Princeton: Replacing RAM with Flash Can Save Massive Power
Fusion-io and Princeton University recently announced a partnership in which the company would enable customers to use Flash memory as an extension of DRAM. The University now added some more information about the technology that adds to the product and developer offering provided by Fusion-io.
The development of the software that enables Flash to serve as DRAM replacement is promoted by the fact that Flash is considerably cheaper and that it uses up to 90 percent less power than its volatile counterpart, Princeton researchers said. The software they created, called SSDAlloc, turns Flash memory into an entity that can behave more like traditional and not like storage memory, which removes a bottleneck that is present when a program switches from DRAM to storage memory.
According to the researchers, SSDAlloc allows developers to bypass "this traditional system of searching for information in storage memory. […]Essentially, SSDAlloc moves the flash memory up in the internal hierarchy of computer data — instead of thinking of flash as a version of a storage drive, SSDAlloc tells the computer to consider it a larger, somewhat slower, version of RAM."
The benefit of this approach is that that the software does not require changes to the program. If you were using RAM and you want to use RAM, you can do that. If you want to use solid state you can use that," said Anirudh Badam, a graduate student who pas part of the developer team.

Nowadays the gains are very little: 90% on power usage of MEMORY (some watts, we are not talking about CPU or GPU). Wonder about the lifespan of the NAND cells in such scenario...
What this article forgets to remind people is that the flash side is only used on items that are not active in the ram so the performance difference does not become an issue.
I was just about to ask that, that strikes me as a little worrisome considering flash has a very limited lifespan.
What this does do is significantly increase the performance of the "disk interface" i.e., whatever version of SATA is currently being used to access flash storage. The result is that the "disk" subsystem's performance is boosted to nearly that of RAM.
because they could not make headlines with that :-))
Read performance per cell of NAND flash is fairly close to DRAM read performance. Write performance is exponentially lower than DRAM, but read performance (arguably more important) is close. Even then, the write perfomrance difference is nothing like the difference between NAND flash and floppy disks.
That might be an accurate way of viewing this. I'd think that a specialized flash memory would be preferable for this, such as some SLC that is specifically optimized for this, to alleviate the reliability concerns. It would be expensive per GB, but it could be cheaper than RAM per GB and if it's done right like the guys from the article have been talking about and working on, then maybe it could be a better solution than what we use now, at least in high-capacity servers such as powerful web servers and such.
That's probably true too, but not even remotely comparable and thus relevant to what is being looked into here.
http://www.newegg.com/Product/Product.aspx?Item=N82E16813131830
http://www.newegg.com/Product/Product.aspx?Item=N82E16813131854
http://www.newegg.com/Product/Product.aspx?Item=N82E16813131858