I didn’t know 3200MT was the optane spec… My PC I really really loved was 3200MT DDR… o_o I have a LOT of things to look up
Good, because I was just trolling you 
Sorry, it was just too tempting. Happy to see everyone has a sense of humor here, within reason at least.
It will be the first and last time I succumb to the troll urge here - but to be clear, no harm was intended
i welcome this with open arms , unfortunately its pSLC not true SLC , so it will be slower than optane for sure , closest to optane we have now is kioxia XL-Flash , first gen was the FL6 , maybe next yr we get some goodies from the XL-Flash team . right now Optane still leading by quite a margin , close second is XL-Flash ( second gen ) and then SLC drives
a few reading stuff for XL-Flash
XL-flash is mostly just normal NAND flash that Kioxia slightly tweaked. Kioxia recently started to brand some of their MLC NAND as “XL-FLASH” and further diluted their performance moniker of NAND.
NAND is kind of a dead end in the write latency department, it won’t ever achieve the performance of Optane; but manufactures will make ever more sophisticated controllers with heaps of RAM that interleave more and more NAND channels together to try and compensate for the inherent lack of performance of NAND.
We do have an Optane replacement in the pipe in SOM (SSM) on the horizon though. Performance metrics have started to be released at the VLSI symposiums: 15 times faster write latency than Optane, 100,000,000 P/E cycles, multi-decade data retention times and most important, it will be more scalable than Optane by removing the entire selector layer stack to greatly simplifying manufacturing.
1 Like
This is the first I’ve heard of SOM. Any links where I can read up on this?
https://scholar.google.com/scholar?q=selector+only+memory
https://research.skhynix.com/publications
See also Micron SXM.
seems like a tech 7yr+ into the future and no gaurentee that its gonna be used for ssd , micron is focusing the tech for their CXL division
If the tech is that good, I’d say the PCIe-NVMe-SSD interface-form factor combination is limiting. Perhaps CXL×SSD is a way forward for the traditional storage paradigm. But that kind of performance, density, and endurance is a mismatch for today’s computing paradigms.
I wonder if it is possible to make an adapter for the optane dimms to get them working with cxl.
I really like your quote from Animal Farm. I am surprised anyone else knows the quote, let alone has read Animal Farm.
1 Like
Do anyone have any info regarding XL-flash / SCM gen5 thats gonna come out soon ? or is it gonna be a bleak future ? I was looking forward for some new info , last yr innogrit showcased an XL-flash 2nd gen5 ssd and that was it !!! someone give me some hope
Finally , some news lads !!! KIOXIA GP Series SSD Extends GPU Memory with XL-FLASH for NVIDIA Storage-Next AI Workloads - StorageReview.com - Kioxia GP series , the FL6 successor !! , SLC and MLC unforunatley , i read somewhere that 800gb SLC and 1.6tb MLC
3 Likes
So like the thing Solidigm P5810 (formerly Intel) had for a while then?
Kinda , but faster speeds and lower latency , but the P5810 wasnt even close to optane latency , xl flash is closer and the second gen seem to close the gap , but fundamentally its stil nand and DWPD sill be 60 i assume
XL-Flash real world latency being around 12-13us for read and 4-5us for write is pretty good, and definitely could be an Optane replacement. Since Optane storage SSDs was around 20us latency.
The tech itself doesn’t match 3D XPoint itself though at all, which is what Optane is built from. The main bottleneck seems to be the storage and communication bus formats themselves now. As XPoint had around 100ns latency, but once put into a usable drive this became 20us. When used in a dimm as memory XPoint maintained the 100ns latency spec.
So it has me curious if XL-Flash has a bit more hidden potential that is held back by current bus tech and processing, just as Optane itself was, and whether Hynix’s XPoint replacement will do any better or be stuck at a similar threshold in a real world usable drive.
1 Like
Looks like it is…
2 Likes
But does it have the endurance too?
That was the advantage of 3DXpoint: Endurance + Performance
Flash seems to be stuck at the performance* part.
*Sometimes, with controller trickery, on closed course, when the moon and overprovisioning are just right…
1 Like
And long term data stability as well. NAND looses voltage as it ages which is why we need to refresh the cell data periodically otherwise data can read VERY slow as it runs through a lot of error correction, or be corrupted entirely. Optane had no issues like this since the voltage doesnt decay over time.
At least being a Kioxia drive meant for enterprise I assume that their controller firmware does periodic cell refreshes on its own in the background, so the drive simply needs to be powered on every so often to retain data integrity.
Despite this still be NAND flash in the basic concept, XL-FLASH like this drive uses is different at the structural level. The bit and word lines are drastically shortened over typical NAND design which lowers latency. XL-FLASH also has wider design so more data can be written/read in one cycle, which increases throughput. So the performane really should be there.
I personally dont see how shortening word lines can truly result in a 10x reduction in latency, as the total length is still short in relative terms, all of it is still on package. So I am guessing that there are other optimizations to the actual word line layout built from the ground up like this that brings significant improvement over the very old design that NAND has. Lets hope some of the tech outlets do a true deep dive of the design with XL-FLASH like they did with Optane. Id be very interested in seeing it.
2 Likes
asianometry just dropped a little video about optane. kinda retreading old info, but pretty interesting none the less.
1 Like