Oh my Tosh, it's only a 100TB small form-factor SSD, SK? The Flash Memory Summit saw two landmark capacity announcements centred on 96-layer QLC (4bits/cell) flash that seemingly herald a coming virtual abolition of workstation and server read-intensive flash capacity constraints. Pop art style illustration of man exclaiming WHAT? in shock/horror/bemusement. Anyone fancy testing …
We see lots of announcements like this here but as usual, little in the way of price is mentioned.
Sigh.
Sigh.
and
Sigh.
Come on el reg... How about not reporting this stuff unless the things like price and availabilty and interface type are mentioned. Otherwise these pieces are pure Marketdroid fluff and worthless.
I think the price is "read-intensive".
QLC 3D NAND just doesn't have the durability for lots of writes, at least in its current forms, so whilst it seems suitable for archival purposes I can't see it really having much use in more general server and workstation workloads.
In view of the relatively lower cost of MLC, perhaps extreme redundancy - let's say up to 10x over-provisioning - might bring the write endurance for an SSD device up to more acceptable levels but NAND, especially MLC types, rely heavily upon error correction and this might become a limiting factor - the SSD device may end up having to spend too much time monitoring, managing and correcting itself.
"Quad level" is a misnomer: with 4 bits per cell, that actually means 16 levels for the electronics to distinguish. So with 16 levels and the cells getting ever tinier, how many electrons does each cell store? Divide that by 16, and you know how many electrons are used to store each bit. Does anyone know what the number is?
For 20nm planer flash the number of electrons per level dropped to under 10. 3D nand uses a larger geometry (40nm?) so there are more electrons per level. It looks like about 20 on the graph near then end of this. As far as I know the geometry is not shrinking any more and the extra capacity comes from more layers per die, more dies in a stack and putting the control circuits on a separate die instead of around the edge so the whole area of the die can contain layers of cells.
"So with 16 levels and the cells getting ever tinier,"
They're not. They stepped back from that and got got bigger when the fabs stepped from planar to 3d, because shrinking cells resulted in slower NAND with substantially lower durability. Cell size hasn't shrunk since, which is why there's all that emphasis on more layers and more bits per cell.
History is littered with "No one will ever need {enter new capacity here}" statements, whether it's speed of travel, range, power of weaponry or computing power. So far, someone's always found a use for more. If you'd asked me even as recently as 2005 whether I personally would really need and exploit a computer with eight cores running at 5GHz, stuffed with 32Gb of RAM and 8Tb of machine storage, I would have laughed. And yet right now I am speccing out 256Gb RAM workstations, because of the difference it will make in my work. No, it's not mainstream office productivity, but I am not a rocket scientist either. Elbow room is always good to have, and many of the things we do will continue to add orders of magnitude to the usefulness of simpy storing data.
I absolutely can see why you make your statement, mind you: I can't currently imagine a legitimate use for 64Tb of local storage ... but ask either of us again, in five years.
As for—
"Actually, I'm waiting for someone to invent a storage system that uses long transparent crystals. That would be cool."
— sounds like a cri de coeur to go in the book along with the flying car, moonbase, International Rescue and trips to Jupiter: things we were promised in the 60s and 70s. I remember the crystals from the original Christopher Reeve Superman (1978?) ... let's hope photonic processing and memory doesn't disappoint. (If you could stop wasting time talking rubbish on Twitter and undermining the west coast like an obsessive termite, Elon, I have a better investment for you ....)
In all fairness, a lot of resource requirements these days come from lazy developers who don't understand shit about optimisations.
I only have 16GB of RAM in this machine, and since I switched to Vivaldi (Chromium) as my primary web browser (which seems to require two to three times the amount of memory per page than an older version of Firefox), I find myself running out of memory more often. Ditto for SmartGit (hello, Java) which uses a laughably high amount of memory for what it does.
And look at how much storage space the average application needs these days. It's insane.
There are some applications that obviously need a lot of memory and storage space - databases and video editing are two that spring to mine - but there's a hell of a lot of waste these days. Won't developers think of the children/environment [delete as applicable]?
</rant>
</oldman> <-- Does this mean I'm dead now?
lazy developers who don't understand shit about optimisations
I've dealt with my share of lazy developers, but I'll bet that most devs understand a fair bit about optimisation, and I'll bet a lot of them even care. They do, however, have to report to management who care about getting it out of the door now. Not in three weeks when it can be 10% faster.
Oh so true.
How long is the 'Technical Debt' list then?
Do you ever get a chance to remove items from it?
Most places seem to never get around to fixing things but concentrating on the new 'shiny-shiny' features that less than 1% would use. (Hey MS I'm looking at Office here)
I've dealt with my share of lazy developers, but I'll bet that most devs understand a fair bit about optimisation, and I'll bet a lot of them even care.
Perhaps lazy wasn't the right word. But in all honesty, I don't know many developers who really understand shit about optimisation these days. Certainly there isn't the same understanding about what a compiler does to your code that there used to be. And as more and more higher level languages become more and more popular, people really don't understand the implications of what happens when they write something one way instead of another.
Back when I was working at [a rather large game developer] not even 10 years ago there was a pervasive sentiment among most of the developers* that they write code, and someone else will optimise it. And to be honest, I haven't seen any differences in attitudes at other places I've worked since.
While it's nice to think people do actually care, in reality most of them don't. The tsunami of shitty Android and iOS apps eating your battery is testament to that.
* There were a couple of *really* good developers there who basically spent every stand up meeting trying to drill into everyone else: Use less memory; Profile your code. (This was for a last-gen console game.) I even ended up writing a memory profiler** to help show people what the consequences of their actions were.
** I've written another once since. If you ask nicely, I'll share a link. ;) Not going to spam it at you right away.
I learnt that early in Uni. An assignment asked us to brute force RSA ciphertext with increasing key sizes. Top marks went to the student who got furthest through the assignment (on like for like hardware). The winner was a guy who used a slightly out of date JDK, but his javac version compiled significantly faster code (c.20%).
Don't expect computer scientists to come to the rescue. From what I've seen, the way coding is taught starts with conceptual elegance, and does not always end with operational efficiency.
Take objects and recursion. Conceptually clean, operationally voracious. Both are abused like catholic altar boys: much code is neither elegant nor efficient. And so when it comes time to optimize, it may be that 'you can't get there from here.'
Quote: lazy developers who don't understand shit about optimisations.
Define optimisations!
Our current client has asked us to optimise our code. The remit being to maximise throughput, whilst minimising DB and network (including drive) access.
This means doing as much as possible in memory, by basically creating in-memory caches for anything that can be cached. Memory is cheap, relatively speaking.
Therefore in this scenario, optimisation means an increase in memory use, not a decrease.
I refer the honourable member to the statement allegedly made by Mr William Gates in 1981 "640K ought to be enough for anybody", often paraphrased as "No-one will ever need more than 640K".
In all fairness to the said Mr. Gates, the remark would have been made in in the context of a machine architecture that couldn't address more than 1MB in any case, so his figure of 640K wasn't completely ridiculous, at the time.
It was still wrong, though, when there were machines like some of the Apricot and Sirius computers that could run DOS with 896K of free RAM.
> f you'd asked me even as recently as 2005 whether I personally would really need and exploit a computer with eight cores running at 5GHz, stuffed with 32Gb of RAM and 8Tb of machine storage ...
Depends. Is it expected to be running windows?
I'm past the half-way point on 64 TB and you'll not find pr0n, nor even large multimedia collections (video or music). The former doesn't interest me at all and damn little of the latter is even interesting. A huge chunk consists of datasets, analyzed results, and reference material (books, papers, &c. collected since 1975). I'd love to have three 64 TB drives, each from different manufacturers, just in case, and call it done. On my current medical trajectory, I'd just about fill that before being planted in a national cemetery being mowed at government expense by a bunch of fellow veterans also with substance abuse issues ;-),
BTW: My mom has promised that I'll be planted with my favorite computer to go down eternity's road. Serious.
“ long transparent crystals. That would be cool.”
I had a thought the other day, suppose a civilisation advanced enough that it could store data in stone tablets rather than crystals. Stone tablets always seem to survive the civilisations that created them and are always being unearthed in archaeological digs thousands of years later. I feel a sci-if story coming on.
More to the point, the 2.5inch format is about the largest form factor you can make and still reliabily dissipate the heat (and that really only in 7mm thickness)
3.5 inch SSDs keep on being announced, but are endlessly plagued with heatsinking issues. The format worked for spinny things but it makes zero sense for solid state and trying to force the SSD square peg into that round hole isn't a good idea.
I'd like to announce my MagicVegaLeapDrive which will pack 300TB into a 1.8" form factor which will be fully compatible with all known interfaces and drive bays.
Pricing to be determined.
Investors please contact the usual address and leave your money in the burning skip out the back.
Pfft! Only 300TB?
I have some 8TB SDXC+ cards here.
Only £39,999.95!
Handy for those times you need to back up stuff *right now*, they have 8 way parallel write and the adapter will plug right into gigabit LAN/mSATA3/etc with the Cortexiphan (tm) onboard adaptive neural network compression chip so the data gets compressed on the fly before committing to Z-NAND.
Chips encased in nano-diamond for the ultimate in advanced cooling, also includes internal X-ray and gamma radiation shielding. Made from isotopically pure 28Si for ultimate efficiency and freedom from point defects that plague lesser manufacturers.
Got just over 6GBit/sec sustained write and card runs at just over 33C when writing.
(Note: may not work in some older devices but will usually read fine)
Not at 32K rez VR interactive it won't be. Now there's a thought for the future history books.. "The ACB - Affordable Consumer roBot only became possible with the explosion in demand for the PPB - personal pornbot.".
"Once people figured out how to make the PPB shower & dress it wasn't long before they got sent to do the shopping."
There'll be tabloid headlines: "My ACB got hacked! Woman demands law against foul dark web menace that turned her ACB into a PPB whilst shopping. A supermarket spokesbot made soothing noises but this tabloid can exclusively reveal the SSB was also hacked and the soothing noises it made were not those officially sanctioned and that all cucumbers purchased by unaffected ACBs are subject to recall."
I'd better stop before this gets completely out of control!
"just 5 computers per household (circa 2018)"
Are you sure? My cheap and nasty smartphone can run rings around the sun workstation I was using when I started work. My wife and kids have more expensive phones than me, so we're up to 5 computers already.
We've got two "smart" TV's, although for the smart stuff we use Raspberry Pi's plugged into them, so that's 2 more. Then there are another 4 raspberry Pis getting used for "tinkering". Does the Einsy board in my 3D printer count as a computer? What about the microcontrollers in our 3 printers, or the ones in the washing machine, the dishwasher, the microwave and the toaster?
Oh, and we've got 5 laptops in the house... and there's 3 tablets.
Have I got up to 5 yet?
Edit:- I forgot about the Router and the two network disks, all of which are running Linux variants.... so there's another 3.
If we use the current consumer price of ~$2.50/GB for mid-range performance SSD's these units could push the price barrier. I expect that configurations in proprietary arrays or part of advanced Hyper-Converged Infrastructure platforms will raise the $/GB even as the greatly increase the performance and scalability of these platforms.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
