Memory, that's a lot of changed bits every second on server. If it is stored on nand... What's life expectancy? Do they use like 64GB nand on 8GB memory module to actually keep it from sudden death?

Memory, that's a lot of changed bits every second on server. If it is stored on nand...

I'm reading that it only writes to NAND when the power is interrupted. I'm still not sure how this would work. Are they saying we wouldn't need to reboot the server? It would just magically skip the 5-10 minute boot process and just jump right back to its previous running state? I can't see that happening....

I'm reading that it only writes to NAND when the power is interrupted. I'm still not sure how this would work. Are they saying we wouldn't need to reboot the server? It would just magically skip the 5-10 minute boot process and just jump right back to its previous running state? I can't see that happening....

Then They have to have capacitors or backup battery which will keep memory operational during writing data to nand. That would work. Data in CPU cache would still be lost, unless even CPU/MB has UPS backup. And as it reaches low power it would trigger 'suspend to ram'. With 'suspend to ram' CPU no longer works on data and as UPS runs out of power, memory will write its nand. Upon recovery of electricity delivery system has to initiate reverse 'suspend to ram'. How that's triggered when entire MB goes out of power, I do not know. But considering all complications in this chain and requirements to have UPS anyway to prevent errors from CPU cache loss... I think UPS + classical software initiated hibernation is practically same solution with less complications. Hell, Even my UPS can tell to PC to shutdown/hibernate when it is running out of juice.

I don't really see the practicality of this application... Most data centers, like the one I'm sitting in, have redundant generators. If the city grid loses power the building can still stay running without interruption for several days.

Memory, that's a lot of changed bits every second on server. If it is stored on nand... What's life expectancy? Do they use like 64GB nand on 8GB memory module to actually keep it from sudden death?

I was thinking the same thing... I can't imagine this lasting much longer than a year or two in the average server. Not to mention, there will hardly be any hardware out there that can actually take advantage of the non-volatile feature. As stated by others, it's better to just use backup power sources and redundancy. Even if it's just a brownout, a few seconds of downtime could cost a fortune for companies that could afford something like this (especially if it has a very limited lifespan, and requires frequent replacement). Is this even ECC or buffered? EDIT: Just an afterthought - seeing as most servers run Linux, it's possible to "freeze" applications. Meaning, you can effectively hibernate the program without closing it or shutting off the computer, where you can then restore the application later on. If retaining a memory status is really so important, to me, it'd make more sense to use a cron job to freeze the application every 10 minutes or so. Depending on what's being processed, 10 minutes of lost time is better than several hours. Still a stupid idea, regardless.

i think this great potential for PC/desktop/workshop users. think of some reduced latency,boot time, power consumption compared to a traditional "harddisk->mobo<-Ram" architecture. if this tech will mature... this will put some order in the void between volatile and non-volatile memory. or if possible unify both volatile and non volatile just like how processors do right now where the memory controller/northbridge is incorporated inside the proc unlike in the past where in it was put in the mobo.

i think this great potential for PC/desktop/workshop users. think of some reduced latency,boot time, power consumption compared to a traditional "harddisk->mobo<-Ram" architecture.

There is definitely more practicality for a workstation user than a server (because workstations tend to have irreplaceable content, but don't rely on a constant uptime) but remember, you need to take into account motherboards that will actually support something like this. Considering how new this technology is, it would be far outside the reach of consumers, even workstation users. For most PC and workstation users, they're better off just doing hibernation or standby. It's only mildly slower. Also, I figure a UPS is cheaper, and more convenient.

There is definitely more practicality for a workstation user than a server (because workstations tend to have irreplaceable content, but don't rely on a constant uptime) but remember, you need to take into account motherboards that will actually support something like this. Considering how new this technology is, it would be far outside the reach of consumers, even workstation users. For most PC and workstation users, they're better off just doing hibernation or standby. It's only mildly slower. Also, I figure a UPS is cheaper, and more convenient.

I believe so too. barebone system/hardware will be taking advantage of this tech. cant wait for something like this to happen. btw... are smartphones also/already using this sort of "unified memory tech"?

I believe so too. barebone system/hardware will be taking advantage of this tech. cant wait for something like this to happen. btw... are smartphones also/already using this sort of "unified memory tech"?

Depends on your definition of unified. But the simple answer is no. Smartphones use SoCs (System on Chips), where at the very least the CPU, GPU, VPU, northbridge, and RAM are all merged into one chip. In some cases, you might find things like NICs, USB controllers, audio controllers, etc are also in the same chip. To my knowledge, there are no ARM or MIPS devices where the NAND storage is part of the SoC. So physically, you could say that smartphone platforms are unified. But at the software level it doesn't work that way at all. Also keep in mind that for some phones (not sure about Windows or iOS), Android is a virtual machine. That means you could, in theory, do a snapshot of the OS, pull out the battery, put the battery back in, and resume your session as though you had something like this nvdimm memory. But snapshots are really just an external form of hibernation.

Then They have to have capacitors or backup battery

They do! "combined with a PowerGem capacitor from AgigA Tech"

I can't imagine this lasting much longer than a year or two in the average server. Not to mention, there will hardly be any hardware out there that can actually take advantage of the non-volatile feature.

I can' see this having a shorter lifespan than any other RAM - the NAND is only written to in the event of power failure. All servers(should) use UPS, so this wont happen very often. The hardware to support this is already available.

or if possible unify both volatile and non volatile

Intel is going to start closing that gap soon with XPoint. After reading more, it sounds like this is for RAID controllers or NAS.

I can' see this having a shorter lifespan than any other RAM - the NAND is only written to in the event of power failure. All servers(should) use UPS, so this wont happen very often. The hardware to support this is already available.

If the NAND is only written to in the event of a power failure, then what's the point of having this? Why not just have an SSD dedicated toward hibernation, VM snapshots, or application freezing?