Thanks for the memory: XPoint put under the analyst microscope Semiconductor market researcher Jim Handy of Objective Analysis has produced an 80-page report looking into what XPoint memory is, how it could be used and what its prospects are. XPoint memory was unveiled by Micron and Intel in July to general amazement. The pair claimed it was 1,000 times faster than NAND, ten times denser …
Does anyone know if this has any sort of maximum tbw like flash? If not and it's near DRAM speed, I dunno, why not use it entirely in place of DRAM in slower mobile and IoT thingies. Unless it's way more current hungry. Personally I'd leap on using it for a total replacement for DRAM if taking the power away paused devices. All kinds of difficult issues in computing get reset. The conceptual frameworks in operating system and compiler design need revisiting. Again.
But I don't recall seeing the article mention how this report got it's information? I'm assuming it was either design docs or an engineering sample straight from Intel? I can understand them not commenting on the actual materials used (probably under NDA), but it doesn't seem like we've learned all that much? Well I guess it's too early days to expect any real benchmarks or use cases (and even if there were they wouldn't be indicative of the final product). Still the 1000x this and 10x that speak makes me skeptical of how trumped up they are. Especially in light of the paper launch so far in advance of it shipping.
I can see one very useful application for this in (true) RAID controllers and generically as a persistent write cache for slower SSD and spinning disks. These applications already exist using different technological means to achieve their goals, and I guess that this new form of RAM will streamline a lot of disparate uses and cause some interesting unifications.
Examples of what I'm talking about:
* "real" RAID controllers have battery-backed write cache that can ensure data consistency by flushing the cache after the system power failure is sorted out
* likewise, many (or at least some) SSDs use a faster form of flash as a write cache
* similarly, hybrid SSD/HDD drives use the faster flash as a write cache (and maybe read cache, too)
* using ext4, you can create an external journal that's stored on an SSD (or similar fast, persistent storage) to get around 2x to 3x better write performance (and still maintain quick crash recovery)
Basically anything that includes any sort of write-ahead log (including databases) should get a very nice boost in write speeds, as least for bursty writes. You'll always be limited by the write speed of the slowest device in the chain if you're doing sustained writes, but with a large enough cache many workloads will never fill it up.
I hope that this ends up being something that ends up just being another resource (like RAM and SSDs) that operating systems will be able to arbitrate the use of rather than it just being closed off within specific bits of hardware like disks and such.
This. And every other instance of battery backed state information. I like building around higher power requirements that once the burst is done, the device/system hits the snooze button and waits for the timer or external interrupt to occur. Oh, and did you notice how stackable the reference graphics make it appear? Density and power requirements are needed to make a comparison with 3D Flash. And, of course, costs/unit/GB.
I love how they're teasing us, Not!
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
