Intel Optane DC Persistent Memory

[olzhabay]

Does anybody know something about newly coming Intel Optane DC Persistent Memory, except this.

Memory amount and expected price? Some insider information?

[8uhbbhu8]

Pretty sure Intel has released no info as it's not even ready for release yet. From what I know they are still working of perfecting it so there is no way a price can be finalized until the y have it running right.

[xentropa]

Optane uses a technology known as Phase Change Memory.   I believe the memory cell is Germanium Antimony Tellurium.  If you give it a quick burst of current, it will liquefy and solidify as a crystalless amorphous structure.  This is high resistance.  If it a lower but longer electrical current is applied, it will recrystallize into its low resistance state.  The cell is switches between 1 and 0 (low resistance/high resistance) by applying a small but long current pulse and a high but short current pulse respectively.  Without proof, but from what I have heard, the short current pulse is around tens of nanoseconds and the longer current pulse is between 100 to 200 ns.

 

Compared to DRAM, it's integration is easier.  The memory, cell is sandwiched between criss cross wires and it is not as thick as DRAM.  This makes it easier to have a multilevel 3D structure and can achieve higher densities as DRAM.  This structure is called the 3D Xpoint.  Unfortunately, each layer of 3D Xpoint requires photolithography, unlike 3DNAND.  Therefore it is not as high density as 3DNAND.  The cost per GB of Optane is less than DRAM but will always be more than 3DNAND.  It's speed and performance is also roughly in between those two as well.

 

Intel has claimed optane's write endurance is about 1 billion write erase cycles, however on the intel ARK site, the 375 GB optane is published with a lifetime of 50 PB writes, which is only about 100,000 write cycles.  From what I know, 1 billion (much less 100,000) write/erase endurance is insufficient to replace DRAM.  To replace DRAM, a memory cell with an endurance of 1 trillion or higher W/E cycles is necessary.

 

Overall in my opinion, optane does seem to fit nicely in between DRAM and NAND Flash in terms of performance cost.  But it doesn't really revolutionize anything, and I don't see it really disrupting the DRAM or Flash memory market, like how Flash seriously disrupted the HDD market.  It's just a nice to have thing.

 

As much as a computer enthusiast as I am, I did not buy it nor do I intend to.

[RobFRaschke]

Optane makes an incredible boost to hard drive performance on small, regularly used files. Random access speeds are incredible compared to current generation nand flash on current controllers and have something on the order of 7 times the write endurance of current nand flash.

 

There's not going to be another storage revolution like there was from spinning disks to flash memory. But incremental increases from new technology could all but bridge the gap in performance between persistent memories like optane and non-persistent memories like DRAM. I can't see it happening in the next 5 years, but maybe in 10. M.2 has already moved boot drives onto the motherboard and are rapidly becoming more reasonably priced. Specialty 288-pin RAM slots that can be used for buffer drives using optane memory have been in the Intel platform plans since Skylake-X planning so we're not far off from it on server/workstation platforms and it all trickles down eventually.

[SansVarnic]

-= Moved to CPUs, Motherboards, and Memory =-

[olzhabay]

1 hour ago, xentropa said:

Optane uses a technology known as Phase Change Memory.   I believe the memory cell is Germanium Antimony Tellurium.  If you give it a quick burst of current, it will liquefy and solidify as a crystalless amorphous structure.  This is high resistance.  If it a lower but longer electrical current is applied, it will recrystallize into its low resistance state.  The cell is switches between 1 and 0 (low resistance/high resistance) by applying a small but long current pulse and a high but short current pulse respectively.  Without proof, but from what I have heard, the short current pulse is around tens of nanoseconds and the longer current pulse is between 100 to 200 ns.

 

Compared to DRAM, it's integration is easier.  The memory, cell is sandwiched between criss cross wires and it is not as thick as DRAM.  This makes it easier to have a multilevel 3D structure and can achieve higher densities as DRAM.  This structure is called the 3D Xpoint.  Unfortunately, each layer of 3D Xpoint requires photolithography, unlike 3DNAND.  Therefore it is not as high density as 3DNAND.  The cost per GB of Optane is less than DRAM but will always be more than 3DNAND.  It's speed and performance is also roughly in between those two as well.

 

Intel has claimed optane's write endurance is about 1 billion write erase cycles, however on the intel ARK site, the 375 GB optane is published with a lifetime of 50 PB writes, which is only about 100,000 write cycles.  From what I know, 1 billion (much less 100,000) write/erase endurance is insufficient to replace DRAM.  To replace DRAM, a memory cell with an endurance of 1 trillion or higher W/E cycles is necessary.

 

Overall in my opinion, optane does seem to fit nicely in between DRAM and NAND Flash in terms of performance cost.  But it doesn't really revolutionize anything, and I don't see it really disrupting the DRAM or Flash memory market, like how Flash seriously disrupted the HDD market.  It's just a nice to have thing.

 

As much as a computer enthusiast as I am, I did not buy it nor do I intend to.

There are durability issues of PCM, that I agree. Will be interesting to see per byte cost of PCM, and compare it with DRAM. Of course, considering all the pros and cons of both. What is most intriguing, is how much will be efficient to use PCM in DBMS and DFS applications, or maybe coupling DRAM with PCM. 

 

[O9B0666]

3 hours ago, RobFRaschke said:

Optane makes an incredible boost to hard drive performance on small, regularly used files. Random access speeds are incredible compared to current generation nand flash on current controllers and have something on the order of 7 times the write endurance of current nand flash.

 

There's not going to be another storage revolution like there was from spinning disks to flash memory. But incremental increases from new technology could all but bridge the gap in performance between persistent memories like optane and non-persistent memories like DRAM. I can't see it happening in the next 5 years, but maybe in 10. M.2 has already moved boot drives onto the motherboard and are rapidly becoming more reasonably priced. Specialty 288-pin RAM slots that can be used for buffer drives using optane memory have been in the Intel platform plans since Skylake-X planning so we're not far off from it on server/workstation platforms and it all trickles down eventually.

pretty well said. In short, it basically feels like it turned a normal 7200rpm drive into a ssd, sort of.