Discussion Optane Client product current and future

[nosirrahx]

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

Intel needs something to go up against Threadripper Pro and the inevitable Zen 3 Threadripper in the HEDT/workstation space. Optane DIMMs could be a very useful weapon here.

Intel can't do anything about the 64 cores/128 threads they are up against but native level 4 cache in the form of Optane DIMMs could eliminate a bottleneck that AMD has no options against.

 

[Dayman1225]

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.

Reviews are out for P5800X if you're still interested

Intel Optane SSD P5800X Review

The Intel Optane SSD P5800X expands on Intel’s fastest tier of SSD storage. We’ve…

www.storagereview.com

 

[nosirrahx]

Reviews are out for P5800X if you're still interested

Intel Optane SSD P5800X Review

The Intel Optane SSD P5800X expands on Intel’s fastest tier of SSD storage. We’ve…

www.storagereview.com

Sometime later this year I might put one in a new workstation. The Zen 3 Threadripper chips are supposed to be out then and the P5800X would be an awesome OS/Apps drive.

 

[IntelUser2000]

Optane P5800X is a monster. The client version would have been nice as it's top even for sequentials. Too bad it didn't sell enough to continue.

Sapphire Rapids leak has information about Optane PMEM 300. 2.6x 2R1W Random performance over 200 series.

I'm expecting nearly 20GB/s Reads and 10GB/s writes. That compares to 8GB Reads and 3GB/s Writes for the 200 series. Thus the endurance over 5 year warranty span may reach 1.5 Exabytes!

If that's for the 1TB version then per cell lifecycle will be at 1.5 million writes.

I really hope good things for the PMEM, and eventually see it for client. TGL-H looked like it'll have it, but doesn't seem like it anymore.

 

[nosirrahx]

Optane P5800X is a monster. The client version would have been nice as it's top even for sequentials. Too bad it didn't sell enough to continue.

Sapphire Rapids leak has information about Optane PMEM 300. 2.6x 2R1W Random performance over 200 series.

I'm expecting nearly 20GB/s Reads and 10GB/s writes. That compares to 8GB Reads and 3GB/s Writes for the 200 series. Thus the endurance over 5 year warranty span may reach 1.5 Exabytes!

If that's for the 1TB version then per cell lifecycle will be at 1.5 million writes.

I really hope good things for the PMEM, and eventually see it for client. TGL-H looked like it'll have it, but doesn't seem like it anymore.

Intel is going to need something like HEDT with Optane DIMM support to compete with Threadripper. Zen 3 Threadripper is going to be an absolute monster and unless something changes between now and the fall my next workstation will be a Zen 3 Threadripper with a P5800X.

 

[nosirrahx]

Remember when Intel killed the 815P successors to the 800P and it seemed a bit odd that they got all the way to the point of giving part numbers to vendors so they could start taking orders?

Yeah its not dead, just renamed:

Intel® Optane™ SSD P1600X Series (58GB, M.2 80mm PCIe 3.0 x4, 3D XPoint™) - Product Specifications | Intel

Intel® Optane™ SSD P1600X Series (58GB, M.2 80mm PCIe 3.0 x4, 3D XPoint™) quick reference with specifications, features, and technologies.

www.intel.com

Intel® Optane™ SSD P1600X Series (118GB, M.2 80mm PCIe 3.0 x4, 3D XPoint™) - Product Specifications | Intel

Intel® Optane™ SSD P1600X Series (118GB, M.2 80mm PCIe 3.0 x4, 3D XPoint™) quick reference with specifications, features, and technologies.

www.intel.com

 

[nosirrahx]

I bought 2 P1600X drives but they arrived after I went away for the winter. I will be home for 1 day on December 13th and might have a little time to fire them up and get some benchmarks.

In particular I will be comparing 2 800P in RAID 0 against 2 P1600X in RAID 0 in my NUC8i7HVK.

 

[igor_kavinski]

I bought 2 P1600X drives but they arrived after I went away for the winter. I will be home for 1 day on December 13th and might have a little time to fire them up and get some benchmarks.

In particular I will be comparing 2 800P in RAID 0 against 2 P1600X in RAID 0 in my NUC8i7HVK.

How much did you pay for them? The P4801X 100GB has better performance with an insane TBW rating. P1600X's is only like 1/10th of that. It needs to sell for at least one third of P4801X's price to be price competitive.

 

[nosirrahx]

How much did you pay for them? The P4801X 100GB has better performance with an insane TBW rating. P1600X's is only like 1/10th of that. It needs to sell for at least one third of P4801X's price to be price competitive.

22110 is the issue, almost nothing is compatible with 22110. Price is not a concern, I buy and test a lot of stuff.

BTW, Intel's choice to release all M.2 drives from that family on the 22110 form factor was kind of a d**k move.

The 100GB drive isn't even the one that really pisses me off either, its the 200GB one that could have been one hell of a great 2280 M.2 drive. If the 375GB model fits on 22110 then I don't buy for an instant that they could not have put both the 100 and 200 on 2280.

EDIT:

If intel cared about consumers this would be the 2280 Optane product stack:

Intel® Product Specification Comparison

 

[igor_kavinski]

22110 is the issue, almost nothing is compatible with 22110. Price is not a concern, I buy and test a lot of stuff.

BTW, Intel's choice to release all M.2 drives from that family on the 22110 form factor was kind of a d**k move.

The 100GB drive isn't even the one that really pisses me off either, its the 200GB one that could have been one hell of a great 2280 M.2 drive. If the 375GB model fits on 22110 then I don't buy for an instant that they could not have put both the 100 and 200 on 2280.

EDIT:

If intel cared about consumers this would be the 2280 Optane product stack:

Intel® Product Specification Comparison

I really do share your frustration about Intel's unfathomably stupid Optane business strategy. It's the best storage technology ever but Intel's greed has done nothing but stagnate it. If they would price it just right within reason, the volume production would eventually bring down the cost and the world would be a better place. Also, their mad asking price has probably deterred hard drive makers from incorporating this awesome technology as a high speed huge cache for their multi-TB hard drives where Optane would greatly improve the user experience.

By the way, P4801X should work with this adapter: Amazon.com: Micro Connectors M.2 NVMe + M.2 SATA 110mm SSD PCIe x4 Adapter with Heat Sink : Electronics

 

[nosirrahx]

I really do share your frustration about Intel's unfathomably stupid Optane business strategy. It's the best storage technology ever but Intel's greed has done nothing but stagnate it. If they would price it just right within reason, the volume production would eventually bring down the cost and the world would be a better place. Also, their mad asking price has probably deterred hard drive makers from incorporating this awesome technology as a high speed huge cache for their multi-TB hard drives where Optane would greatly improve the user experience.

By the way, P4801X should work with this adapter: Amazon.com: Micro Connectors M.2 NVMe + M.2 SATA 110mm SSD PCIe x4 Adapter with Heat Sink : Electronics

I'm all about the adapters, I have 4 905P 22110 M.2 drives in my workstation via Asus Hyper 16X running in VROC 0.

Sequential and 4KQ1T1 are both outstanding. So yeah, I am a huge Optane fan and really hate how they mismanaged the entire product run.

These P1600X drives are going into a NUC8 extreme which only has 2 2280 ports. I'll end up repurposing the 2 800P drives in other systems. They are great for anything functions that do a huge number of small reads and writes.

 

[nosirrahx]

Installed 2 P1600X SSDs into my NUC8i7HVK in RAID 0. Here is the gap in performance between 2 800P and 2 P1600X.

The jump in both sequential and 4KQ1T1 is obviously pretty great but as has always been the case, there is no way the price is worth it. This use case is still just a cool tech toy that lets you have good sequential and outstanding 4KQ1T1. This system is crazy fast in this configuration.

 

[IntelUser2000]

If they would price it just right within reason, the volume production would eventually bring down the cost and the world would be a better place.

It's not just greed, there are legitimate reasons for it, and that's the insane production cost. If they cut the prices like some are suggesting, they'd have bled money like with contra-revenue Atom chips. The price point that people would want and Intel won't lose money needs 10x the volume. Cutting the price now by half won't get them 10x the volume.

Ashraf Eassa(well known Intel leaker before he got snatched away to work at Intel) said the 900P drive's selling price barely made Intel any money.

The real thing they need to focus is the PMEMs because the performance justifies the price. It's 10x the endurance and 50x faster in latency terms compared to the SSDs.

They will get there if they don't abandon development. Right now I'm not sure. The success of the next generation PMEMs determine whether Intel continues with them. They need to stick with it until it succeeds, period! Growth will come steady but consistently.

 

[nosirrahx]

It's not just greed, there are legitimate reasons for it, and that's the insane production cost. If they cut the prices like some are suggesting, they'd have bled money like with contra-revenue Atom chips. The price point that people would want and Intel won't lose money needs 10x the volume. Cutting the price now by half won't get them 10x the volume.

Ashraf Eassa(well known Intel leaker before he got snatched away to work at Intel) said the 900P drive's selling price barely made Intel any money.

The real thing they need to focus is the PMEMs because the performance justifies the price. It's 10x the endurance and 50x faster in latency terms compared to the SSDs.

They will get there if they don't abandon development. Right now I'm not sure. The success of the next generation PMEMs determine whether Intel continues with them. They need to stick with it until it succeeds, period! Growth will come steady but consistently.

The way you deal with insane production costs is to engage the enthusiasts that must have the best of the best. Optane as a caching solution was always seen as a low end option for people that could not afford a real SSD.

If Optane was positioned as the must have drive for enthusiasts they could have sold a lot of them. Just look at what people are willing to pay for scalped hardware.

I think it was also a mistake keeping PMEMs away from the HEDT platform, witch was more or less on life support due to Threadripper. Had Intel released a X399 platform with newer chips and PMEM support they would have had an option AMD could not compete with, since as it stands X299 could not compete with the threads of Threadripper.

Out of everything that annoys me about Optane I think the worst offender was the lower capacity M.2 P4801X drives. There was no reason for these to be 22110 form factor instead of 2280, like the 800P and P1600X drives.

Oh well, I guess we will have to wait a few years to have an enthusiast focused SSD with blistering 4KQ1T1 performance. If not Optane, someone else will crack the code.

 

[IntelUser2000]

If Optane was positioned as the must have drive for enthusiasts they could have sold a lot of them. Just look at what people are willing to pay for scalped hardware.

I kinda agree with this. I think the should have brought a client version of the Memory Drive so you can expand RAM in a way. Of course the caching module sequentials are too low to sell it like that but the 9xx series would have worked. The caching should have been an option of course.

I think it was also a mistake keeping PMEMs away from the HEDT platform, witch was more or less on life support due to Threadripper.

I don't know what's up with that since they said numerous times in investor presentations that workstations would come with PMEM(not talking about the 28-core "workstation"). You remember the various media outlets covering the Optane PMEM workstation with CrystalDiskMark benchmarks?

They did say it had "platform difficulties" so they might have had a legitimate technical reason not being able to bringing it. Maybe we'lll see it with Sapphire Rapids HEDT? I do hope it expands and not dies off.

Out of everything that annoys me about Optane I think the worst offender was the lower capacity M.2 P4801X drives. There was no reason for these to be 22110 form factor instead of 2280, like the 800P and P1600X drives.

That's true. That's probably just to save on redesigning the board. the lower capacity had lower performance as well.

You mean the 905P M.2 too right? Since that was 22110.

 

[nosirrahx]

You mean the 905P M.2 too right? Since that was 22110.

I've got 4 of these in VROC RAID 0 on an Asus Hyper 16X and they are loaded with chips, they could not shrink them to 2280 unless they went with chips on both sides. The 100 and 200 GB P4801X have dead space and could easily be on the 2280 form factor. You are probably right about redesign cost though, easier to just use the same PCB and remove chips.

 

[IntelUser2000]

I've got 4 of these in VROC RAID 0 on an Asus Hyper 16X and they are loaded with chips, they could not shrink them to 2280 unless they went with chips on both sides. The 100 and 200 GB P4801X have dead space and could easily be on the 2280 form factor. You are probably right about redesign cost though, easier to just use the same PCB and remove chips.

I believe based on the 1.5TB M.2 versions of the 905P that if they got the same 8 dies per package it could have worked.

Although doing that on 2280 would require them doing rejigs on the controller so it supports the same amount of channels.

On a side note I hope they skip 1TB and go straight to 2TB capacity for the Crow Pass PMM coming with Sapphire Rapids. Samsung has 512GB module already which is high as the top Optane. Sure they are much more expensive but they lose the more memory part which is important for in-memory supercomputers.

 

[nosirrahx]

I'm starting to think I might be the only person to actually benchmarked these P1600X drives. I searched today for other reviews, picked the only chart on google image search and ......... its this thread.

 

[IntelUser2000]

Things are not looking too good for Optane. Intel is very quiet about it. I know we'll see the Optane PMM 300 series in Sapphire Rapids. The success of it will be hampered by the Sapphire Rapids chip itself.

They should have opened it up so it works on EPYC as well.

Interesting thing is that we're supposed to see Donahue Pass first before Crow Pass coming with Emerald Rapids. Donahue Pass was supposed to be the 4th generation so why first?

It does tell us Optane is ready before the Rapids chips.

 

[nosirrahx]

In the latest episode of 'Intel has no idea what its doing with Optane' it seems that when Intel sold off its SSD division, there was a miscommunication about Optane and NAND being 2 different things:

Product Support Forums

At this point Intel's DSA tool still tries to get you to install the Solidigm Storage Tool on systems with Optane drives even though the documentation clearly states that this tool does not support Optane drives. It took them a month just to fix the formatting bug in DSA, so at least the information looks correct when it tries to install incompatible software.

Its funny that this bug took and existing bug (that Intel has refused to fix for more than a year now) and made it much worse. If you have a NUC with 2 Optane drives in RAID 0, Intel's new tool for updating firmware will not be able to see the individual drives. This is funny because Intel's old tool for updating firmware does this with no issue. What is even funnier is that if you actually install the Solidigm Storage Tool on a system with 2 Optane drives in RAID 0, the tool flat out crashes. The insanity of this is that DSA clearly sees through the RAID because it recommends the Solidigm Storage Tool (erroneously) for the Optane drives and then promptly crashes when you try to use the suggested tool.

I don't think Intel tests anything on real hardware anymore. There are a lot of NUC supports threads currently that are all obvious cases where a BIOS update was pushed out after passing simulated testing since on real hardware the bugs are insanely bad.

 

[igor_kavinski]

This is what I deal with whenever a new requested feature is added to our dotnet based core application. The errors I report to the development team are so moronic that they would have caught them on their own before the UAT phase if they had only bothered to test basic functionality. It seems they just write code, see that it's compiling error-free and then expect users to report bugs which wouldn't exist in the first place if they would just use the application for a few minutes.

 

[nosirrahx]

So, I did a really dumb experiment that totally violates RAID protocol by RAIDing together 2 totally different SSDs. I put a P1600X and a 980 PRO into RAID 0 just to see what would happen. Turns out, its actually kind of amazing. Both the sequential performance and 4KQ1T1 performance were better than the 980 PRO on its own. The 4KQ1T1 performance was roughly the average of the P1600X and 980 Pro while the sequential was only a little higher than the 980 PRO.

Obviously there is no way I would trust this setup for long term stability and the price : capacity ratio is beyond terrible, but it was surprisingly fast under all scenarios.

More or less this is a totally useless experiment beyond proving that there was a potential to create pretty amazing hybrid Optane/NAND drives. The way Intel did this with the H10 drives was totally different and more or less just combined 4X->2X+2X bifurcation with Optane caching. I think what totally kills the possibility of doing a hybrid drive correctly is the cost of developing and producing the controller. In effect you would need to put a NAND controller, an Optane controller and a RAID controller into one package. Add in the cost of Optane chips and the price is just impossibly high.

 

[igor_kavinski]

Intel should have outsourced/licensed the manufacture of Optane to some company like Samsung. They would have done it far cheaper and actually could have come up with interesting new ways to put Optane's strengths to good use. Alas! They are too dumb and the upper management probably shoots down a lot of good suggestions their employees have. Same reason why Intel has been trying to compete with AMD since 2017 and failing with pie on their face.

 

[nosirrahx]

Intel should have outsourced/licensed the manufacture of Optane to some company like Samsung. They would have done it far cheaper and actually could have come up with interesting new ways to put Optane's strengths to good use. Alas! They are too dumb and the upper management probably shoots down a lot of good suggestions their employees have. Same reason why Intel has been trying to compete with AMD since 2017 and failing with pie on their face.

They did shoot down a lot of stupid stuff. There was a period of about 1 year where a cheap 2TB SATA SSD + 58 GB Optane drive was cheaper than a 2TB NVMe SSD and the 4KQ1T1 performance was way better. Not only did Intel not promote this, they flat out said it was not supported and was not the intended use of the 800P SSDs.

I like the idea of outsourcing Optane. I'm sure someone could have come up with an enthusiast focused boot drive. The secret sauce would be the controller mirroring small files to the Optane part allowing Optane to do what it does best, read small files quickly. Small Optane drives were never the problem. The problem was using Optane to do both the stuff it was great at and the stuff it was not great at. The other problem was offloading a lot of the work to the system forcing the user to deal with things like drivers and setting Optane up. A real solution makes the secret cause invisible to the user.

 

[coercitiv]

I guess it's RIP Optane.