Discussion Optane Client product current and future

[nosirrahx]

Alder Stream next generation Optane SSD to have "3 times the throughput and 4 times lower application latency".

If that means 6GB/s read/write they will have quite a nice drive out there. Don't know how they are getting the latency that low though. Maybe they're talking about comparisons in specific usage scenarios.

I kind of want to test these new drives, I wonder how long I will need to wait.

 

[IntelUser2000]

Looks like Optane DIMMs are coming for Tigerlake H.

This year we'll likely see a successor to Optane H10 hybrid drive.

 

[nosirrahx]

Looks like Optane DIMMs are coming for Tigerlake H.

This year we'll likely see a successor to Optane H10 hybrid drive.

I wonder if the cancellation of the M15 and 815P has more to do with the legal trouble Intel and Micron are having over 3D Xpoint royalties than it does with their claim of low demand. The M10 and 800p drive also had almost no demand.

I am super interested in getting my hands on a Micron X100 drive, if it ever materializes.

 

[IntelUser2000]

The M10 and 800p drive also had almost no demand.

I think that's different. M10s were actually quite often used as caching for retail systems. Most of them were the low capacity 16 and 32GB versions and they were just called "Optane Memory", and unless you tried, you couldn't figure out it was M10.

I agree 800P was crap. That should have been clear from the start. Zero price cuts since last generation. The higher end 900P was sold for less per GB than the 800P. No wonder no one bought it. I thought 900P was badly priced.

That said, one generation of bad selling may not have been enough, and 800P sealed the coffin. Plus SSD prices were crashing in 2018 and 2019, rendering caches to an extreme niche. Second year that is equal to Intel throwing away money.

I am super interested in getting my hands on a Micron X100 drive, if it ever materializes.

Throughput and TDP are interconnected for 3DXP devices. If its 18W for 900P, it may be several times that for the X100.

The pictures had an auxiliary power connector. The possibility of going over the PCI-Express' 75W limit is there!

I suspect the balance is same for Alder Stream. If it has high throughput expect high TDP, possibly reaching 100W!

That's why ultimately Optane for SSDs are a short term deal, because it keeps the disadvantage of it being very expensive while the interface severely bottlenecks the medium.

 

[Edrick]

That's why ultimately Optane for SSDs are a short term deal, because it keeps the disadvantage of it being very expensive while the interface severely bottlenecks the medium.

The 900p is a killer product. Yes it is expensive, but there is always a premium for the best. As for interface, I have no idea why they crippled the first series with 4x PCIe 3.0 lanes. The second generation with 8x PCIe 4.0 lanes will be a monster. If they can get costs down, there there is no reason to think these can not be a high end NAND replacement.

I doubt they will surpass 75W, but even so, are HEDT customers going to care about that when they are rocking 250W GPUs and 300W CPUs?

 

[Edrick]

And does anyone know when the Micron X100 drive will be available?

 

[nosirrahx]

And does anyone know when the Micron X100 drive will be available?

I keep checking and not seeing anything. I am pretty sure it will be the OS/apps drive in my workstation when I rebuild but so far I am not seeing anything.

I wonder if there is some kind of deal in place where Micron and Intel will release their next gen 3D Xpoint products at the same time?

 

[Billy Tallis]

And does anyone know when the Micron X100 drive will be available?

I keep checking and not seeing anything. I am pretty sure it will be the OS/apps drive in my workstation when I rebuild but so far I am not seeing anything.

I wonder if there is some kind of deal in place where Micron and Intel will release their next gen 3D Xpoint products at the same time?

The X100 is not going to get a wide release. It's a product only for major partners and will not be sold publicly through the retailers and distributors that carry Micron's usual enterprise/datacenter SSDs. Also, the X100 as a product isn't managed by Micron's Storage Business Unit.

 

[Edrick]

The X100 is not going to get a wide release. It's a product only for major partners and will not be sold publicly through the retailers and distributors that carry Micron's usual enterprise/datacenter SSDs. Also, the X100 as a product isn't managed by Micron's Storage Business Unit.

Well that sucks. Thank you for the info.

 

[IntelUser2000]

The 900p is a killer product. Yes it is expensive, but there is always a premium for the best. As for interface, I have no idea why they crippled the first series with 4x PCIe 3.0 lanes. The second generation with 8x PCIe 4.0 lanes will be a monster. If they can get costs down, there there is no reason to think these can not be a high end NAND replacement.

I'm not saying that the 900P isn't nice, but I'm trying to look at in terms of Intel's perspective and my wishes that they'll be successful with it. That means financial success if its not clear.

The simple reality is that cost of production is too high right now for them to be any sort of a NAND replacement anytime soon. It's said that the 900P is not making a lot of money for them so that should give you a rough idea how it is.

One of the biggest reason for that in any product is due to lack of volume. Even NAND went through the same thing. Unless they get into within an order of magnitude of volume it has absolutely no chance. And just an order of magnitude is a tall order right now.

The advantage of Optane is latency, which allows for potential DRAM replacement/complement. PCI Express severely bottlenecks the media, which is why its nowhere near ideal. The cost of the media is being "wasted" by the interface. It limits the pricing, which limits revenues for Intel and ultimately makes small sense for them. It's same for the customers.

If it was on DRAM though, it can be 30-50x faster in terms of latency over the SSD Optane. The differences are between enabling new usage application versus continuing the same thing we have been doing for decades.

I assure you a 64GB DIMM Optane would find far many users at the same price because it can work as slow memory than a cache or a drive which is a glorified SSD. Even for the cache versions if they brought the server Memory Drive Technology to PCs it would make more sense being called Optane "Memory".

One can imagine say Optane Client PMM being used for DRAM addition applications and on top of that Intel can use software/drivers to cache storage to load things faster. You might be able to allocate 32GB portion for slow memory while the rest can be set for a super-fast cache. DIMM opens up opportunities that will justify its cost.

Speaking of transfer rates, super fast transfers on SSDs are totally wasted. Storage is called "cold data" because its accessed seldomly by the CPU. DRAM is called "hot data". DIMM Optane will be far "warmer" and thus its transfer rates taken advantage of since it'll be used for compute. This is part of the reason why I believe SSD benchmarks are so useless on client. How can you test it other than say it feels fast and does(seldomly done) file transfers faster?

I doubt they will surpass 75W, but even so, are HEDT customers going to care about that when they are rocking 250W GPUs and 300W CPUs?

It again has to do with reaching volume sales. 9xx series are 18W because that's what the client SSDs are at. Reach 75-90W then you start to lose quite a lot of them. Forget about having an NVMe version when SSDs are becoming a challenge to cool at 20W already.

 

[IntelUser2000]

Optane DC Persistent Memory 200 series "Barlow Pass" launched!

There are conflicting infos but here's what we know.
-2nd generation 3D XPoint with 4 layers.
-Rated Read bandwidth for the fastest device improves by 19%, and write bandwidth by 37%. Correspondingly, the endurance is at 497PBW.
-Capacities are same at 128GB/256GB/512GB

 

[IntelUser2000]

Correction on Barlow Pass. It was natural to assume it would use 2nd generation 3D XPoint, but it doesn't. Some articles say its likely the reason for the small bandwidth improvement.

Alder Stream is confirmed to use 2nd generation 3D XPoint. That means Crow Pass, the 3rd generation or, Optane DC 300 PM will use 2nd generation 3D XPoint, along with DDR5 compatibility.

Some analysis is indicating that Intel is possibly no longer losing money on it. That is good news!

Is Intel Ready To Make Money On Optane? - Business Quick Magazine

Jumping to the next level getty During Intel’s last earnings announcement the company provided information to indicate that 3D XPoint…

bq-magazine.com

My analysis added to the graph from the article:


I said from the beginning that it wasn't SSDs, nor the client cache modules that would make the difference, but the DIMM versions. The cache modules have to be cheap and it depends on how long the HDDs would stay relevant. SSDs(and caches) also significantly hinder the one strength of Optane, and that's super low latencies.

DIMMs suffer from no such issue. Latencies are indeed pretty much 1000x better than NAND(over 500x versus best NVMe, 100us vs 180ns) allowing for entirely new usage scenarios, or simple expansion of RAM. It also allows higher price due to higher production costs to be justified.

According to their analysis, its in 2023 where byte-shipments of Optane and DRAM will be within an order of magnitude. That should also be the year where it really shows its advantages against DRAM.

 

[Edrick]

Anyone have an update on Alder Stream?

 

[nosirrahx]

Anyone have an update on Alder Stream?

Last news I head was about the lawsuit going forward against Intel and Micron about 3DXpoint.

3D XPoint patent suit against Micron and Intel is allowed to proceed – Blocks and Files

A US lawsuit filed two years ago by the liquidator of a bankrupt patent company against Intel and Micron has been allowed to proceed. The ruling issued on October 1 by Judge Thomas J. Tucker, exposes Intel to claims that its has no right to its core 3D XPoint technology. ECDL Trust accuses...

blocksandfiles.com

 

[kurosaki]

Last news I head was about the lawsuit going forward against Intel and Micron about 3DXpoint.

3D XPoint patent suit against Micron and Intel is allowed to proceed – Blocks and Files

A US lawsuit filed two years ago by the liquidator of a bankrupt patent company against Intel and Micron has been allowed to proceed. The ruling issued on October 1 by Judge Thomas J. Tucker, exposes Intel to claims that its has no right to its core 3D XPoint technology. ECDL Trust accuses...

blocksandfiles.com

So this is why it's so expensive!

 

[nosirrahx]

So this is why it's so expensive!

Actually I think this has more to do with things getting delayed and/or canceled. If they came up with an amazing and widely adopted use case for 3DXPoint right now, it could bite them in the ass. I kind of wanted to get my hands on that X100 drive from Micron but I still cant find anything indicating that it ever existed.

 

[IntelUser2000]

So this is why it's so expensive!

No, LOW volume is why its so expensive.

Yes it seems redundant to say low volume makes it expensive, which means low sales, meaning its hard to get out of the loop.

If you go to my post above, there's indication that the server persistent memory has been selling enough to lower costs to start selling it profitably.

I was always adamant from the beginning it was born for the DIMM versions, because a new, expensive technology can only justify itself if the performance is super high.

As good as the SSD versions are, its just too much of a limiter. Alder Stream doesn't change anything fundamentally, because the latency is still going to be in the ~10us range.

 

[kurosaki]

it seems redundant to say low volume makes it expensive, which means low sales, meaning its hard to get out of the loop.

If you go to my post above, there's indication that the server persistent memory has been selling enough to lower costs to start selling it profitably.

I was always adamant from the beginning it was born for the DIMM versions, because a new, expensive technology can only justify itself if the performance is super high.

As good as the SSD versions are, its just too much of a limiter. Alder Stream doesn't change anything fundamentally, because the latency is still going

Oh, sorry, missed the "" at the end. Wasn't that serious in my statement.
But the issue of the hen and egg persists. If you want to sell a lot of Optane disks, the prices has to go down, but we cant sell the disks to an all to great loss either. I'm just looking forward to the day we all have PCI-e gen 6, 15 TB Optane boot drives at a reasonable cost. Or rather, I hope under my lifetime I will be able to witness such a feat.

 

[Edrick]

Oh, sorry, missed the "" at the end. Wasn't that serious in my statement.
But the issue of the hen and egg persists. If you want to sell a lot of Optane disks, the prices has to go down, but we cant sell the disks to an all to great loss either. I'm just looking forward to the day we all have PCI-e gen 6, 15 TB Optane boot drives at a reasonable cost. Or rather, I hope under my lifetime I will be able to witness such a feat.

I have been enjoying my Optane drive for a few years now. When you consider Optane drives are far cheaper than SSD's were back when they were first released, its not too bad actually. No need to wait for the future! Then again, I am getting very impatient waiting for Alder Stream and/or the X100.

 

[IntelUser2000]

@Edrick In 2009 I paid $800 cdn for the 80GB X25-M.

I'm hesistant at buying 900P 280GB for $500.

Because the Optane drive(SSD) doesn't change anything fundamentally. The X25-M did, which is make computers viable to use, and not behind freaking Smartphones in responsiveness. The 10us latency, great QD1 figures, and consistency are all great features, but too little for the price, especially considering the latency and QD1 can't be fully taken advantage of, because the way applications are coded.

Also, the technology is seriously capped by PCI Express, and the coding for storage. The DIMMs are 30x better in latency and can displace DRAM in majority of client scenarios.

The thing that has a chance to change things fundamentally is the DIMM version. There's no doubt about it. Yes, software has to catch up, but even then.

The DIMM versions have the potential for Averages Joes to get interested in buying a new computer, and actually benefit them which hasn't happened in perhaps 15 years.

It'll take simple as Windows supporting an Optane DIMM specific hibernate feature so it can be a truly zero power "sleep" mode.

 

[nosirrahx]

@Edrick In 2009 I paid $800 cdn for the 80GB X25-M.

I'm hesistant at buying 900P 280GB for $500.

Because the Optane drive(SSD) doesn't change anything fundamentally. The X25-M did, which is make computers viable to use, and not behind freaking Smartphones in responsiveness. The 10us latency, great QD1 figures, and consistency are all great features, but too little for the price, especially considering the latency and QD1 can't be fully taken advantage of, because the way applications are coded.

Also, the technology is seriously capped by PCI Express, and the coding for storage. The DIMMs are 30x better in latency and can displace DRAM in majority of client scenarios.

The thing that has a chance to change things fundamentally is the DIMM version. There's no doubt about it. Yes, software has to catch up, but even then.

The DIMM versions have the potential for Averages Joes to get interested in buying a new computer, and actually benefit them which hasn't happened in perhaps 15 years.

It'll take simple as Windows supporting an Optane DIMM specific hibernate feature so it can be a truly zero power "sleep" mode.

I wonder if we will get Optane DIMM support with the X299 successor, whenever that actually happens?

By then Intel will be up against next gen Threadripper based on Zen 3 and likely 64 cores/128 threads again.

Memory was the weak point with Zen 2 Threadripper so Intel going all in with both DDR5 and Optane Dimm support could capitalize on that weakness.

In particular Threadripper has absolutely crippled maximum supported memory. Supporting Optane DIMMs could make that gap absolutely laughable.

 

[IntelUser2000]

Maybe.

The CPU team especially on the desktop is hopelessly behind AMD. It's almost better expecting DDR5 Optane to be not proprietary and AMD supporting it.

Some evidence on Tigerlake-H supporting some form of Optane DIMMs.

 

[nosirrahx]

It looks like the 5800X drives will be here soon but it looks like the 905p/900p direct successor either wont happen or wont happen until PCIe 4.0 is officially supported by Intel's desktop CPUs.

That would be kind of funny, needing to buy an AMD CPU to use a "915P" (or whatever they would be called) to get optical performance.

I might actually buy a 5800X if I get curious enough.

 

[nosirrahx]

900p/905p successor not happening is official. All of the announced P5800X products are U.2 at this point further limiting consumer space compatibility.

 

[biostud]

Of intel wanted to stay in competition, they should support optane Dimm in mainstream. Imagine 32gb ram + 64Gb optane Dimm.