News Micron to Sell 3D XPoint Fab and Cease Further Development Efforts

[Admin]

Micron announced that it is exiting the 3D XPoint business entirely, with plans to sell its fab by the end of 2021 and cease all 3D XPoint product development.

Micron to Sell 3D XPoint Fab and Cease Further Development Efforts : Read more

 

[Kamen Rider Blade]

=:::::::::::

 

[Chung Leong]

Hopefully Intel will buy the fab then start adding support for persistent RAM to their consumer CPUs.

 

[eklipz330]

the death of an incredible technology. sucks.

 

[Dsplover]

Just waiting for the 905P 960GB M.2 to drop.
I use a mixed batch of seq reads and random seeks.
Would love to see the effect of the high random seeks and extra low latency these offer. Just not at the current price.

 

[cyrusfox]

Just waiting for the 905P 960GB M.2 to drop.
I use a mixed batch of seq reads and random seeks.
Would love to see the effect of the high random seeks and extra low latency these offer. Just not at the current price.

As far as I can tell they never released an M.2 drive above the capacity of the M.2 380 GB. If you want large capacities need to go with U.2 or PCIE card for that size.

 

[Dsplover]

SSD w/ M.2 adapter. My bad.

 

[escksu]

Just waiting for the 905P 960GB M.2 to drop.
I use a mixed batch of seq reads and random seeks.
Would love to see the effect of the high random seeks and extra low latency these offer. Just not at the current price.

I have to tell you that you won't see a big increase in performance. Although the 905P does perform very well on benchmarks, you won't see your programs loading 2x faster or anything like that. You will only see a slight increase instead.

This is due to inefficient nature of windows I/O itself. So, hardware can't overcome this inefficiency.

 

[Dsplover]

My programs are loaded into RAM like the OS.
Everything else is streamed from an NVMe or SSD.
The target buffers are RAM based, so the extra random seeks would increase my polyphony.

I get decent results as is, but when I use high polyphony voice stacking sometimes there are dropouts.

We use to avoid this years ago by having multiple 10k Cheetah SCSI Drives.
The theory was more seeks, more polyphony.

SSDs changed that to a degree. I think the extra dough might alleviate these occasional Pops. I’m the only one who would likely hear that, but I do strive for perfection.

Cheerz

 

[danlw]

Sounds like 3DXpoint is going to join things like IPv6 and Dvorak on the ash heap of great ideas that didn't get adopted because the older technology is too entrenched for the change to actually happen.

 

[Kamen Rider Blade]

3DXP totally deserves to be the dominant drive for OS usage only across the industry.

Too bad Intel pushed it in the wrong way and tried to fit it into a niche to combat NAND Flash directly.

It's super low latency and consistent latency regardless of how full the 3DXP drive gets should be pushed and marketed as a OS Drive.

Doesn't matter that their cost per GB is higher than regular NAND Flash.

Normal productivity programs and OSes don't need that much space.

Games can be shunted onto the NAND Flash drive.

Regular NAND flash has a perfect place as fast & expensive storage.

HDD's are Cheap & Slower Bulk storage.

 

[TerryLaze]

Hopefully Intel will buy the fab then start adding support for persistent RAM to their consumer CPUs.

Sadly intel already stated that they will sell it off last year.
https://www.tomshardware.com/news/intel-sells-nand-fab-ssd-business-sk-hynix-9-billion-dollars

 

[JamesJones44]

There goes another US fab (unless Uncle Biden compensates the Taiwanese or Chinese buyers for wage differences...), all the while we complain about chip shortages day in day out.

 

[CerianK]

3DXP totally deserves to be the dominant drive for OS usage only across the industry.

Too bad Intel pushed it in the wrong way and tried to fit it into a niche to combat NAND Flash directly.

I almost shed a tear over this news... hopefully it is purchased by someone that understands the full potential and importance of the technology.

 

[hotaru.hino]

I figure 3DXP would be best suited in areas where high IOPS is particularly important, but the only use case I can think of at the moment is in enterprise class servers and data centers. It doesn't make sense in the consumer grade space since IOPS isn't really a problem there.

But 3DXP just has the problem with any other engineering solution: if it isn't significantly better in practice or cheaper than what already exists, people see little point in investing in it.

 

[Giroro]

It's still hard to believe that they never even tried bringing 3dXpoint into a traditional consumer product. At least outside of Intel trying to put their worst optane drive controller on a board with their worst flash controller - H10 was still basically two drives in glued together into tiered caching... but by that point their 16GB optane drives were on an old controller that was being outperformed in most metrics by almost every other SSD. It was like a beta or prototype product that left anybody interested waiting for "the real thing".
Was their some fundamental technical reason that they couldn't make a decent controller for those Optane cache drives? Is 3D XPoint really so expensive to produce that they couldn't even try to make a decent capacity drive at a price that was remotely affordable? Is it too hard to design a controller that can natively use both flash and 3D Xpoint?
3D XPoint seems like the ideal technology to replace the SLC cache and maybe even the DRAM in flash based SSDs. It would definitely boost the performance of DRAMless drives. It should be in every SSD on the market so it's a real bummer that Intel focused development on trying to make Optane an exclusive server-only product. I guess it was too late by the time Micron got a chance to do something with the license.

Hopefully, whoever buys the tech will put it to good use.

 

[jchang6]

there was never much demand for 3D XPoint because the number of situations in which it has impact are few, and often contrived. Intel should have realized that system performance can be improved with greater effect with low latency DRAM, which might entail bringing DRAM inside the processor package - also at some set size say 4-8GB, and this might have made for a better case for 3D XPoint as tier-2 memory

 

[supremelaw]

In the interest of scientific experimentation, if nothing else, I would like to have seen a few radical enhancements to standard server and workstation chipsets, to allow a fresh OS install to a ramdisk hosted by Optane DIMMs.

Along these lines, one configuration that came to mind was those dated triple-channel motherboards: the third channel could be dedicated to such a persistent ramdisk, and the other 2 or 4 channels could be assigned to current quad-channel CPUs.

The BIOS could be enhanced to permit very fast STARTUPs and RESTARTS, and of course a "Format RAM" feature would support fresh OS installs to Optane DIMMs installed in a third channel.

By way of comparison, last year I migrated Windows 10 to a bootable Highpoint SSD7103 hosting a RAID-0 array of 4 x Samsung 970 EVO Plus M.2 NVMe SSDs.

I recall measuring >11,690 MB/sec. READs with CDM. I continue to be amazed at how quickly that Windows 10 workstation does routine maintenance tasks, like a virus check of every discrete file in the C: system partition.

p.s. Somewhere in my daily reading of PC-related news, I saw a Forum comment by an experienced User who did something similar -- by installing an OS in a VM. He reported the same extraordinary speed launching all tasks, no matter how large or small.

 

[Kamen Rider Blade]

I hope Intel buys out the rest of 3DXPoint production and push it to the moon.

The tech really deserves to be the OS drive of all PC's, even AMD & Apple based Systems.

 

[supremelaw]

I hope Intel buys out the rest of 3DXPoint production and push it to the moon.

The tech really deserves to be the OS drive of all PC's, even AMD & Apple based Systems.

Because I'm now retired, I would like to be involved, even as a "volunteer" consultant, in a scientific experiment:

The experiment is intended and designed to test the feasibility of implementing a HEDT which supports a fresh OS installation to a ramdisk hosted by Optane DIMMs.

For example, consider adding a "Format RAM" feature in the motherboard BIOS, and related features which detect that an OS has already been installed in that ramdisk.

Perhaps the National Science Foundation would support a few computer science Ph.D. students who might find this idea challenging?

A triple-channel architecture might make it easier to support different DRAM clock frequencies for the Optane DIMMs vs. standard volatile DIMMs, as Patrick Kennedy has observed in his video on Micron's recent decision.

Patrick says that larger servers are being required to down-clock otherwise faster volatile DRAM DIMMS so that all DIMM slots run at the same frequency as the Optane DIMMs.

This does not appear to be a limitation that is insurmountable.

 

[supremelaw]

Here's a copy of my REPLY to Patrick Kennedy's video on the Micron decision:

[begin quote]

Patrick: love this video as I do all your videos.

Question: you are the 1 in a million IT experts who would be able to answer this question off the top of your head:

Have you encountered any systems that supported a fresh OS install to Optane DIMMs?

I posted a similar question at another IT website reporting Micron's recent decision.

What originally came to my mind was a re-design of triple channel chipsets, which allowed the third channel to host persistent Optane DIMMs for effectively running an entire OS in a ramdisk that is non-volatile.

In other words, using Windows terminology, the C: system partition would exist on that Optane ramdisk.

To implement this hybrid approach correctly, DRAM controllers would need to operate at different frequencies, so as to prevent the problem you described which down-clocks all DRAM to the same frequency as the Optane DIMMs.

Yes, enhancements would also need to be added to a motherboard's BIOS, chiefly by adding something like a "Format RAM" feature which supports a fresh OS install, and subsequently detects if the OS is already installed in an Optane ramdisk running on that third channel.

FYI: I filed a provisional patent application for such a "Format RAM" feature, many years ago, but that provisional application expired.

[end quote]

 

[TerryLaze]

Because I'm now retired, I would like to be involved, even as a "volunteer" consultant, in a scientific experiment:

The experiment is intended and designed to test the feasibility of implementing a HEDT which supports a fresh OS installation to a ramdisk hosted by Optane DIMMs.

For example, consider adding a "Format RAM" feature in the motherboard BIOS, and related features which detect that an OS has already been installed in that ramdisk.

Wasn't the idea that the final optane dimms would transparently show up as a system drive?
Making a ram drive would kinda negate the whole concept of bypassing the OS system calls/ not doing disk-ram ram-disk but just ram period since ram and disk would be the same.

 

[hotaru.hino]

Patrick says that larger servers are being required to down-clock otherwise faster volatile DRAM DIMMS so that all DIMM slots run at the same frequency as the Optane DIMMs.

This does not appear to be a limitation that is insurmountable.

It's not that it's insurmountable. It's just there's no real benefit. The reason why the DRAM DIMMS had to clock down is because memory controllers only generate one frequency domain for all memory. Sure you could make a memory controller that could generate different frequencies, but then you have the issue that the internal bus operates at one frequency and you have to sync up all of those different frequencies to that one. Syncing would just cause the performance to drop to the slowest thing anyway if Optane memory is considered at all times. And if you're going to segment Optane memory from DRAM, you may as well just relegate Optane to secondary storage.

If the idea is to run directly off of Optane instead of copying data over to the DRAM, unless Optane performs as good or better than DRAM, you're going to hit a performance loss overall.

 

[Kamen Rider Blade]

https://en.wikipedia.org/wiki/POWER10

https://www.devever.net/~hl/omi

This is why I'm hoping more of the industry would want to jump on the "Serialized" Memory interface vs the current "Parallel" Interface paradigm that we have now.

IBM has stated that "Serialized" memory (OMI) has a (< 10ns) latency penalty, but ALOT more bandwidth can be brought out from each memory module (200-300%) for a given speed.

And "Serialized" Memory is Memory format agnostic, so you can attach standard DRAM / Flash / Optane as you feel like.

That's a world I would like to see happen beyond IBM's little eco system.