RAM versus SSD on MacBookAir

[MacInWin]

But you don't know if it would be expensive, or super expensive, or super super expensive.

Actually, we do know. If SSDs could be made fast, cheap and persistent, someone would build one and sell it as a world-beater. Given no one has done that, it's safe to conclude that persistent memory will always be slower than volatile memory used in RAM. (And the physics supports that, too, as the fast, volatile ram is so ephemeral. Writing a permanent change to a storage device is going to be slower just because of the processes required.)

Now, can we rule out some miracle bolt from the blue discovery suddenly changing the entire scenario? No, and you never can do that. But for now the computational physics is getting very close to where quantum mechanics starts to be a really challenging problem and there has not yet been found a way to have fast, cheap, persistent memory. Maybe if we find a room-temperature superconductor then we can all have miracle machines, but until that day...

R&D has to have some basis in reality, unless you are the Government or a University that can throw money at a study of the whichness of what for no real reason. Companies will do R&D in some interesting technologies, but with the goal of improving what they do. Witness IBM's quantum computing project: https://www.research.ibm.com/ibm-q/

If technology improves over the next 50 years as it has in the previous 50, quantum computing will be ubiquitous. Or maybe there will be one massive quantum computer and all of our devices will just use it as the central server for the real work and our local devices (watch, phone, pad, desktop) will just be input-output devices. Who knows?

 

[Raz0rEdge]

Sure, an R&D program to make really zippy SSDs might cost a lot. But you don't know if it would be expensive, or super expensive, or super super expensive. You also don't know if once the R&D program were complete, if the product produced would be unaffordable to consumers. Might be super super cheap! Given that we don't know anything about future costing of dream-products, I choose to live in the current day, and I will repeat. With current day technologies, SSDs are too slow and too prone to degradation. That is why, in the current day, we don't use them instead of RAM. The smartness of product providers is that they don't want to offer a product that is slow and prone to raid degradation.

NOT THE SAME TYPE OF MEDIA. You cannot use the SSD flash media as RAM, not now, not ever.

 

[pigoo3]

Given no one has done that, it's safe to conclude that persistent memory will always be slower than volatile memory used in RAM. (And the physics supports that, too, as the fast, volatile ram is so ephemeral.

This was one of my points earlier Jake. That as SSD hardware gets faster...very likely RAM speed will improve as well. So there will always be a "speed gap" between SSD's and RAM. Even if it were theoretically possible for an SSD to be fast enough (and durable enough) to be used as dual purpose (storage & RAM)...consumers would still always demand the greater/faster performance of even faster RAM.

More than likely some yet unknown (or on the drawing board) leap in hardware technology may be what gets us to a product that can act as both non-volatile storage & as RAM.

- Nick

 

[pm-r]

NOT THE SAME TYPE OF MEDIA. You cannot use the SSD flash media as RAM, not now, not ever.

Even if you could, or possibly gets suggested again that you could, I would suggest that any such user would be pretty P'd off with its short write/read life, and God only knows how they would deal with garbage collection and leveling stuff etc.





- Patrick
======

 

[pigoo3]

...God only knows how they would deal with garbage collection and leveling stuff etc.

I say ship it all to Brentwood Bay, BC!

- Nick

 

[Doug Lassiter]

NOT THE SAME TYPE OF MEDIA. You cannot use the SSD flash media as RAM, not now, not ever.

I don't think anyone is talking about using SSD architecture in place of RAM. What I'm talking about is using NAND flash non-volatile memory architecture (which is the basis for SSDs) as RAM. It is certainly true that you can do the reverse. That is, if volatility isn't an issue, you can build a SSD out of DDR RAM. You could even power it with a battery to have some level of non-volatility.

Actually, we do know. If SSDs could be made fast, cheap and persistent, someone would build one and sell it as a world-beater. Given no one has done that, it's safe to conclude that persistent memory will always be slower than volatile memory used in RAM.

If indeed the physics is such that non-volatile memory will always be slower, then I accept that. Otherwise, it might just be that no one has made the R&D investment in order to try. Not entirely clear why it would be a "world beater". If I had one SSD memory module in my computer instead of an SSD plus a RAM chip, it isn't going to shake things up much. Not at all clear that major R&D investment would be justified.

 

[MacInWin]

Not entirely clear why it would be a "world beater".

It would be a world beater because if you could get an SSD with the speed of RAM, your entire computing experience would be transformed by the speed increase.

Trust me, storage companies are doing R&D all the time on how to build faster, cheaper storage. Companies like IBM, Intel, HP, Samsung, even Apple, spend a lot of money trying to get things to move faster. One of the major throttling points in computing is always read and write operations to long term storage. That's why hybrid drives came out, and why the amount of cache on the hard drive was critical for performance. Anything to speed up that slow operation will transform the whole experience of the user. And the first company to find the way to make an SSD with RAM speed stands to make a ton of money, not just from folks like us, but from the industrial users. Think about a banking operation that could retrieve transactions from storage at the speed of RAM. Or a stock trading company who can make decisions quicker because the data in storage is retrieved at RAM speeds, or a currency trader watching the value of currency change millisecond by millisecond. Those folks would pay a lot for faster storage (and already do). In fact, some of those folks are paying for mid-range and mainframe computers with massive amounts of RAM just to try to hold as much data in RAM as they can and avoid the slow r/w cycle to storage. There are even folks optimizing routines dedicated to "guessing" what you will need from storage next and pre-staging it in RAM so that this pre-emptive read can make the data available faster. And other folks are working on storage routines that take from RAM and stage the written material in a RAM-like cache before it goes to storage, just to free up the CPU/RAM to get on with other things. The challenge in all that dancing is that if power cuts out while that written material is in the cache, but not written to storage, it's gone.

Bottom line: Companies are spending massive amounts of money on increasing storage speeds and capacity.

BTW, the IBM Watson computer that won on Jeopardy was built from 2880 CPUs operating in parallel and had 15 terabytes of RAM. That way none the processors had to go to storage as the entire database was in RAM. Data was loaded to it via scanners, at 2 million pages every three seconds. Indexing the database took months of processing so that Watson could correlate the data to get to an answer. I had a friend who worked on the project in the voice interpretation section so Watson could understand Alex Trebeck as he read the answers.

Frankly, the only breakthrough I can think of to make RAM and storage the same would be superconducting. In a superconductor, the current never fades, so once the bit is set, it stay set. That let's an SSD not have to have long term writing components embedded on the chip. Just send a bunch of electrons and they will stay there. But right now, to get that you need to be well under 1 degree Kelvin. If they find a room temperature superconducting material, that would be as revolutionary as the first transistor was. (And yes, a ton of money is being spent on superconductivity research, too.)