Thursday, May 7th 2020
Intel Ready with 144-layer 3D NAND On its Own, Talks 4-layer 3DXP, "Alder Stream" and "Keystone Harbor"
Intel's memory and storage products division now has a completely independent NAND flash technology development team post its split with Micron Technology, with which it was part of the IMFlash Technologies joint-venture. Intel is close to gaining a technological lead over Micron with a new 144-layer 3D NAND flash chip which will ship roughly around the time Micron begins pushing out its 128-layer 3D NAND chips. SK Hynix will begin shipping its 128-layer 3D NAND flash chips later this year. KIOXIA will put out 112-layer chips before the turn of the year. YMTC is developing its portfolio at a breakneck pace.
The 144-layer 3D NAND flash chip by Intel can handle up to four bits per cell (QLC), and can be configured to function as TLC or SLC, at lower densities. Intel will launch its first SSD based on this 144-layer QLC NAND flash chip, codenamed "Keystone Harbor," later this year. Development is underway at Intel for PLC (5 bits per cell) technology, which should drive up densities by 25 percent. Intel is also close to launching its second generation 3D X-point memory technology.
The 2nd gen 3D X-point memory chip by Intel will feature four layers on the physical media, as opposed to two on the first-gen. The first Optane product based on the new chip is codenamed "Alder Stream." The drive will launch later this year in a single-port variant, with dual-port variants planned for 2021. The drive uses an advanced new controller that features PCI-Express gen 4.0 host interface. It's likely that "Alder Stream" will be part of Intel's big next-gen Xeon "Sapphire Rapids" processor and "Eagle Stream" platform launch, which features PCIe gen 4.0.
Source:
Blocks and Files
The 144-layer 3D NAND flash chip by Intel can handle up to four bits per cell (QLC), and can be configured to function as TLC or SLC, at lower densities. Intel will launch its first SSD based on this 144-layer QLC NAND flash chip, codenamed "Keystone Harbor," later this year. Development is underway at Intel for PLC (5 bits per cell) technology, which should drive up densities by 25 percent. Intel is also close to launching its second generation 3D X-point memory technology.
45 Comments on Intel Ready with 144-layer 3D NAND On its Own, Talks 4-layer 3DXP, "Alder Stream" and "Keystone Harbor"
I probably will use such adapters in my future builds, just to make it more convenient to pull them out. The M2 form factor might make sense in laptops, but in desktop they are just a pain.
When I build computers, they are usually in service for 8-10 years, but their role changes over time. And this is one of the biggest strengths of desktop computers; many things can be adapted or changed, each usable part can be reused somewhere else. So graphics cards are usually swapped a couple of times, SSDs and HDDs swapped a lot, etc. Then it's a pain if I have to unplug half of the stuff to swap an SSD, especially if I'm troubleshooting something. I'm also tired of small cases, so unlike londiste, I will do my future builds in spacious towers, do only basic cable management, etc. Computers to me are not meant to be pretty, but serve as tools. ;)
Depends on what you use the computer for but I am noticing that I swap parts and do random testing and troubleshooting less and less every year. There is mostly no need for that. Swapping GPU is the same regardless of case size (excluding some extremes like my A4SFX maybe), CPU is more likely to be replaced with motherboard and when it comes to drives how often would you want to swap those and why? For testing HDDs/SSDs, a USB dock has been my go-to method for well over a decade now and troubleshooting M.2/mSATA etc drives is PITA one way or another.
Edit:
For mass storage, NASes are pretty good these days. I would not run games or storage-dependent work off of them but a cheap enough Synology box for example will nicely saturate a 1Gbit Ethernet bandwidth which isn't much slower than what drives are realistically capable of.
I would bet constant testing and swapping and adapting is pretty niche thing in itself :)
Click the spoiler for details.
Back on topic,
QLC is bad enough, but with PLC I would be really worried about file integrity. I would certainly put OS and other stuff on separate drives, but this still sounds scary. I do wonder if controllers will move things around more to reduce data rot. Still, I think NAND flash is in dire need of a replacement.
What I am a bit vary of is that 2nd gen XPoint might actually get a nerf of some kind. XPoint itself is overspecced for anything it is used for today in consumer or enthusiast space. The exception is probably the Optane DIMM technology. NVMe controllers cannot seem to use XPoint speed and latency to fullest. This leaves some space for simplification or relaxed specs.Depends on what you use these for. 5 years is a long time when we look at how SSDs have evolved. In 2009 the reasonably priced SSDs were 120GB, maybe 240GB with especially write speeds that are low by today's standards. Manufacturers have learned from their mistakes for the most part when it comes to SSD endurance. Today, I am quite convinced that endurance rating claim in manufacturer spec is legitimate, at least for major manufacturers.
Price remains a factor and QLC has its place. I have a 2TB 660p for my secondary drive (mainly for games) which I got for ~160€ last year. Intel spec says it is good for 400 TBW and in about 9 months I have put 10TB on it. If what Intel claims is even remotely accurate, I am fine for a long while. The downsides of QLC I can live with in this particular configuration - unless I fill the drive to 95+%, it's faster than a SATA SSD. Filling the last 5% is indeed very painful but awareness of that helps :)
In fact, where does all this "QLC is unreliable as shit" attitude come from? Are there any real tests or experiences published that I'm not aware of?
If QLC is used for longterm storage, you barely write to it anyways and if used in a redundant array with a proper file system in a NAS, you can scrub your stored data to detect and repair flipped bits as often as you like? Nobody said that QLC is good for every application...
QLC definitely has much less write cycles compared to TLC. In a write-heavy environment this is important. For consumer use cases, might not be too crucial. The other downside of QLC is its speed, particularly write speed. Due to the accuracy of voltage levels needed for that many levels, writes are painfully slow. The approach controllers and drives take to mitigate this is the same as with TLC Flash - SLC cache. Some part of the drive is treated as SLC and that allows very fast write speeds. After writes to the drive finish, drive itself will move the data around to QLC and frees up the SLC cache for more writes.
While I can't claim to have the empirical data to support a scientific conclusion, the trend I'm seeing is really concerning. In the past couple of years I've seen a decent number of SSDs which become very unstable after 2-3 years of use, and most of them nowhere near their endurance rating. I've seen a lot of HDDs go bad in the last 25 years, but nothing close to this failure rate.
There are at least two major problems with SSDs; stability of the cells and data rot. Especially data rot gets a lot worse when they cram more bits into each cell. I wish SSD specs would list what the controllers do to deal with data rot, if they do anything at all.
The cheapest ones are the slower SATA based Samsung 860 QVO and ADATA SU630 around 200€ for 2TB. Still too much.
SATA based 2TB QLC should be 150€ max and PCIe versions should be around 200€ like they were last year.
The retailers are ripping you off.
I see $239 and $369 respectively.That 665p price is a bit high doesn't make sense when you look at the 1TB being $139
www.newegg.com/intel-660p-series-2tb/p/N82E16820167461
www.newegg.com/intel-665p-series-2tb/p/N82E16820167470
US pricing is meaningless to me as it does not include VAT and shipping costs.
But still $239 is too much for 660p.