it the other way, this must be a lie"?

I look at the photo of the 32GB part next to the 120GB part and the Optane chips have the same part number and everything. It seems pretty convincing but I guess that we're about to find out. When do these things ship?

EDIT - the 900P has a controller which is marked "SLL3D", which implies that it is an SLC controller. This seems to imply that more bits are coming.

EDIT2 - here's the detailed photo of 32GB next to the 120GB.
 
Last edited:

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
So, you're saying that it is impossible for anyone to increase capacity by stuffing more than one bit into a cell? Because that's what you're saying. When the first MLC drive came along, was your argument that "because we've always done it the other way, this must be a lie"?

No, you are going way ahead of yourself, because I never said that.

The original doubt was based on the fact that 900P does not resort to multi-bits to achieve capacity, but more dies per package. Since the chip names are different on there, and its similar for the 800P, that is out the window.

800P is coming in 30GB, 60GB and 120GB capacities. The suggestion means Intel would use single cell for 30GB, dual cell for 60GB, and quad cell for 120GB. Does that make sense to you?

His speculation that error correcting bits are the reason for odd capacities can be countered by the fact that retailers list in 30, 60, and 120GB capacities. The M10, is listed at 32GB and 64GB, but not 800P.

It looks to me that the packages on the 32GB Optane Memory and 120GB 800p SSD are the same height:

You can't tell by the physical dimensions. The thickness is in microns. That's how they stack 50+ layers on vertical NAND without affecting package size.
 

Billy Tallis

Senior member
Aug 4, 2015
293
146
116
I think the author of that article missed the fact that even last year's Optane Memory 16GB and 32GB drives internally identify themselves as 800p. There's no reason to suspect these larger capacities are anything different from a normal capacity increase with more dies, especially since they're using the same controller. The 16GB packages seen at CES probably just mean that Intel was using 32GB drives for some of their demo units.

You can't tell by the physical dimensions. The thickness is in microns. That's how they stack 50+ layers on vertical NAND without affecting package size.

Extra layers of 3D NAND on a single die are far thinner than extra dies within a package. It is sometimes possible to spot a height difference between single-die packages and 16-die packages with the naked eye, but it is also common to find that single-die packages and 4-die packages are the same height.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
Extra layers of 3D NAND on a single die are far thinner than extra dies within a package. It is sometimes possible to spot a height difference between single-die packages and 16-die packages with the naked eye, but it is also common to find that single-die packages and 4-die packages are the same height.

Yep, this webpage does list a package with 16 dies being thicker than one with eight dies.

P.S. I didn't realize NAND wafers were so thin. According this (on page 18) the wafer thickness for NAND is only 60um. (This, in contrast, to 300mm wafers which are 775um thick)

So if 3DXpoint follows the same thickness as NAND I could see package thickness being made the same whether there was 1 or 4 dies in it.
 
Last edited:

cbn

Lifer
Mar 27, 2009
12,968
221
106
800P is coming in 30GB, 60GB and 120GB capacities.

I didn't realize 30GB 800p existed till I Google searched for it. (I thought only 60GB and 120GB existed).

The presence of the 30GB capacity rules out any chance of QLC being used. (re: one 3DXpoint die with QLC would be 512Gb (64GB))
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
So I guess new Optane and 300-series chipsets aren't launching today.

I did just listen to Rob Crooke's speech at Morgan Stanley TMT conference. For 3D NAND, he emphasized the role of the technology is foremost achieving most bits/mm2. They seem to be doing that well with the recent SSDs like the 760p, achieving excellent performance at low cost, and low power. For 3D XPoint based products, for this year its mostly about proliferating the products to meet cost targets better, but with the primary goal of improving performance.

Density wise, he was talking about QLC NAND, with greater vertical stacks and even the traditional shrink to do so. With Optane and density it was about increasing the current 2 stack layer into higher ones, and traditional shrink. There were analyzations done that said 3D XPoint gets better with lithography shrink and its metrics improve. Like in performance and durability.

Optane Memory 16GB and 32GB drives internally identify themselves as 800p.

Billy, weren't they called 8000p? The early leaks also had them with higher specs. The write was almost double. And thanks for your thoughts on the matter.

There's a fairly recent PCN that states some PCB size changes for the 800P models. No way to tell why it was done. I'm still of the belief that they'll release it all with 300-series chipsets, because it makes sense.

DIMMS reportedly uses 15-18w of power depending on the workload and because of their dimensions require reworked mobos.

I've seen that figure before. I think they'll launch Cascade Lake in early Q3, but announce Apache Pass DIMMs coming later in the year. That allows unnecessary delaying of the platform, without having to risk the DIMMs falling short of expectations because it launched too early. Exciting tech for sure, but lots of hurdles to clear.
 
Last edited:
  • Like
Reactions: Dayman1225

cbn

Lifer
Mar 27, 2009
12,968
221
106
Thinking ahead about the future of client I am wondering how little DRAM a NVDIMM-P (with Optane) could have?

Will it be possible to have as little as 2GB DDR per stick? 4GB DDR per stick? Or will the minimum be 8GB per stick when DDR5 arrives?

I bring this up because I can imagine some scenarios where two NVDIMM-P DIMMs (with 2GB DRAM (or 4GB DRAM) + Optane each) could replace two 8GB DRAM DIMMs plus M.2 NVMe SSD.

I guess it depends on how well DDR4 or DDR5 scales with a lower amount of dies?

EDIT: Here is a DDR4 2400 SO-DIMM with 2GB capacity ( I didn't even know these existed). I wonder if we could even get 2GB at DDR4 3200 speed eventually* ? (This, of course, on a NVDIMM-P)

*though I think even DDR4 2400 would work fine.

P.S. For anyone who hasn't heard of NVDIMM-P Anandtech has an article with info here.

Another specification that JEDEC plans to finalize in 2018 is the NVDIMM-P that will enable high-capacity memory modules featuring persistent memory (flash, 3D XPoint, new types of storage-class memory, etc.) and DRAM. The capacity of today’s NVDIMM-Ns is limited to the capacity of regular server DRAM modules, but the NVDIMM-P promises to change that and increase capacities of modules to hundreds of GBs or even to TBs. The NVDIMM-P is currently a work in progress and we are going to learn more about the tech in June.

nvdimm_2_575px.png
 
Last edited:

18436572

Junior Member
Mar 2, 2017
9
2
41
The presence of the 30GB capacity rules out any chance of QLC being used. (re: one 3DXpoint die with QLC would be 512Gb (64GB))

Why couldn't that just be a blem chip with MLC? while the 60GB chip is perfect chip with QLC?

I'm very curious now that I can see that the current 16 and 32GB Optane products identify themselves as "8000p", as was pointed out by IntelUser2000 above. To me, this says that they are the higher-performing product with the 800p coming in with lower performance. This, combined with the photo of the new 800p, does really make it look like the 800p is storing multiple bits per cell - either MLC, TLC or QLC with some combination of wear-leveling and ECC.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Who said that 800P is lower performance than current Optane Memory?

The read and write sequentials are higher than Optane Memory, the latter twice as much. That's just based on preliminary samples.

The craziest expectation is that 1-bit/2-bit/4-bit cells would be used in the same product line. Also while its technically not impossible as stated even by IMFT themselves, they also stated early on that first few generations won't be going towards multi-bit versions. They also mention about 8-stack versions, which would allow 128GB per chip.

http://www.tomshardware.com/reviews/3d-xpoint-guide,4747-3.html

Rob Crooke, the manager for Intel's NVMe said this week that density for 3D XPoint products are about using lithography shrinks and stacking(die stacking like 3D NAND), with the current focus on improving Optane performance.

That's indeed what they are doing with 900P and P4800X. The 1.5TB versions require 4 die stacked chips to achieve it. They can use the same thing on the 120GB capacities. With 30GB capacities featuring two chips having only 1 die. With the M10 version, which is supposedly mobile-focused the capacities available range from 16 to 64GB.
 
  • Like
Reactions: Dayman1225

18436572

Junior Member
Mar 2, 2017
9
2
41
Who said that 800P is lower performance than current Optane Memory?

Two things: [1] the naming convention (8000p versus 800p) and [2] the part numbering ("MEM" versus "SSD"). If the 800p was to have better performance than the 8000p, then they'd be marketing it as such (e.g. "8100p").

The read and write sequentials are higher than Optane Memory, the latter twice as much. That's just based on preliminary samples.

Aside from write endurance, the specs are not currently advertised. Let me know if you've found them somewhere so that I can inspect for myself.

The craziest expectation is that 1-bit/2-bit/4-bit cells would be used in the same product line.

They aren't the same product line - otherwise, they'd have the same model and part numbering conventions.

Without worrying about that other stuff above (which will become public soon enough), how do you reconcile the identical part numbers on the Optane chips between the 32GB 8000p "mem" product versus the 800p "ssd" product? If you need to know anything about a fab, know that the package markings are automated. A production line cannot be changed without the part number changing. You can't expect to see the same part number on a single 128Gb Optane chip and a stack of 128Gb Optane chips. If this was allowed to happen, the manufacturer would lose their ISO certification, instantly. The sheer numbers are just too big.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
IntelUser2000 said:
The read and write sequentials are higher than Optane Memory, the latter twice as much. That's just based on preliminary samples.

Aside from write endurance, the specs are not currently advertised. Let me know if you've found them somewhere so that I can inspect for myself.

According the Anandtech article endurance of 800p is about half per GB compared to the 16GB and 32GB Optane Memory.....but is that enough of a reduction to believe it could be QLC? ( Using NAND as an example P/E changed from 3000 to 750 just going from MLC to TLC at the 2Xnm node. Same proportional relationship for 3xnm and 5xnm as well. But this isn't MLC to TLC transition, but rather SLC to QLC....with the idea thrown in that die quality might have also improved)

The 16 GB and 32 GB Optane Memory are both rated for a write endurance of 182.5 TB over five years, equivalent to 100 GB per day. The new 58 GB and 118 GB Optane SSD 800p drives are both rated for 200 GB per day over the same five year warranty period.

Furthermore, how does a person explain the presence of a 30GB 800p SSD if Intel is using QLC? Smaller than normal die would be the only explanation that I can think of (and judging by package size on the 60GB and 120GB models Intel would not be using a small die on those)
 
Last edited:

cbn

Lifer
Mar 27, 2009
12,968
221
106
About the only way I could imagine a hypothetical 60GB and 120GB QLC Optane having only a hypothetical 50% reduction in endurance coupled to 550-600 MB/s write (as specified by IntelUser2000 in post #22) would be if....

1.) The Sequential write per die was slowed down compared to normal hypothetical QLC Optane would do. This to increase endurance to ~half the level of SLC Optane.

2.) The dies were made smaller. This to increase parallelism per GB and regain overall write performance lost by the running each die slower.

But again, I don't see the 800p using small 3DXpoint dies based on the package size seen in the pictures.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Aside from write endurance, the specs are not currently advertised. Let me know if you've found them somewhere so that I can inspect for myself.

Of course:

http://ranker.sisoftware.net/top_ru...f5c8f8deb68bbb9de5d8e8ceabcef3c3e596ab9b&l=en

The page shown above has a drive named INTEL SSDPED1D280GA. That's the 280GB 900P drive. So most of the first page numbers are various capacities of the 900P drives, and get roughly in line with the write spec, which is 2GB/s. You will also find lower numbers though on following pages, with one being around 1600MB/s.

You can also find INTEL SSDPEK1W060GA. The top result gets 588MB/s. That's the 800P. Generally the 800P results end up at 500-600MB/s.

Without worrying about that other stuff above (which will become public soon enough), how do you reconcile the identical part numbers on the Optane chips between the 32GB 8000p "mem" product versus the 800p "ssd" product?

You are right about this, and the confusion arose from me forgetting the 2 die stacks had different markings on the 900P drive.

But its easily explained. As Billy said earlier, the media samples floating around could just use the 30GB version. That requires same number of chips and same number of dies per chip as the current 32GB Optane Memory device.

Your assertion though requires a more extraordinary claim. Something they are not doing, and never stated would do at this early point in manufacturing.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Information updated.

https://www.pcper.com/reviews/Stora...8GB-and-RAID-Review-3D-XPoint-Goes-Mainstream

My predictions were correct. The 800P uses 2 and 4 stack dies for the higher capacities, not MLC or QLC.

800P is also Brighton Beach, not Carson Beach.

The pricing is disappointing. In particular because the higher end 900P was priced pretty attractively. It should have been $99/$169 at most. Reviews do state it can be used in place of Optane Memory as a caching device though, and the 58GB version might work quite well due to being big enough.

The 800P uses NVMe power state, called L1.2. It does not seem to be as comprehensive as latest NVMe SSDs, and thus fall under lower battery life device for Notebooks. Though that might be fine, as the yet to be released M10 device is supposed to be targeted at mobile.
 
Last edited:
  • Like
Reactions: cbn

Billy Tallis

Senior member
Aug 4, 2015
293
146
116
  • Like
Reactions: cbn

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
TH result is showing battery life on the range of lower battery life HDD and NVMe SSD drives. 760p NAND SSD on the other hand seems like a top drive to get for notebooks.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
With Carson Beach being a new controller (and available in BGA SSD according to that old roadmap) maybe that will be the controller with the low power to match the best NAND based NVMe SSDs?

P.S. My guess is that Carson Beach will have four channels (slotting in between the two channel of Stony Beach/Brighton Beach the and seven channels of Mansion Beach). If true then maybe 120GB Carson Beach will have 1200 MB/s Sequential write.

One more thing, I noticed Brighton Beach is NVMe 1.1.....but I wonder if Carson Beach will have NVMe 1.2 and support for Host Memory Buffer? If so, will we see any improvement using System RAM? A tiny improvement?