Intel have announced inventing a new type of memory technology, the first in decades. They claim it has 10 times the density of existing SSD memory; they are starting fabrication at 128Gb per dye. Intel’s press release says it is three orders of magnitude more durable as well, and three orders of magnitude faster.
This is a huge deal. With this, terabytes are the new K (or new M, if you prefer). But let’s leave that aside for the moment.
That speed means that this new RAM is faster than current system RAM, by about an order of magnitude. Which means there will no longer be an important difference between system RAM and long-term storage. The whole idea of long-term (disc: large, but slow) vs. short-term (RAM: fast, but small) storage becomes completely artificial. It’s all large, it’s all non-volatile (meaning stays when powered off), it’s all fast.
This changes everything about software development. If – and it is a huge if, like, this huge:
…what they’re announcing is true, there hasn’t been a breakthrough like this in the field in decades. This isn’t “better USB drives.” This is a new universe.
I mean, what do you compare this to? The hard drive, maybe? It kind of undoes the hard drive as a separate device, but I don’t think that goes far enough. It resets so many basic ideas about software and hardware development that you may in fact have to go all the way to the very concept of interactive computing to get something bigger. (Interactive, as opposed to batch, where you submitted code and waited for printed output to see what happened.)
No, Really, It Actually Was Like This
And this is just the first generation. At 10 times current storage density, that’s a big skip ahead in Moore’s Law (doubling every two years? Nah, let’s double three times next year alone, instead), and is getting close to “all the storage you want forever.”
You’re also seeing a massive elimination of fragility. That has tremendous value in and of itself. People think of the internet as forever, but that’s not true – that’s only for things that enough people care about to store individually that the resulting redundancy makes up for the intrinsic fragility of previous computer storage systems. 1000 times more durable than existing SSD puts it far, far past the lifespans of magnetic media – it gets you into paper range.
I have archive drives. They are about to be utterly obsolete. The whole concept of ‘delete’ is now pointless, except organisationally.
This changes the way file systems are written. This changes the way processors are made. They’ve got a bus that can handle the RAM bandwidth internally, so presumably – hopefully! – that extends out to a motherboard system bus. Think about that. The first products are going to be PCIe storage expansion, but that’s just because PCIe is the fastest interface out there now. It’s far behind what this new memory type can deliver.
I mean, once you’re designing for this kind of storage, do CPU caches have a point anymore? I don’t even know. (And as an aside: VMs become hilarious. SURE HAVE SOME WHY NOT) If the RAM bus that will go with this is moving at the speed of the RAM – which is, again, faster than anything we have now – then the only savings in the end is raw distance and speed of electron travel. That’s certainly not zero, so may make it worthwhile. I don’t know.
In my head I’m seeing processors basically treating all of storage as on-chip cache, and doing it with as many processors as you want. Hell, do new types of processors optimised for this storage paradigm even need registers anymore? I imagine so, but it’s a question worth asking.
So, yeah. If they deliver, you can short Seagate, short Western Digital, short Hitachi – or, as Fishy said, “short Thailand.” In this environment, what even is magnetic media lol other than “It’s been fun, guys; you are now niche players.”
Test quantities this year to developers. Product next year. PREPARE FOR REBOOT!