UltraRAM inches closer to mass production: non-volatile memory with DRAM speeds on the horizon

[Redaktion]

Developed by researchers from the Lancaster University in UK, UltraRAM is almost ready for mass production and could soon offer a unified solution that acts as RAM and storage. The technology relies on compound semiconductors that are usually used for photonic arrays like LEDs and lasers.

https://www.notebookcheck.net/UltraRAM-inches-closer-to-mass-production-non-volatile-memory-with-DRAM-speeds-on-the-horizon.591657.0.html

[Anonymousgg]

More expensive than DDR would mean dead on arrival.

It doesn't have enough endurance to act as a true universal memory.

[Bogdan Solca]

More expensive than DDR would mean dead on arrival.

So DDR5 is dead on arrival too if it cost 2X more than DDR4?

[vertigo]

Given the expectation that this will be pricey, confirmed by the suggestion it will be more than DDR, it will be extremely cost-prohibitive as a storage replacement, costing far more than SSDs of similar capacity. And aside from possibly servers, SSDs provide more than enough speed as it is. So I don't see this providing any benefit to storage in all but the most extreme, obscure cases. Which means it's only really good for the working memory (RAM) functionality, but if it's only similar in speeds but costs more, then it's not going to do much there, either. It would have to be either cheaper or faster, and it doesn't sound like either will be the case.

The only benefit I see to it, which granted is not insignificant, is that it would allow a device to be powered down completely, like hibernation, without having to first write RAM to storage, so it would be faster to put a device to sleep and wake it back up (though with the speed of SSDs hibernation only takes a couple seconds anyways), and it wouldn't use any power when "sleeping" because it wouldn't have to keep the URAM powered. And it wouldn't need a hiberfil.sys, which would free up a little bit of SSD space. All of this would certainly be beneficial, but only minimally.

And then there's security concerns. With a completely new technology, it's quite possible that after years of use, and many devices using it, they'll discover it has flaws that result in security issues, similar to the row hammering flaw with DRAM.

Another potential use, and probably the best one, would be for HDD cache, as it would provide massive speed improvements and, unlike SSD or DRAM caches, would protect against data loss in the event of power failure when writing.

[_MT_]

So DDR5 is dead on arrival too if it cost 2X more than DDR4?

The market is willing to pay for (usable) performance. But paying more for the same performance? That's a tough sell. Optane is a lot cheaper than DDR4. That's the selling pitch. You can get more space for your datasets for the same money, actually improving performance compared to having less memory. At the cost of raw performance, but traditional SDRAM is still there and acts as a cache, minimizing the practical performance hit compared to an all-SDRAM solution. That's something you need to realize. Solutions like this are aimed at machines with huge memory pools. Even DDR4 is very expensive at those capacities. And that's the very reason people keep looking at flash, be that as storage or flash-as-RAM solutions. It's such an important consideration that people are willing to sacrifice persistence. Offering something even more expensive just doesn't make that much sense, unless it has the performance to match (but that requires processors to go with it; you're not going to get paid for theoretical performance). Persistence in itself is not worth that much money. If it costs more than the sum of SDRAM and flash storage, it's just too expensive. And heterogenous approach allows you to save money by buying less RAM. You typically don't need a 1:1 ratio. If you're going for a converged, homogenous solution (RAM and storage in one), you need at least as much of it as you'd need storage in a conventional system, which is why it has to be pretty cheap. Or it can't replace storage.

[_MT_]

Another area of interest is "superfast" storage. Which is another angle that Optane works. Such a solution can be more expensive than standard flash storage (as I wrote, the market is willing to pay for performance), but nowhere near as expensive as DDR4. Unless it can match DDR4 for performance. The specifics are application dependent. Some applications really benefit from capacity, other applications can be very latency sensitive. However, it will never displace standard storage if it's significantly more expensive. There will just be some applications where the extra performance is worth it. Just to clarify the ending of my previous posting. It doesn't have to be cheap as chips to succeed. But again, more expensive than DDR4 really is a tough sell. If you succeed in creating storage that's as fast as DDR4 for the same money per GB, then your market is going to be something like in-memory database servers. But they're not going to pay double the price just to get persistence when flash is cheap as chips by comparison. I guess unless the database is so write heavy that flash doesn't cut it. Perhaps it might work out then. I have never seen such a system.

[_MT_]

Another potential use, and probably the best one, would be for HDD cache, as it would provide massive speed improvements and, unlike SSD or DRAM caches, would protect against data loss in the event of power failure when writing.

An interesting application, until you consider that it competes with capacitors. All you need is to keep the power up long enough to dump cache and park heads (huge caches don't help here). Getting interrupted in the middle of an operation is always risky even with persistent memory as registers and cache within a processor are still volatile. Having data and knowing what to do with it are two different things. Being able to properly clean up before a shutdown is handy.