A requiem for Optane, Intel's KV cache killer that could have eased the RAM price crunch

History is littered with the corpses of technologies that were ahead of their time, and Intel’s Optane storage and memory products are certainly among them. Expensive, badly misunderstood, and awkwardly priced, the technology struggled to find a place in the market, surviving just five years before Chipzilla pulled the plug. …

[Jou (Mxyzptlk)]

3D XPoint was a missed opportunity...

Having it attached like DRAM, practically no endurance problem, four times the amount for roughly the same price and speeds which were, by that time, fantastic without comparison.

The problem: Intel marketing strategy. Only the most expensive Xeon CPUs were enabled for that type of memory-bus attached storage. If every Xeon, from the lowest to the highest, would have had it, it would have been a very big hit. Whether to use it as "swap", or "persistent storage cache", for database transactions files, or actually as RAM expansion since even MS-Windows has the ability to handle two memory speed classes.

We had a lot of cases where with smaller servers which could have used it. By that time most prominent in the lower segment: MS-Exchange, possibly combined via storage-spaces tiering or whatever, we could never put 3D XPoint in use. And for the higher segment: Was not needed there, since the price for more RAM was cheaper than whatever ran on those monster boxes and added complexity was rather frowned upon (i.e. not on the supported list).

But, actually, Intel should have even enabled 3D XPoint for nearly every CPU, not just Xeons, not repeating their stupid ECC-capability CPU politics.

[TReko]

Re: 3D XPoint was a missed opportunity...

Very true - this is what many greedy companies fail to understand - it is often the tinkerer on a low-end system that ends up building something really useful with your technology, which will move to the bigger end of town.

Pricing them out hobbyists is stupid.

[Yet Another Anonymous coward]

Re: 3D XPoint was a missed opportunity...

Same reason we don't have desktop ECM memory. You reserve support for expensive Xeon processors so the memory stays expensive so never becomes ubiquitous

Trying to explain to non-technical users and regulators that sometimes the computer can just randomly change data and it's not our fault and there's nothing we can do.

[Jou (Mxyzptlk)]

Re: 3D XPoint was a missed opportunity...

I bet you mean ECC. This is why AMD. For over a decade now, nearing two decades, real ECC for desktop computers.

My Ryzen 2700x was built with 64 GB ECC RAM for that reason. Then upgrade to Ryzen 3900x, then to 5950x. My plan was to get the 16 core Ryzen with 3D-Cache on both chiplets as soon as it gets available, but AMD released that CPU a year to late, the RAM prices say "no". The 9950x3d2 CPU price is fine, but RAM costing over double than the CPU, let alone with real ECC OVÄR FAIVE TAIMES the CPU?

Oh, and I tested the ECC error correction with RAM overclocking on all three Ryzen. The computer suddenly gets very slow, I get an event log entry, but I can still do a clean shutdown. The server variants have better ECC correction hardware I suppose, 'cause THAT speed difference is not desired there.

[Yet Another Anonymous coward]

Re: 3D XPoint was a missed opportunity...

Yes, sorry - knew there was something wrong but when I spotted it I had missed edit window.

Is there any reason why some competitor to Intel (cough AMD) don't do ECM in the memory controller together with a bit of extra cache so you can do ECC with regular RAM? It wouldn't seem to cost too much performance or silicon

[Jou (Mxyzptlk)]

Re: 3D XPoint was a missed opportunity...

That type of ECC, i.e. use five or nine regular RAM modules, or rather chips a few decades ago, for this was used "back then", when you combined 36 RAM chips or 72 RAM chips on a memory board the size of A4 / Letter. Sometimes the ECC logic was even on that big board AFAIK. The current ECC implementation is practically identical, just shrinked.

But if you want to take modern standard memory modules you need at least five or nine modules for a reasonable implementation, which means a lot of board layout and so on, for at least three better five to nine RAM channels on one CPU. That was easier when the RAM bus was small, i.e. only 16 bits or less.

[Yet Another Anonymous coward]

Re: 3D XPoint was a missed opportunity...

The number of bits per parity is presumably a trade off between cost and number of double bit errors you want to detect.

Even have 1 parity per 32 bit word is better than nothing and only reduces your available memory by 3%

Like how drives used to reserve space for bad blocks.

PS. I wonder why memory errors haven't gone up massively as we made gates smaller and lower voltage, presumably the amount of damage a passing muon could do is now much larger?

[Jou (Mxyzptlk)]

Re: 3D XPoint was a missed opportunity...

This is true. The "classical" ECC memory uses a hamming 127 cut off to 64, but if you go a bit higher, which i suspect is done for the internal ECC of DDR5, you can use a hamming distance of 4095 where ~0.3 % parity bits needed - ralistically though 0.6% since it would be cut off at 2048. Or 132071 cut off to 65536, which would requite only 1 partiy bits per 512 bytes (unless I miscalculated, it is evening here).

The standard of 8 parity per 64 bits is probably for simplicity when the technique was introduced and the circuits still had to stay simple.

[prof_peter]

It's HARD to introduce a new memory technology

To build a competitive memory/storage product you need cutting-edge fab facilities, or you can't get the density. If it's a weird technology like Optane you need to devote an entire fab to it, you can't just interleave your Optane runs between runs of DRAM or logic or something else. That means you need to consistently sell a very large amount of the product.

Since list price is for chumps, and Optane wasn't a commodity product, it's hard to figure out the price Intel charged most of their customers, and even harder to figure out Intel's cost. But Optane was significantly denser than DDR5 DRAM, fitting a lot more bits on each square mm, so it seems like it was inherently cheaper.

The problem is that they didn't have a strategy for selling it - they had an abomination that seemed like it was shaped by internal Intel politics rather than any serious market considerations. They stumbled over themselves to create barriers against buying it. The result? They didn't sell enough of it for their partner (Micron) to keep the fab (Lehi, Utah) open, then they didn't exercise their option to buy the fab when Micron exercised their option to get out, so the fab got sold to Texas Instruments and is now set up for something completely different.

[AtomicWombat]

Re: It's HARD to introduce a new memory technology

This seems to happen every time Intel tries to do something different, especially if it's high performance. They don't know how to price it, they make it almost impossible to get, and then they're surprised when it doesn't IMMEDIATELY take off like a rocket. Itanium was like that, Xeon Phi was like that, Intel's brief forays into Lustre and Infiniband were like that...

[TReko]

Software was a problem

A big part of Optane's failure was the way all modern operating systems and the applications that run on them are designed.

They have grown up in over 60 years where there was fast, temporary storage (RAM), and slower permanent storage (disks).

Optane changed this - permanent fast memory - you could, for example simply, u power down your server, then power back on again later and it just resumed where you left off? Why do you need to boot the OS, why do you need to save a file. There's just one sort of memory fast and permanent. Our computer architecture wasn't built for this.

[Piro]

Re: Software was a problem

VMware has made a big fuss about memory tiering to NVMe recently, to save on memory costs.

https://blogs.vmware.com/cloud-foundation/2025/06/19/advanced-memory-tiering-now-available/

Then you wouldn't have to give a damn about the guest OS, they don't need to know anything. As long as your hypervisor is aware and can shuffle the data, Optane would have been absolutely perfect for this. Optane would be a seriously appropriate technology for 2026.

[xyz123]

Re: Software was a problem

Imagine an OS that was always "loaded". anything time-critical or temporary could be pushed into RAM, but the bulk of the OS sits in Optane type memory.

Instantaneous ON. Instant "reboot". just clear the temporary values from RAM.

[Anonymous Coward]

Re: Software was a problem

I dunno, sounds pretty vulnerable.

[Phil O'Sophical]

Re: Software was a problem

The 1960s called, they want their core memory back.

[NXM]

Re: Software was a problem

That's how it still works in cheap embedded processors. Firmware in flash, data in ram, no chance of firmware being overwritten by out-of-control or hacked code. I still don't understand why PCs don't work the same way.

[Christopher Reeve's Horse]

Re: Software was a problem

Haven't used a computer (knowingly) with the OS hardcoded on silicon since my Atari ST. Can imagine how much fun Microsoft's patch Tuesdays would be if they had to post out new chips to customers on every update!

[Tron]

Re: Microsoft's patch Tuesdays

The point being that you wouldn't need all those OS patches. If it worked when you turned it on, it would keep working. Any tweaking would happen in software. It's analogous to a system never being online not needing protection from online hacks.

If anyone had the cash they could build a new system design (mobo and OS) that excluded hacks by design. It would use a core OS on ROM and a more interesting mix of memories with strong read and write rules. You could even call code portions from ROM in software, making software much smaller. ROM-based stuff could be on cards/cartridges. The switch from 8/16 bit OS on ROM to OS on RAM/disk was an obvious path but opened up a hornet's nest that we are still living with. We may need a redesign exploiting a hybrid of current tech and past designs to deal with contemporary issues. But all the cash is being wasted on AI.

[Anonymous Coward]

Re: Microsoft's patch Tuesdays

> It would use a core OS on ROM and a more interesting mix of memories with strong read and write rules.

It looks like you're describing UEFI, or it's precursor our old friend BIOS.

[doublelayer]

Re: Microsoft's patch Tuesdays

The reason we patch operating systems isn't because something got in and rewrote them and we need to undo that damage. It's because they had a problem and we want to fix it. If we can't rewrite that bit, we can't fix problems we find in it. All your OS versus software quibbling does is to lock part of it down, preventing us from fixing that part, and change what you're calling the parts you can write. Those things would be as vulnerable to rewriting attacks but as easily patched as operating systems are now. Your suggested change would have no positive effects and negative ones on any part you actually made impossible to change.

You can't exclude all hacks by design. You can include a lot of hardware security layers which are difficult to get around, and that would help a lot with security. As with most security, it would also make the devices a lot harder for people to use as they like. Security and convenience often oppose one another, and the solution that introduces tons of guards on security tend not to be purchased by people wanting to get their work done today. I speak as someone usually pushing the more secure option. A lot of attention is needed to keep those safeguards from preventing users from doing what they need the computer for, and as your solution already mistakes how the security part would be implemented, I have doubts you're considering the convenience elements properly.

[Heya]

Re: Software was a problem

How much flash do you provision, and what happens when 640kB is not enough?

Your cheap embedded doohickey is cheap, embedded and inflexible. Horses for courses.

[Anonymous Coward]

Re: Software was a problem

Swings and roundabouts.

[Jou (Mxyzptlk)]

Re: Software was a problem

For embedd you have a limited specific scope. ROM might be 16k or less, RAM might be 4k or less, and the "disk" storage might bei 4k or less. Even if you 16-fold those numbers you are still in a specific range where you can verify the code much easier. If you 128-fold those numbers you reach the region where Windows 3.11 might run on such an x86 embedd CPU.

[Anonymous Coward]

Re: Software was a problem

They can, but nobody bothers with it.

If you log out then "hibernate" you achieve exactly what he's after. Logging out closes all running apps and sends you to the login screen, then hibernate dumps the "login window" state and reloads it when you power on again. Rocket Surgery it ain't.

[Someone Else]

Re: Software was a problem

I can imagine it...I actually lived it.

The memory technology was called "core"; you know, those teeny, tiny doughnut-shaped ferrite things you may have (or probably did not) read about in technology history books. Core based systems really did remember where you were, and would come back up after a (even unexpected) power down and just pick up where they left off. Yes there was some derring-do about sensing that the power was going off, and storing registers in a key known place before the machine fully lost power, but heck, that was easy (I believe capacitors were involved...). Thing is, core was slow, and with RAM access times 2, 3 and soon 10 times faster, everybody wanted the speed, and...well, we'll deal with the entire system reload thing as a cost of doing business.

Every once in a while, we hear of folks pining for the capability of core-based systems. Optane was one of those pinings. Nowadays, there isn't so much pining, because we're too busy doom-scrolling or chatbotting to even think about such things.

[doublelayer]

Re: Software was a problem

You speak as if the bulk of boot times is waiting for disks. Some of that is starting peripherals, including hardware in the same box as the CPU, which have actually shut down. No memory change will reduce that. We already can copy RAM to disk and shut down, meaning a relatively fast restore to the existing state when you power back up. Using Optane for that would make that process quite a bit faster. But it wouldn't be instantaneous because the hardware startup would be the same. In most of the systems where Optane would be financially viable, they don't care much about boot times because they hardly ever reboot unless they are changing something in the operating system and, if they are, they will need to reload it from scratch anyway.

[Anonymous Coward]

Re: Software was a problem

It's ok, you don't have to imagine it. In the years you've been living in a cave, hibernation modes were added to most major operating systems...some of them you can do something like this:

gnome-session-quit --no-prompt && systemctl suspend

obviously substitute the gnome-session for whatever DE you're using and achieve exactly what you're after.

I mean, if you're on Windows you probably still have to imagine it...but everything else...not so much.

[williamyf]

Re: Software was a problem

In reality, in modern computing (and especially on servers, where optane was aimed) you do not want that.

Database or KV cache in memory space persisting over reboots? A-OK. OS persisiting over reboots? Almost useless. In this day and age, the main reason to reboot a server is to apply a patch, and that, by definition, means the OS can not persist. Servers are always on, even for most HW changes.

The "save a file" part is partialy true. You still want to save files to block storage to simplify backup and data retention tasks. But other than that, no need.

[Anonymous Coward]

It failed because it was 10x the price of NAND and RAM at the time. Let me guarantee you that it would be 10x the price of NAND and RAM now...

There is a shortage of all kinds of storage at the moment.

[xyz123]

You're correct. Even tupperware storage has more than doubled in price.

I presume from all the AI datacentre builders bringing their own lunch to work.

[Anonymous Coward]

Tupperware costs a fortune because it's now owned by private equity. Educate yourself, you're literally handing your money over to US billionaires who are laughing at you.

https://ledger.worseonpurpose.com/brands/tupperware

[williamyf]

You forgot the joke alert.

Having said that, 2 months ago I went to my MBA 20 year reunion, and people from all over were complaing about inflation. A result of all the (needed and justified) COVID liquidity injection for stimulus and benefits/bonuses.

That is the explanation of the increase in price of plastic pots

[xyz123]

Intel should open-source Optane for zero cost.

Someone else develops and uses it for AI to replace hefty DRAM useage, DRAM prices fall...more ordinary customers upgrade their PC and buy an Intel CPU.

Intel wins, the optane person wins...there's no loser

Intel gains a LOT of Kudos and loyalty from customers for help with ram pricing.

[Anonymous Coward]

That wouldn't solve the cost problem though.

Optane was always something that was on my radar, but the cost of it made it very unattractive. Had it been priced appropriately, it would have been a no brainer...but instead it was priced and positioned as a premium solution.

This is not just a problem with Optane, it's a problem in general with performance in the tech space...the premium should be on longevity of components instead it's on raw speed, which is dumb because usually high performance parts don't have the endurance of mid tier parts that are usually only a little slower.

The result is you have this ultra premium high cost performance tier for everything that brings you a 10% advantage for near double the price of the next tier down but also around half the lifespan...it's kinda silly.

The same thing doesn't apply elsewhere for any other products because it is generally acknowledged that "performance" is usually a trade off, not the pinnacle. Like power tools for example...yeah you can get a batshit insane drill bits that can cut through the hardest materials, but they're not built to last, they're built to be used a handful of times and are priced accordingly...you also usually get more than one in a pack to offset the durability.

[phuzz]

Ah yes, then it would be completely free for anyone that had a billion dollars worth of chip fab, just lying around unused.

[williamyf]

Agree 107% with you. And not just any old fab. We are talking here a VERY especialized fab, that would not work for anything else. Not for logic chips, not for DRAM, not for flash.

[doublelayer]

It's somewhere between very difficult and impossible to "open source" something like Optane. There are only a few companies with the ability to manufacture, let alone advance the design of, something like that, and open sourcing makes almost no difference to them as they all have plenty of money with which they could purchase it, either as a license or to own outright. Funnily enough, not one of them, not even Micron who already has the manufacturing experience, seems to have decided this would make sense, even though obtaining a canceled product from Intel who was/is in financial distress wouldn't have been that hard.

Nor would it be the miracle solution that the article and your comment suggests. I agree with the author that, if it had still been here, LLM systems would probably want it. Other than that, almost no difference. It still needs fab capacity to manufacture. That still takes a long time to build. That still gets locked into one product for a while. At best, we could have had one more fab making Optane, but another possibility depending on exactly what hypothetical we're using is that one of the fabs now making HBM instead was configured to make Optane and the same capacity limit applies. Optane didn't have a simple solution to fabrication limits.

[GlenP]

My personal Dell Inspiron has 32GB of Optane memory, acting as a storage cache. Does it make much difference? Probably not! It's not an option I would have deliberately chosen but the special offer was too good to ignore.

[RobDog]

Same here

A HP budget laptop I bought has it installed. I never knew until I had to reinstall it and needed special drivers.

[wh34ty73.pm]

I too have a Dell with Optane

Added aftermarket by me, to speed up the god awful rust disk.

And it flies. You could never tell this was a HDD, except the odd time first boot after an OS update. Night and day from what it was before the Optane went in.

[Martin an gof]

Re: I too have a Dell with Optane

But the question is, would you have got the same performance by simply swapping the rust for an SSD, probably at lower cost?

M.

[Steve Graham]

I have a 16Gb module that came out of my sister's dead Dell. I have it in a USB adapter but haven't thought of a use for it. Rescue boot drive, maybe.

[Uncle Slacky]

Not Linux-friendly

I first discovered what a PITA Optane was when I tried to dual-boot with Linux - ended up just not using the Optane storage at all (but took a couple of days of hair-pulling before I realised why Linux wouldn't boot).

[MazeFrame]

A case study in foot-shooting

What Intel managed to do here, is to create an amazing technology, and then absolutely suffocate it in corporate politics.

Having watched from a distance when Intel kept the world at 4 cores/8 threads until AMD caught them with their pants down in 2016, this is just that again.

Having to make back the R&D and fab-construction cash is fair enough, it kept early adoption slow since the new thing was also unproven and expensive. Locking this essentially magic tech to the most premium of products or the worst arbitrary SKUs killed it.

Had they just kept this cooking, updating the products every couple of years, they would be making massive bank right now. Instead, we lost this step forward to be stuck with flash.

[Nik_S]

If you build it they will come

"If you build it they will come" pretty much summed Intel's marketing strategy which relied on OS and application vendors adjusting their products for a 3-tiered memory architecture. Sadly, the software vendors had more of a "If they come, we'll build it" approach and wouldn't spend NRE adapting their OS and applications to make efficient use of it until they saw significant adoption, so at best it just became a glorified nonvolatile RAM disk with limited utility.

[gaston]

History repeats

Intel like many tech companies has abandoned winners too soon. They had external financial pressures that caused them to throw out good and bad. Yes Optane was too expensive. I am not sure how rugged Optane is but I can use cases in automotive and embedded markets. Optane seems well suited for R/T applications. Maybe someone will reinvent a similar technology a few more years down the road.

[thomasvenables]

release it again?

What I dont understand is why Intel couldn't produce it again. They have the techs and the patents. Spit a few out linked to a B580 (or equivalent) with 256gb memory instead of 16gb. If they could resist pricing it in the millions, you'd have an AI era defining card. And (one would assume) the software would follow if the price was low enough to allows hobbyist in. CUDA is the default because its easy.

[doublelayer]

Re: release it again?

They would have to sort out manufacturing. They didn't make it the first time, even though they do manufacture other devices, and the facility where Micron used to do that isn't run by Micron anymore and has been changed to do something else. So while they could make it again, they'd have to create a fab, either from scratch or by buying and reworking an existing one, and neither of those are easy or quick. One other problem is that they would also need processors to support them, and since they dropped it a few years ago, their current ones don't. They could fix that too, and likely somewhat faster than spinning up manufacturing for Optane itself, but that's another complication.