News - ST-MRAM hits the market, damn expensive but damn fast

[HEXUS]

How does 400,000 IOPS per chip sound to you?

Read more.

[brasco]

typically offers 800 to 3,000 IOPS per module

That surprised me, I would have thought RAM would need to be really high even at current day speeds. Learn something new everytime I come here

[chates]

the 800 to 3000 IOPS relates to flash memory rather than RAM I believe (if I'm reading it right) but it is a massive jump!!

[watercooled]

Micron are also shipping an alternative to NAND; phase change memory: http://www.micron.com/products/phase-change-memory

[MaddAussie]

I can see this blurring the border between RAM and a HDD.. Interesting times

[Biscuit]

I can see this blurring the border between RAM and a HDD.. Interesting times

Yeah, once HDD catch up to RAM in access speed, is there any real need for RAM anymore?

[watercooled]

Interface and latency immediately spring to mind, you can't have system RAM on SATA or something, and latency through any current desktop storage interface is still miles higher than to system memory.

But in theory, with significant architectural changes, and if NVRAM matches DRAM in performance, durability and latency, and matches existing storage technologies in price, then it's possible to have just one memory system. I'm not holding my breath though.

[Tact.X]

I reckon in the future we'll just use flash memory (or an alternative like this) as a storage device and partition a certain amount to act as ram...

[Kumagoro]

If you have ram with nand storage then you woudn't really need an ssd.

Has anyone tried to calculate what process size and wafer size would make nand as cost effective as hdd are now or would be against projected hdd advancements.

[watercooled]

NAND is not a suitable replacement for DRAM, not least because of its very high latency vs DRAM (orders of magnitude) and limited write durability, which will only get worse with smaller nodes. Also remember HDDs continue to offer better value so they're not a static target; NAND might reach the price/GB HDDs are at now a few years down the line, but HDDs will also be significantly cheaper by then.

IMO NAND is an important stopgap until we get technologies like this article mentions, but as several large companies are already saying, it has a limited future. It's simply not a suitable replacement for HDDs in some scenarios.

[Attila the Bun]

...but as several large companies are already saying, it has a limited future. It's simply not a suitable replacement for HDDs in some scenarios.

I don't suppose these "several large companies" include Western Digital by any chance

[watercooled]

Not sure, but they include MS IIRC. There's not really any reason to doubt it though if you look at the details TBH; limited write cycles which will only continue to get worse as process shrinks to improve price/capacity, etc...

[Delarth]

This would be wonderful for a lot of people. Power goes out? No problem since nothing in the RAM would have been lost you could be back to work in a couple of seconds.

[watercooled]

You'd still have to consider CPU states and cache, which are volatile ATM, but it should be possible to quickly flush them to RAM on power fail with a capacitor, like some SSD's do now.