News Optane's Last Gasp: Intel's Final Persistent Memory Roadmap Leaks

[bit_user]

So Long Optane, will be a long time if ever we see anything like you again.

Aren't there already NVDIMMs that are essentially DRAM that's backed up NAND flash and a super capacitor (i.e. to flush the contents on power loss)? That's going to get a lot more feasible, in the CXL era, if not.

This is probably why Intel decided to kill Optane - when they realized they couldn't compete on either cost or performance, with such devices.

Question: Is there no technology that would combine memory and storage into one device?

There is:

What is NVDIMM (Non-Volatile Dual In-line Memory Module)? | Definition from TechTarget

An NVDIMM (non-volatile dual in-line memory module) couples high-speed DRAM performance with onboard NAND flash for persistent storage.

www.techtarget.com

I'd guess reasons why they wouldn't do anything like that, to compete in the storage and memory market at the same time

No, it's not competitive with DRAM. Anyone buying such devices must be willing to pay a premium for the nonvolatile functionality.

well I suppose you lose a bit of fault tolerance:

One option is to use mirroring. High-end servers support mirroring (i.e. like RAID 1) of DRAM. No reason they couldn't support it for NVDIMMs.

Another option is to use other fault-tolerance strategies, like checkpointing the system state to a secondary storage subsystem with some redundancy (e.g. RAID 5 or better).

Question is would it have high endurance as system memory?

That depends mostly on the strategy for persisting the memory contents to flash. If you only do it on power-loss, using an onboard battery or supercapacitor, then you'd get all the endurance of DRAM.

There could still be a market for ssd's for more storage even if there were a hybrid memory/storage drive to run the OS and apps from.

Yes. Most people don't need NV memory, and it would command a price premium vs. regular DRAM and SSDs. In terms of GB/$, it would be extremely non-competitive with SSDs. Consumers are not the main market, for this tech.

make everything middle so manufacturing costs would be lower and everyone pays a reasonable price for the same thing.

Did you ever compare the GB/$ of DRAM vs. SSDs? Try it, some time. There are reasons why DRAM is a lot more expensive. If you build storage devices with 1:1 a DRAM-to-NAND ratio, it will not be more reasonably priced than what we have today. And scaling up volumes of such a product won't change that.

We had those. Hybrid drives.
A small solid state segment and a large spinning drive.

Different in two important ways:

1. They used DRAM + NAND + HDD, rather than DRAM + NAND.
2. Their ratio of DRAM to persistent memory was like 1:10000. I think what @DavidM012 means is 1:1, so that you get the effect of nonvolatile DRAM.

Thats what Optane was supposed to be.

We should clarify that Intel used Optane technology in two very different ways:

1. For conventional SSDs, with conventional NVMe interfaces.
2. For PMem modules, like those described in the article. It's only these PMem modules that had the potential you're describing.

 

[bit_user]

There is no non-volatile memory technology on the market that has good random performance now that Optane is a dead end (there's also the absolutely absurd endurance which no tech can come close to).

NAND-backed DRAM would win both of these contests, easily.

I'm unsure if Intel just bet on the wrong usage for the technology, or if the storage side was just never in the picture for volume sales.

Intel bet that:

• Optane would have higher endurance, making it more suitable as a DRAM replacement
• Optane would scale better, in 3rd dimension (i.e. more layers)
• NAND flash wouldn't have such an aggressive GB/$ curve - likely, they underestimated how rapidly the layer count of 3D NAND would increase.

Because of this:

• Optane is less suitable for in-memory database usage.
• Optane is less price-competitive vs. NAND flash & even DRAM

or if the storage side was just never in the picture for volume sales.

Maybe I'm misinterpreting what you mean by this, but Intel actually needed much higher sales volumes. I think Optane has never been profitable for them, since day 1. If it were, either they'd be keeping it, or they'd have sold it off to someone like Micron or SK Hynix (i.e. along with their NAND flash assets, that have now become Solidigm).

I know Gelsinger never had any interest in Intel doing anything with memory tech (or as he put it making dumb chips) which is why he was quick to bail on the NAND business.

He's a businessman. He can't just make decisions on such a whim. There needs to be a business case and a strategy behind everything he does or doesn't do. Failure to uphold his fiduciary responsibility leaves him vulnerable to lawsuits by Intel's shareholders. This isn't an idle threat, either. Qualcomm's management have been the subject of several such lawsuits.

I firmly believe consumers are the real losers here as enterprise can just throw money at circumventing the issue while consumers will be stuck with just increasing sequential R/W on NAND which is mostly irrelevant.

Not enough consumers want to pay what Optane costs. Those costs haven't been anywhere close to tracking the cost of NAND flash-based storage, which suit the vast majority of consumers well enough.

Intel needs to sell the IP to the Memory Manufacturers and get everybody in the industry involved to use it.

If it's not competitive with conventional NVDIMMs, then there's no business case. Then, Intel could offer but there would be no takers.

The way to drop the price was advancing the technology as from what I'd read their second revision of 3D XPoint managed to cut manufacturing prices almost in line with the increased density. The problem is that this means it'd be another 2 generations before they could get close enough to NAND to be a more broad alternative. So it's one of those things where Intel would have to keep developing it, bring a new fab online and then market the hell out of its storage capabilities.

You assume this is even possible. Intel hasn't been clear on the exact physics of Optane, so it's possible they hit a wall that would prevent it from scaling down enough to ever be cost-competitive with NAND flash.

Intel is big enough that they could have kept it going,

Nobody is big enough to defy physics. And selling Optane at a loss isn't a viable long-term business strategy.

 

[bit_user]

To my knowledge, there exist at least 2 emerging memory type technologies that are also non-volatile,
...
the 2 main reasons why those technologies are noy yet spreading much are :
...
2. Technical limitations of 2.1 memory controller and 2.2 emerging memory itself : memory controllers are integrated with the CPU or chipset, and so are gating the access to new memory.

There's already a standard for NVDIMMs, as I understand it. You just have to build DIMMs that conform. Then, obviously, there need to be CPUs which support it.

The problem is that you need to make the technology competitive, and that takes investment. The technologies you mentioned both have niches in which I think they are competitive (e.g. embedded), but that won't necessarily lead to something that can take on NAND flash in the mainstream market.

 

[bit_user]

Your PC & data could just be a virtual machine in a database in a datacentre somewhere if the remote connection was fast enough.

Game streaming has already been around for more than a decade, which is essentially what you're talking about. You should read up on it, if this interests you.

They made a cpu out of optics didn't they?

Yes, but I think they're not competitive with silicon CPUs, for general-purpose computation. I'd guess density is the main issue. I think the only place they're used in actual products is in high-end networking gear.

quantum computers are developing so some peoples are going to get to use supercomputers but it seems like humble consumers are getting the crumbs and cast offs.

Quantum computers aren't simply better versions of conventional machines. They work in a different way, have massive limitations, and are suited only for specific classes of problems. They're also extremely temperamental and require elaborate cooling & RFI isolation that would never be practical or affordable for a home user.

There are always a few groups looking at QC technologies more resilient to RFI and that don't require super-cooling, but I think those aren't anywhere near competitive with the state-of-the-art.

Basically, don't worry about quantum computers, for your own purposes. They are best thought of as a special type of supercomputer that requires special programming and is mainly good at solving specific technical, financial, and scientific problems.

 

[bit_user]

Shame we will get M.2 ssds that need heatsinks with fans

Optane drives burn a ton of power, though. Did you see in the article, where the new Optane PMem DIMMs burn 15 W each ??? And all for a measly 6 GB/s, sequential? Even a PCIe 4.0 SSD can do better than that for less power!

That's not to say Optane doesn't have its strengths, but you had to pick power...

 

[rluker5]

Optane drives burn a ton of power, though. Did you see in the article, where the new Optane PMem DIMMs burn 15 W each ???

And yet are kept cool by short, flat heat spreaders in high airflow cases in high density layouts. This is likely for a high write operation where a large portion of the pcm is being melted.
Right now my Hynix p41 and my 800p both hit 59c while running CDM 8 with 3 repeats and the Hynix has a heatsink twice as large. My 800p is the only thing comparable as my 900p and 905p both have large heatsinks.
I suppose that would need some discrete cooling for those that watercool and some pretty good heatsinks even for those with good airflow in their cases for an Optane dimm setup.
But it would be moving both ram and nonvolatile storage to a higher tier in memory performance. Something beyond the capabilities of our current traditional setups.

Also the 4yr older pcie gen3x2 dramless cacheless introductory Optane m.2 beats the pcie gen4x4 nand in 4k q1t1 by the same ratio the nand's cache wins in seq1m q8t1. A ratio of 5. Since I just tested them, both on chipset m.2 slots in the same pc.

 

[bit_user]

And yet are kept cool by short, flat heat spreaders in high airflow cases in high density layouts.

You're missing the point, which is that M.2 SSDs don't use anything even close to that much power. Yet, consider the Phison PCIe 5.0 SSD they just previewed:

Phison E26 SSD Preview: PCIe 5.0 SSDs Are Finally Here

The fastest drives are just around the corner

www.tomshardware.com

In only 11.4 W peak power (7 W average), it managed > 10 GB/sec sequential read & write, ~67% faster than the upcoming Optane PMem module.

This is likely for a high write operation where a large portion of the pcm is being melted.

Well, Intel doesn't list the idle power of their PMem devices, but consider their latest NVMe drives have an idle power of 3.8 W (400 GB) to 6.3 W (3200 GB). That's a difference of 2.5 W for 2.8 TB, which amounts to 0.9 W per TB, presumably just for housekeeping of the Optane. Much worse than conventional consumer NVMe drives:

• https://ark.intel.com/content/www/u...00x-series-400gb-2-5in-pcie-x4-3d-xpoint.html
• https://ark.intel.com/content/www/u...00x-series-3-2tb-2-5in-pcie-x4-3d-xpoint.html

I'm not saying Optane is bad, but maybe it's not great, if power is your main concern.

 

[DavidM012]

Basically, don't worry about quantum computers, for your own purposes. They are best thought of as a special type of supercomputer that requires special programming and is mainly good at solving specific technical, financial, and scientific problems.

Why can't I open a wormhole to the other side of the universe? I wanna Stargate

Game streaming has already been around for more than a decade, which is essentially what you're talking about. You should read up on it, if this interests you.

Kinda hard to beat the value of meh pc though. only now is it really growing long in the tooth and it didn't cost a lot to build 7 years ago, and it can't be replicated again with today's prices. Spent about £500 on it in that time. Though it has played and still runs all my old faves. from homeworld to fallout so it's dying hard anyway.

My internet subscription cost more than meh pc basically but now probably looking at a new build of $1000 or so which is more than I've paid for any pc lately, my first pc in 1997 cost £1000 which was a p120, with a 2gb hdd. So don't know if a subscription to a gaming service will be the same value. Still need a client of some description and I doubt they really want to do anything cheap. I mean $20 dollars a month is still more than what I spent on pcs really.

So when you think about it, it is a fundamentally tough market and while some demand is there it is kinda slow. Not a great environment to break in revolutionary new tech that commands a premium.

It just so happens now is the time when technology is advancing, gfx are improving, new cpus are launching etc. so just have to reconsider things soon.

Well the cost of technology it's all usually hidden away in consumer devices all the electronic circuits are built into digital fridges, boilers, washing machines etc. so consumers don't think about it like they do pcs.

I think, they will cheerfully exploit anything they find on the shelves at a bargain price anyway. Don't ask them to think too deeply about developing tech. Probably best to let the engineers, engineer.

 

[bit_user]

Why can't I open a wormhole to the other side of the universe?

As I understand it, wormholes don't work like that. You need to create the wormhole in one place, and then move the endpoints in space.

Also, wormholes aren't thought to be able to transfer matter in an intact form. Maybe information, though.

I wanna Stargate

Most sci fi is a lot more fi than sci.

Wikipedia is probably a good place to start, if you want to know what actual theory predicts: https://en.wikipedia.org/wiki/Wormhole

 

[DavidM012]

I think they solved that problem by having the ancients seed the galaxy with stargates a million years ago so stargate command didn't have to invent the network they just co-opted it being the last surviving technologically advanced species by the end of the series.

And the stargate was a matter/information converter or something like a star trek transporter. Yes though it be more fi than sci it still spawned a multi million dollar franchise and basically most pc games have evolved from literature, or movie scripts from novels, comics and spin offs.

Fun and imagination seems to be an industry. You couldn't make it up but you could and sometimes the truth is stranger than fiction. Why dumb things work?

 

[DavidM012]

Besides all that I suppose customers have already spent their PC or IT budget on sd ram and nand rather than a promising looking new tech called Optane.

Who am I to lecture but Intel probably needed a working product poised to capture some of the market. Like dolphins hunting mackerel you just have to corral them around I suppose.

How you can do r&d and take such an epic risk at the same time, pish.

 

[falcon291]

Question: Is there no technology that would combine memory and storage into one device? So if you could have a hard disk at the speed of ram and memory with the capacity of a hard disk wouldn't that speed up the entire system? It would be like loading your games on a ram disk and since Optane can be non volatile couldn't they just redesign the architecture a bit to have a fast hybrid memory/storage interface to the cpu and be done with it?

That's no reason you couldn't hybridise storage and memory and design yet another new mobo interface?

Even the fastest M2 PCI Express drives are a lot slower than memory. And memory itself is not that expensive anymore, so no hybridizing storage and memory does not make much sense. At the server level, these solutions can make some sense, but even then, increasing the memory and/or replacing HD with a SSD is possible, and make better sense than Optane.

SSDs (both M2 and SATA) are cheap and getting cheaper, and Optane makes sense only in certain circumstances and there aren't many. It can only make sense for my gaming Hard disk that is 6 TB, and even then, its effect would be minimal.

Optane is on the wrong path at the storage drives evolution. Much like dinosaurs.

 

[falcon291]

I used a hybrid drive for years, its performance was outstanding for the time, a lot faster than a 5400 rpm 2.5" drive. But as soon as SD drives got cheaper, I replaced it and never looked back. We don't need to cache our M2 PCI Express drives and it is the end of the story for Optane.

 

[DavidM012]

Yes but if you can use optane both for dimms and storage because it's non volatile maybe storage speed could've caught up to ram speeds

so nand and dram are different chips nand is non volatile so hybridising nand and dram doesn't make much sense but maybe if there was some substrate that was suited to both then you could have ram with the capacity of a hard drive and a hard drive with the speed of ram.

Why would such a thing be useful well sometimes you want more memory and you wouldn't say no to faster storage if it was possible and older drives only went obsolete because data grew, windows is 30gb, steam games often are 50-100gb or more and you can't really know if the industry is doing all it can to be data efficient and power efficient as possible like coding tighter- rather they've gone the other way, power consumption is up again, and software is bloated, large maps and gfx need loads of vram etc.

I s'pose nvme is fast enough now but the internet connection is still pretty bad in some places so it might be preferable to get steam games on a usb drive. That might solve a couple of problems: If you had backups on a usb flash drive or micro sd card you wouldn't need terabytes of storage for a steam library and far as durability goes you could copy the back ups to a new drive every while.

Well so long as you have a fast usb port and flash drive to load 100gb game so it's convenient to load when you want rather than a 5,0mb/s usb 2.0 flash drive or a 4.0mb/s broadband up 'til recently there were only naff internet deals so maybe now things might be changing for the better. Gigabit ethernet has been around for years but when the internet connection was only 4.0mb/s, it's kind of annoying to lag behind the times.

Still usb flash drives probably aren't as fast or durable as might be liked either though they are cheap well iI just think the trouble is data storage could go in several directions, the network, or nand drives, or usb flash drives so you'd have to choose which method to invest in to develop for the mass market.

A fast cheap durable flash drive that could do 100mb/s would be better than a 4.0mb/s internet connection that takes 4 days to download 1 100gb game from the steam store.

Basically your snail mail naff broadband connection makes them a profit if they don't have to send out dvds or flash drives and add overheads to the cost of games . >_< Like literally 2023 and only now is my isp offering a fibre connection so annoying.

 

[thestryker]

NAND-backed DRAM would win both of these contests, easily.

Does any such persistent storage product exist on the market today?

Intel bet that:

• Optane would have higher endurance, making it more suitable as a DRAM replacement
• Optane would scale better, in 3rd dimension (i.e. more layers)
• NAND flash wouldn't have such an aggressive GB/$ curve - likely, they underestimated how rapidly the layer count of 3D NAND would increase.

Because of this:

• Optane is less suitable for in-memory database usage.
• Optane is less price-competitive vs. NAND flash & even DRAM

Intel would have known the route NAND was going given that they were competing at the very top of the market. If you look back at the way Intel launched Optane it was all about providing that stopgap between RAM and storage which required the low latency of persistent memory rather than PCIe based storage. What they didn't bet on was the capacity scaling of DRAM (and price collapse) which is where they thought they would have an edge.

Maybe I'm misinterpreting what you mean by this, but Intel actually needed much higher sales volumes. I think Optane has never been profitable for them, since day 1. If it were, either they'd be keeping it, or they'd have sold it off to someone like Micron or SK Hynix (i.e. along with their NAND flash assets, that have now become Solidigm).

You are because I poorly worded it. They spent a ton of resources on Pmem and the storage side was always an afterthought, but it was never clear why. Whether it was cost related or due to how much was invested in Pmem we will probably never know.

He's a businessman. He can't just make decisions on such a whim. There needs to be a business case and a strategy behind everything he does or doesn't do. Failure to uphold his fiduciary responsibility leaves him vulnerable to lawsuits by Intel's shareholders. This isn't an idle threat, either. Qualcomm's management have been the subject of several such lawsuits.

My point is that he didn't think Intel should be in that business so it was an easy target for cuts.

Not enough consumers want to pay what Optane costs. Those costs haven't been anywhere close to tracking the cost of NAND flash-based storage, which suit the vast majority of consumers well enough.

You're talking right now, and I'm talking long term. NAND as it stands, and there's no indication it will change, has unfixable issues which absolutely hold back storage innovation for the masses. Barring a new technology actually bearing fruit consumers are stuck with it for the foreseeable future.

You assume this is even possible. Intel hasn't been clear on the exact physics of Optane, so it's possible they hit a wall that would prevent it from scaling down enough to ever be cost-competitive with NAND flash.

You're assuming it was a problem with the technology rather than a purely business decision due to the amount of money it would have cost Intel to continue manufacturing 3D XPoint (remember they never produced any themselves). You could be right, but then again you could be completely wrong because they've never disclosed the particulars behind the decision.

 

[Diogene7]

There are several drawbacks to non-volatile Intel Optane memory that prevented it to spreading :

1. It is an Intel propietary technology supported by a proprietary Intel memory controller : The memory controller is NOT an open standard that could support different types of non-volatile memory, neither from different manufacturer.

However there are now (2022/2023) ongoing work at JEDEC (for memory working on a memory bus) and CXL to standardize the protocols to work with different kind of non-volatile memory (Carbon nanotube NRAM, Magnetic MRAM, RRAM, PCRAM,…) from different companies (start-ups,…)

2. Cost due to not sufficient enough scale : Optane was in somewhat High Volume Manufacturing (HVM) but by only one manufacturer (Intel) versus many manufacturers for NAND, RAM,… allocating financial resources to manufacture and improve them, which help reduce costs…

3. Technical limitations like high power consumption (not sure, but I think Optane DC PM RAM modules were consuming maybe in 10W / 15W) and limited R/W endurance (much lower than DRAM, I would say more than 10E6 (1000 000 times) less than DRAM).

So if you would use R/W on Optane memory intensively (like in a Data Center (DC)), then it could wear it quickly and fail maybe in a matter of a few months…



To my knowledge, many R&D labs from goverment and private companies across the world are investigating different type of MRAM (STT-MRAM, SOT-MRAM, VG-SOT-MRAM, VC-MRAM,…), like goverment labs IMEC in Belgium, and even commercialized by start-ups (Everspin, Avalanche technology,…) and foundries like Samsung, Globalfoundries, TSMC,…

To my knowledge, MRAM seems to require much less power to Read / Write, and can have an endurance that exceeds 10E10 or more…

And thanks to the ongoing standardization work at JEDEC and CXL, different Non-Volatile Memory (NVM) type from different manufacturers (therefore including MRAM) increase the Total Adressable Market (TAM) size (volume) for MRAM.

For Nantero carbon nanotube NRAM:
To my knowledge, only Nantero (created in 2001) is meaningfully allocating financial ressources. Even though maturity has increase, it is a very slow process due to that only 1 company is trying to do it all alone…

 

[Diogene7]

I not a specialist, but indeed New emerging Non-Volatile-Memory (NVM) won’t yet benefit of High Volume Manufacturing (HVM) scale as DRAM / NAND…

And in the next 10 years (up to 2030), I would think that MRAM would stay a (growing) niche (mainly SRAM replacement and a bit of DRAM), and even after that the next lowest hanging fruit is likely DRAM due to catching up to the improving NAND density/cost would require tremendous amount of financial resources (or an important technology disruption for example that would makes it possible to produce MRAM re-using NAND manufacturing tools (so re-using existing amortizedcapex))…)

To recoup investment done in MRAM, I think start-ups have first to find premium niche markets where customers are willing to pay a premium for the benefits that MRAM Non-Volatile Memory (NVM) nature could bring.

Then Data Centers, Artificial Intelligence,… may be big opportunities to gain higher volume manufacturing and increase revenue/profits, which could then be re-allocated to improve the technology.

Regulations (laws) that would make mandatory to have bi-stable NVM memory in devices could also help accelerate the transition further (like regulations help the transistion from ICE to EV cars).

 

[rluker5]

You're missing the point, which is that M.2 SSDs don't use anything even close to that much power. Yet, consider the Phison PCIe 5.0 SSD they just previewed:

Phison E26 SSD Preview: PCIe 5.0 SSDs Are Finally Here

The fastest drives are just around the corner

www.tomshardware.com

In only 11.4 W peak power (7 W average), it managed > 10 GB/sec sequential read & write, ~67% faster than the upcoming Optane PMem module.



Well, Intel doesn't list the idle power of their PMem devices, but consider their latest NVMe drives have an idle power of 3.8 W (400 GB) to 6.3 W (3200 GB). That's a difference of 2.5 W for 2.8 TB, which amounts to 0.9 W per TB, presumably just for housekeeping of the Optane. Much worse than conventional consumer NVMe drives:

• https://ark.intel.com/content/www/u...00x-series-400gb-2-5in-pcie-x4-3d-xpoint.html
• https://ark.intel.com/content/www/u...00x-series-3-2tb-2-5in-pcie-x4-3d-xpoint.html

I'm not saying Optane is bad, but maybe it's not great, if power is your main concern.

OK you are right on the power.
HWinfo64 doesn't list power for my drives, but the temps on my drives on my basically idling system are 27c for the ram which has a fan, 30c for the hdds, 44c for the u.2 905p (solid aluminum case with fins) 49c for the 800p (in a cheap, sub pcie bracket height aftermarket passive cooler), 43c for the 900p pcie (good passive pcie cooler, is os drive), 47c for the Hynix P41(under the motherboard supplied dual m.2 length heatsink) and 64c for a bare gen3 nvme drive. The Optane drives 800,900,905p are warm considering their coolers.
Optane does use more power than ram and likely more than a pcie gen4,5 ssd and efficiency varies by metric.

I used a hybrid drive for years, its performance was outstanding for the time, a lot faster than a 5400 rpm 2.5" drive. But as soon as SD drives got cheaper, I replaced it and never looked back. We don't need to cache our M2 PCI Express drives and it is the end of the story for Optane.

Your m.2 is a hybrid with cache. That's why their performance drops off over time with continuous load. But if you never run it long enough to leave the cache it doesn't really matter that much does it? You may get superior performance with a Samsung sata with rapid mode ram cache enabled btw. At least the numbers look a lot better on benches.
If you wanted a fast drive that wasn't hybrid and didn't rely on a cache for speed there is only Optane to choose from.

 

[USAFRet]

You may get superior performance with a Samsung sata with rapid mode ram cache enabled btw. At least the numbers look a lot better on benches.

Yes, the benchmark numbers look impressive.
Actual user facing performance, not so much.

I tried it, saw no real benefit.
So I turned it off.

 

[bit_user]

Fun and imagination seems to be an industry. You couldn't make it up but you could and sometimes the truth is stranger than fiction.

The universe is incredibly weird and wonderful, as it is. It's fine if people want to go down their sci fi rabbit holes, but you could also just appreciate it for all the fascinating things we already know about it.

 

[bit_user]

Besides all that I suppose customers have already spent their PC or IT budget on sd ram and nand rather than a promising looking new tech called Optane.

Intel announced Optane in 2015. They spent a lot of time talking about it, and believe me: everyone in the industry heard the message!

It wasn't for a couple more years that they actually started shipping products, but they've been on the market for at least 5 years. In that time, multiple IT budget & planning cycles have come and gone.

 

[bit_user]

Intel would have known the route NAND was going given that they were competing at the very top of the market.

Intel overestimated Optane and underestimated 3D NAND. It's as simple as that. These strategic directions were probably set years before 3D NAND came onto the scene.

And, in case you don't remember, Intel went through several rounds of walking back their initial claims about Optane, and the initial products featuring it shipped more than a year late. It's very obvious that the technology had real issues and limitations that probably aren't going away just by throwing money at them.

My point is that he didn't think Intel should be in that business so it was an easy target for cuts.

He can't just do stuff on a whim. And even if he just thought it weren't the proper strategic direction for Intel, he'd have spun it out or sold it off if it were viable.

You're assuming it was a problem with the technology

Yes. That's how the business works. If the technology truly had potential to be cost-effective, it wouldn't be killed off. It would live on, in some form or another.

 

[bit_user]

Your m.2 is a hybrid with cache. That's why their performance drops off over time with continuous load. But if you never run it long enough to leave the cache it doesn't really matter that much does it? You may get superior performance with a Samsung sata with rapid mode ram cache enabled btw. At least the numbers look a lot better on benches.

Source: https://www.tomshardware.com/features/phison-e26-ssd-preview-pcie-5-ssd​

Heh, Samsung 990 Pro has lower and shorter sustained write performance than many of its competitors. The Kingston and Acer drives are really out ahead, on both counts (excluding the Phison prototype).

If you wanted a fast drive that wasn't hybrid and didn't rely on a cache for speed there is only Optane to choose from.

There are some datacenter SSDs with very strong sustained write speeds. I'm not sure about zero falloff, but the Solidigm D7 P5620 claims to sustain at least 3.7 GB/s, in its largest incarnation.

D7-P5620

Mid-endurance PCIe 4.0 SSDs with capacities up to 12.8 TB, designed for mixed-workload applications.

www.solidigm.com

 

[rluker5]

Source: https://www.tomshardware.com/features/phison-e26-ssd-preview-pcie-5-ssd​

Heh, Samsung 990 Pro has lower and shorter sustained write performance than many of its competitors. The Kingston and Acer drives are really out ahead, on both counts (excluding the Phison prototype).


There are some datacenter SSDs with very strong sustained write speeds. I'm not sure about zero falloff, but the Solidigm D7 P5620 claims to sustain at least 3.7 GB/s, in its largest incarnation.

D7-P5620

Mid-endurance PCIe 4.0 SSDs with capacities up to 12.8 TB, designed for mixed-workload applications.

www.solidigm.com

1M QD32.
Anybody ever use that in real life at home? try QD1.
And write? That only matters if you were to be using the device to store data from a faster drive, like those 1TB dram drives you were mentioning are the next big thing. 4gigabits/second internet is needed to outpace SATA.

Also did you notice that Tom's stopped showing Optane in SSD reviews? I suppose there is a very small market for it and lots more want to play with their new toys that will still be a tier behind it in practical use.

But back to your response from my post: is there a cache in these drives or not? How many pcie gens will pass before nand can get to 1st gen Optane SSD 4k q1t1 latency? 10?

I get that Optane loses on practical price/perf. But it only loses to dram on heavy use, consistent performance, gaming, light use, etc. It is a shame that it won't be further developed and have it's production scaled up so costs go down and everyone gets better performance.
Optane usually doesn't have an attached cache because the only thing that would work is dram. Like in that server board pictured in the article. If you gave a 5800x a similar sized cache of dram that the others have of dram+slc, what would the following chart look like?

That chart is Optane at it's weakest.

Why are you so enthusiastically championing mediocrity?

Edit: You can see the jump in performance from pcie gen3 to pcie gen4 in Optane. Where do you think pcie gen5 would land? My guess is the slowest point wouldn't even be slow enough to be on the chart you posted. But we'll never know. Yay for nand!

 

[USAFRet]

Why are you so enthusiastically championing mediocrity?

What is/was the consumer price difference between the 800GB Optane and a 1TB 970 EVO?

There are many things that were faster/more stable/longer lasting in the coorporate world vs the consumer world.
Does that make up for the price diff?