Samsung's Z-NAND to Compete Favorably With Intel's Optane

[Raevenlord]

As big data usage is becoming ubiquitous, and workload data-sets increase in both size and complexity, new ways of connecting processing resources to storage are being developed. Intel and Micron's partnership in developing 3D XPoint memory came as a way for computer systems to reduce their bottlenecks in storing data for processing, with a particular emphasis on reducing latency. Samsung, however, has enough resources to try and provide alternatives for the emergent market needs, and being one of the most important players in the NAND industry, it seems the company is betting on the Z-NAND wagon.

For now, Z-NAND as it is being developed by Samsung, is expected to be a new rendition of SLC (Single-Level Cell) NAND, with increased controller tweaks and improvements to achieve greater IOPS in both random and sequential workloads. SLC has already been widely used in the SSD market, though in recent years it has been giving way to density-oriented technologies, such as MLC and, more recently, TLC NAND, in an effort to lower the $/GB equation. Z-NAND is a return to the SLC roots, with some very relevant tricks up its sleeve - while 3D XPoint's call to fame was sometimes up to 10x lower latency (in the order of 10/10μs), Z-NAND is also bringing latency to levels hitherto unknown to NAND memory - specifically, to the 12-20/16μs realms.






Comparing Intel's 750 GB Optane P4800X with Samsung's upcoming 800GB, SZ985, Z-NAND based solution, Samsung's drive delivers higher random read IOPS than Intel's Optane (750K vs 550K), but falls short on write IOPS (175K vs the same 550K for Optane). R/W bandwidth, however, seems to be in Z-NAND's favor: it delivers 3.2 GB/s in both metrics, while the Optane P4800X makes do with 2.4 GB/s and 2 GB/s, respectively, while endurance between both technologies should be equivalent.

View at TechPowerUp Main Site

 

[StrayKAT]

All things being equal, Optane has one small advantage: It doesn't lose speed at high capacity.

 

[Nokiron]

Pretty pointless numbers for consumers, this does not say much unless they show numbers of bandwidth and latency with very low queue-depths.

That's where Optane excels.

 

[bug]

Pretty pointless numbers for consumers, this does not say much unless they show numbers of bandwidth and latency with very low queue-depths.

That's where Optane excels.

If they compare this to P4800X, it's an enterprise product. So yes, designed to be pretty pointless for consumers at this time.

 

[R-T-B]

All things being equal, Optane has one small advantage: It doesn't lose speed at high capacity.

Honestly, SLC NAND shouldn't as extremely either. We just haven't seen it in so very very long.

 

[RejZoR]

I wonder if Optane (non-volatile) memory could replace volatile DDR cache on SSD drives, making them basically 100% proof against power failures. Even if power goes out, everything would remain in it and would get written/scheduled when controller gets power back like nothing even happened. Optane is slowe than DDR memory, but on SSD drives, I don't think it would matter as it's still faster than NAND. It would make sense. This way you don't need to include power failure capacitors at all. Just stack of NAND for storage and DDR cache replaced by Optane.

 

[R0H1T]

All things being equal, Optan (recent)e has one small advantage: It doesn't lose speed at high capacity.

You mean it supposedly doesn't lose speed when the drive's nearing full capacity, I've posted this earlier as well but none of the tests actually compared the drive when it was full. We also don't know the secret sauce behind 3dXP right now.

Pretty pointless numbers for consumers, this does not say much unless they show numbers of bandwidth and latency with very low queue-depths.

That's where Optane excels.

For consumers an Optane drive is already way overkill, this ZNAND drive as well.

I wonder if Optane (non-volatile) memory could replace volatile DDR cache on SSD drives, making them basically 100% proof against power failures. Even if power goes out, everything would remain in it and would get written/scheduled when controller gets power back like nothing even happened. Optane is slowe than DDR memory, but on SSD drives, I don't think it would matter as it's still faster than NAND. It would make sense. This way you don't need to include power failure capacitors at all. Just stack of NAND for storage and DDR cache replaced by Optane.

You don't need optane for that, a few tantalum capacitors are more than enough.

 

[Nokiron]

For consumers an Optane drive is already way overkill, this ZNAND drive as well.You don't need optane for that, a few tantalum capacitors are more than enough.

Optane is still a tangible improvement over current NAND if you have sat in front of a system with one. There is no doubt that you can point to the system with an Optane OS-drive next to one with a regular M.2. SSD was "overkill" in the beginning as well.

Optane-drives does not need capacitors at all though, so you get more benefits using Optane instead of capacitors right?

 

[R0H1T]

Optane is still a tangible improvement over current NAND if you have sat in front of a system with one. There is no doubt that you can point to the system with an Optane OS-drive next to one with a regular M.2. SSD was "overkill" in the beginning as well.

In a blind test I doubt 90% of the people out there could differentiate between a 960 Pro or Optane, as an OS drive, unless we're talking about writing massive amounts of data to the same drive.

Optane-drives does not need capacitors at all though, so you get more benefits using Optane instead of capacitors right?

I think so but I'm not 100% sure.

 

[Nokiron]

In a blind test I doubt 90% of the people out there could differentiate between a 960 Pro or Optane, as an OS drive, unless we're talking about writing massive amounts of data to the same drive.
I think so but I'm not 100% sure.

You do actually feel it, if you thought normal NVMe drives were responsive they are nothing compared to Optanes. I have used a P4800X 375GB for a week or so now at work and I really feel that I am switching to a slower drive when using my 750s at home.

 

[R0H1T]

You do actually feel it, if you thought normal NVMe drives were responsive they are nothing compared to Optanes. I have used a P4800X 375GB for a week or so now at work and I really feel that I am switching to a slower drive when using my 750s at home.

That might just be your work (environment) though, because we're talking microseconds.

Basically for tasks that aren't all that read/write latency sensitive, you will be hard pressed in noticing impressionable difference between the two drives I mentioned, or any top of the line NVMe drive for that matter.

 

[bug]

That might just be your work (environment) though, because we're talking microseconds.
...
Basically for tasks that aren't all that read/write latency sensitive, you will be hard pressed in noticing impressionable difference between the two drives I mentioned, or any top of the line NVMe drive for that matter.

Maybe so, but nowadays, besides those "tasks that aren't all that read/write latency sensitive", there's all kinds of stuff the OS runs in the background.
And we don know latency is felt, because SSDs don't do so well at QD1.

 

[Gasaraki]

You mean it supposedly doesn't lose speed when the drive's nearing full capacity, I've posted this earlier as well but none of the tests actually compared the drive when it was full. We also don't know the secret sauce behind 3dXP right now.
For consumers an Optane drive is already way overkill, this ZNAND drive as well.You don't need optane for that, a few tantalum capacitors are more than enough.

Most of the Optane drive reviews on the net have tested the drives when full.

 

[natr0n]

The more the merrier.

 

[R0H1T]

Most of the Optane drive reviews on the net have tested the drives when full.

That's not how I remember it, none of the reviews explicitly state testing drives that are almost full ~

The difference between the Optane 900P and everything that came before it is simply unbelievable.

On the two opposing sides of the chart, we see how performance is typically measured with 100% read and 100% write (displayed as 0% read) workloads. Mixed workloads reside in the middle. These mixtures are a critical area for enterprise SSD testing, but many consumer SSD reviewers also examine it due to its direct impact on application performance.

The charts in this section also show steady-state performance, an often-overlooked area. Many of our performance tests are void of this steady-state data. Your SSD would have to be nearly full to experience this type of low performance. Performance degrades as the drive fills and NAND-based SSDs slow down. 3D XPoint doesn't suffer from that issue, at least not on the normal scale.

http://www.tomshardware.com/reviews/intel-optane-ssd-900p-3d-xpoint,5292-2.html

This can be partly mitigated by keeping/sparing more of the drive for over provisioning.

 

[bug]

That's not how I remember it ~http://www.tomshardware.com/reviews/intel-optane-ssd-900p-3d-xpoint,5292-2.html

This can be partly mitigated by keeping/sparing more of the drive for overprovisioning.

Tom's Hardware, what a blast from the past...

 

[Gasaraki]

In a blind test I doubt 90% of the people out there could differentiate between a 960 Pro or Optane, as an OS drive, unless we're talking about writing massive amounts of data to the same drive.
I think so but I'm not 100% sure.

If you actually bothered to read the reviews, all the reviewers said there was noticeable difference between Optane and regular SSDs. Optane is tuned for Q1 performance while regular SSDs ramp up speed at deeper queues. Most consumers will be sitting at Q1.

 

[R0H1T]

If you actually bothered to read the reviews, all the reviewers said there was noticeable difference between Optane and regular SSDs. Optane is tuned for Q1 performance while regular SSDs ramp up speed at deeper queues. Most consumers will be sitting at Q1.

How many applications sit doing just QD1 w/r all day? As for reading the review maybe you can find real world differences beyond QD1 tuned Optane?

Hint ~ try hdtune, maybe that'll answer your query.

 

[Nokiron]

That might just be your work (environment) though, because we're talking microseconds.
View attachment 94145
Basically for tasks that aren't all that read/write latency sensitive, you will be hard pressed in noticing impressionable difference between the two drives I mentioned, or any top of the line NVMe drive for that matter.

Well, i'm talking about general desktop use. Not specific workloads, otherwise I would have mentioned specific programs.

There are obvious differences, I promise you.

 

[R0H1T]

Well, i'm talking about general desktop use. Not specific workloads, otherwise I would have mentioned specific programs.

There are obvious differences, I promise you.

I'm sure there are, but let's say you're working on gigabytes (or TBs?) of data per day, using Optane, & so you would notice the difference. Would you still notice any major difference if your workload was say a 10th or just 100th of what you do?

It would help if you could tell us you what you do on the 4800x & 750, nothing specific but perhaps the same application & same workload?

 

[Nokiron]

I'm sure there are, but let's say you;re working on gigabytes (or TBs?) of data per day, using Optane, & so you would notice the difference. Would you still notice any major difference if your workload was say a 10th or just 100th of what you do?

It would help if you could tell you what you do on the 4800x & 750, nothing specific but perhaps the same application?

Most definitely. I mean, just browsing around in Windows and surfing the web is a lot more responsive.

I notice the increased zippyness constantly.

 

[bug]

How many applications sit doing just QD1 w/r all day?

Pretty much all of them? I mean, apps don't constantly access the storage, but when they do, it's usually at QD1. Even a humble browser will write everything it loads on the disk for caching. And those are all small writes (and subsequent reads), stuff that SSDs like the least.

 

[StrayKAT]

You mean it supposedly doesn't lose speed when the drive's nearing full capacity, I've posted this earlier as well but none of the tests actually compared the drive when it was full. We also don't know the secret sauce behind 3dXP right now.
For consumers an Optane drive is already way overkill, this ZNAND drive as well.You don't need optane for that, a few tantalum capacitors are more than enough.

I think one of the reviews mentioned how/why, but now I can't remember. I know it's not electron based like NAND, but that's it.

 

[R0H1T]

Pretty much all of them? I mean, apps don't constantly access the storage, but when they do, it's usually at QD1. Even a humble browser will write everything it loads on the disk for caching. And those are all small writes (and subsequent reads), stuff that SSDs like the least.

That might have been the case previously but modern day browsers rely more on memory (cache) these days, including chrome. Most of the disk access that I've observed is due to extensions & plugins.

I think one of the reviews mentioned how/why, but now I can't remember. I know it's not electron based like NAND, but that's it.

3dXP no doubt is technologically superior to traditional NAND, coming 20 years later that is to be expected. Having said that we have absolutely zero info as to why that is, & how 3D SLC compares to Optane.

 

[bug]

That might have been the case previously but modern day browsers rely more on memory (cache) these days, including chrome. Most of the disk access that I've observed is due to extensions & plugins.

Have you checked the size of your browser's on-disk cache? Is it zero?

Plus, I'm not even sure why you're trying to tell Nokiron, the only guy here who has actually experienced Optane, that he's not seeing any benefits.