News UltraRAM

[igor_kavinski]

UltraRAM breakthrough could finally fuse RAM and storage into a single package | TechSpot

I personally hate news like that. The chances of something like that actually translating to a real selling product is 1 in a million.

 

[IntelUser2000]

Western digital made a presentation that shows the relation between performance(in terms of latency, because to increase bandwidth you can just put more channels in parallel), and persistence, the ability for the data to stay relevant when the power is cut.

Roughly speaking, the faster the storage medium the more voltatile it is. Platter Drive--NAND SSD--Optane--RAM. The performance improves dramatically but the volatility increases.

NAND SSDs are JEDEC rated to have 10 year data retention after the cycles have been exhausted. Optane on the other hand, are rated by Intel to have 3 month retention. So after 3 month, they aren't guaranteeing the integrity of your data. Perhaps Intel is conservative and real world is longer at like a year or so, but that's a drastic decrease. Hard drives are tested over decades to show data rentention times greater than 10 years. The super slow tape drive is one of the most reliable drive in terms of data retention.

Technologies exist that have very good persistence and super low latencies but are prohibitely expensive. Notice that Optane is noted as one of the new technologies having the greatest potential for being a part of the memory hierarchy.*

So the triangle of trade off seems to be Persistence, Volatility, and Cost.

*This doesn't guarantee the chance of success for Optane. This depends on Intel's execution. The full advantages of lower cost over DRAM isn't realized because the volume isn't there.

Before people go on about Intel selling at a loss to get the volume there, think about this: The volume needed is upwards of 10x. People complained the 900/905P drives being expensive right? Well, those drives were selling at a breakeven point for Intel.

As the contra revenue program indicated, bleeding money causes the said company to quickly abandon it. The only way to go is organically. Clockwork yearly release cadence, persistence, working for the greater good(supporting non-Intel platforms) to go on without abandoning it too soon(Intel is noted for this). Then in 3,4,5 years the volume might start to be there.

 

[00Logic]

Western digital made a presentation that shows the relation between performance(in terms of latency, because to increase bandwidth you can just put more channels in parallel), and persistence, the ability for the data to stay relevant when the power is cut.

Roughly speaking, the faster the storage medium the more voltatile it is. Platter Drive--NAND SSD--Optane--RAM. The performance improves dramatically but the volatility increases.

NAND SSDs are JEDEC rated to have 10 year data retention after the cycles have been exhausted. Optane on the other hand, are rated by Intel to have 3 month retention. So after 3 month, they aren't guaranteeing the integrity of your data. Perhaps Intel is conservative and real world is longer at like a year or so, but that's a drastic decrease. Hard drives are tested over decades to show data rentention times greater than 10 years. The super slow tape drive is one of the most reliable drive in terms of data retention.

Technologies exist that have very good persistence and super low latencies but are prohibitely expensive. Notice that Optane is noted as one of the new technologies having the greatest potential for being a part of the memory hierarchy.*

So the triangle of trade off seems to be Persistence, Volatility, and Cost.

*This doesn't guarantee the chance of success for Optane. This depends on Intel's execution. The full advantages of lower cost over DRAM isn't realized because the volume isn't there.

Before people go on about Intel selling at a loss to get the volume there, think about this: The volume needed is upwards of 10x. People complained the 900/905P drives being expensive right? Well, those drives were selling at a breakeven point for Intel.

As the contra revenue program indicated, bleeding money causes the said company to quickly abandon it. The only way to go is organically. Clockwork yearly release cadence, persistence, working for the greater good(supporting non-Intel platforms) to go on without abandoning it too soon(Intel is noted for this). Then in 3,4,5 years the volume might start to be there.

In most cases: "If it sounds too good to be true, it probably is."
But statistically; once very 30 years a real breakthrough comes along and everyone living by that saying misses it...

Remember transistors vs ...vacuum tubes was it?

ie:
This one looks very interesting and worth reading and keeping an eye on:
High Density.
Low power.
Fast. (as fast as DRAM in testing. Probably faster in production)
Non volatile (retention 1000 yrs, 10 million program/erase cycles
Proven in/on silicon! And easy to make.

ULTRARAM: A Low‐Energy, High‐Endurance, Compound‐Semiconductor Memory on Silicon - Hodgson - - Advanced Electronic Materials - Wiley Online Library

 

[igor_kavinski]

A big issue is inertial resistance of the established players in the storage industry. They would rather keep churning out the same old thing and keep making the relatively "easy" money. We need Elon Musk to get excited about this technology. It sucks that he grew out of PCs and gaming a long time ago.

 

[IntelUser2000]

The technologies that people see as revolutionary, are really just enablers to continue progress. And the enablers only get productized when the conventional technology is reaching limits of scaling.

Look at how long EUV took to arrive. And now we're going to need things like double patterning for EUV!

Elon's foray into electric cars, Apple's debut of the iPhone, are significant but the pieces of the puzzle all existed. Yes initially they were few years ahead of the competition, but nothing that couldn't be done.

Remember Apple actually tried the tablet 30 years ago and that failed miserably. The vision simply surpassed available tools and materials.