Future computers

[coughtryb]

I heard an interesting idea at work the other week regarding future computers. In theory, couldn't flash memory be used in place of both active memory and a HDD? Being non-volitile in nature there would be no need for separate devices right? can't we folfill both purposes with 1 device?

Obviously, advances in size and changes to popular OSs would have to be made first. But think of the benefts...fewer moving parts yeilds less heat and less electric comsumption; cold booting a PC would be as fast as turning on a monitor since you don't have to load information across a system board. I'm interested to hear what you all think about this. Is this a direction all PCs may take?

 

[dorion]

ITs possible but Flash lags behind in terms of Size on Hard drives, and in speed for ram. Also flash still has a lower lifetime than hard drives and ram, as soon as flash can stand the massive amount of writes that ram takes you might have something worthwhile. Anyways flash isn't the only alternative. I'm looking foward to when the finally get M-ram on the market.

 

[pcy]

Hi,


This is the key point:

Originally posted by: dorion
... Flash lags behind in terms of Size on Hard Drives, and in terms of Speed for RAM.

We will always want a memory hierachy - a smaller quantity of faster (hence more expensive) memory and a larger quantity of slowe cheaper stuff.


It may be that at some point a totally new technology (and it might be Flash RAM) replaces Hard Drives, and it's equally possible that the same fundamental technolgy (though probably deployed differently) could also replace RAM.

However, the speed, price and functional differences in the requirement for working storage (currently RAM) and data storage (currently HD) are so large that is seems to me unlikely that one technology will ever perform both tasks.



Peter

 

[BrownTown]

flash is way slower than DRAM, plus its has a write limit which makes it useless for RAM which is constantly being rewritten. Its only advantage is that it is not volitile, but since that is not an advantage as far as main memmory goes it has no advantages in that arena. Using Flash for a hard drive is a different story, but with perpendicular recording you can get so much denser recording with a HD that i doubt anyone will want to completely replace them since it would mean less capacity.

 

[Jdog1718]

Price is a huge problem with this idea. The other of course is design. Just a few gigs of Flash Memory can cost hundreds of dollars. If cost and design were not a factor, wouldn't we have everything just integrated onto the CPU itself. No data bus to worry about, and its fast. Obviously this would require users to purchase an all in one CPU/Memory-Hardrive that would be outragously expensive and would prevent anyone from upgrading just one component. There's more problems with this idea than I care to discuss. The architecture should stay pretty much the same. Have CPU cache thats very small... but fast, main memory thats more spacious and slower, and then a long term storage that the slowest. We are, however, seeing the laptop harddrives that are supposed to contain flash memory in addition to platters that will improve battery life and speed up boot times. So it may be possible, but economics will play a bigger role than pure technology will.

 

[snes tor]

Yeah theres an article of this in PCWorld. They reviewed harddrives with 4gb of flash memory in them for cache.... windows start up , standby, and such. They got good reviews, but they cost way to much. and power consumption on a notebook only increased 19% ... about 25 minutes

 

[pm]

Another interesting question is whether a replacement for Flash - one that was faster and more robust and cheaper to manufacture - could work. Floating-gate flash technology is approaching the scaling limit a lot faster than CMOS. Something will replace it.

I know which one I'd bet on: Phase-change RAM. A much higher limit on the numbers of writes, prototypes which are already more dense than commercial Flash and which are being build in large-scale devices, and a read speed that's only about half of DRAM. Also the distinction between a "1" and a "0" is huge - leading to the likelihood of of a phase-change memory element (a single memory "bit") holding 3 or more states. Intel's Strataflash can hold 4 states... based on what I have read of "PRAM", it's not unlikely that it would be able to hold 4, or perhaps even 8 states, per memory cell. If this were possible (and I know that I'm piling "ifs" on more "ifs" but...), if PRAM could hold 8 states per cell, then the density would be impressive brining manufacturing costs much lower.

FRAM and MRAM have gotten more press than the rather poorly named "PRAM" but MRAM and FRAM prototypes are pretty small (4Mb/8Mb) and still look like... prototypes - whereas Samsung's new 512Mb PRAM device actually looks like something approaching productization. Samsung is saying that it will be available in 2008 and will compete primarily against NOR flash.... but that will just be the beginning (IMO).

 

[pm]

How about a future where the CPU and memory are on die together (see today's news item at CNet), and then there's some form of reasonably sized nonvolatile (NV)memory (like Flash, FRAM, PRAM) which holds the OS and memory workspace - so the entire OS, and then a backup of the CPU RAM. Then when you power off, you just move the CPU RAM contents to NV-RAM, and when you turn back on you reverse this, leaving you where you left off. And for reboots, you reboot the OS from NV-RAM entirely for a massive speed up in boot time.

It's an idea anyway.

 

[SickBeast]

Originally posted by: pm
How about a future where the CPU and memory are on die together (see today's news item at CNet), and then there's some form of reasonably sized nonvolatile (NV)memory (like Flash, FRAM, PRAM) which holds the OS and memory workspace - so the entire OS, and then a backup of the CPU RAM. Then when you power off, you just move the CPU RAM contents to NV-RAM, and when you turn back on you reverse this, leaving you where you left off. And for reboots, you reboot the OS from NV-RAM entirely for a massive speed up in boot time.

It's an idea anyway.

What about a CPU/GPU tandem on a dual-core chip (with a memory controller built into the CPU), similar to what AMD/ATI is currently (more than likely) building?

I would imagine that the graphics division at intel is doing some interesting research right now. Methinks DirectX 10 will be an interesting graphics battle.

Out of curiousity, how much RAM could fit within the space of two "core dies" using 0.65nm technology? I'm thinking an interesting thought might be the above scenario, with two other cores for memory/cache (one for graphics, the other for the CPU).

 

[Genx87]

I have heard this for 15 years, the problem I understand is

A. Flash doesnt have the write and read lifetimes of HDs. They lose the ability to save information way too fast and thus would have a lifetime of a few days or weeks, compared to years like a HD.
B. Flash while able to have high bandwidth has horrible latency when seeking compared to current disks.
C. Capacity, Flash is right now is just a few GBs compared to nearly a TB on a HD.

 

[gsellis]

Y'all might be interested in this Tom's review

Tom's Flash Drive Preview.

Me, I would like to eventually see a hybrid. A Flash drive in a RAID 1 config with platters. Faster read/write with a secondary, higher retention media as a backup store. Not cache, a second copy. High performance and better reliability for the win.

 

[pm]

Originally posted by: Genx87
B. Flash while able to have high bandwidth has horrible latency when seeking compared to current disks.

Maximum worst-case latency on a NAND non-sequential random read access is on the order of 20us (2.0E-5 seconds), for example on the currently shipping 8Gb (1M x 8bit) NAND Flash (model number K9K8G08U1A) is 25us (2.5E-5 seconds).

This compares to a hard disk average (not worst-case) seek latency on a currently shipping Seagate Barracuda 7200.8 ST3400832AS 400GB 7200 RPM hard disk of 11ms (1.1E-2 seconds).

Worst-case NAND flash access latency 440 times faster than a 7200RPM hard disk. And on the lower-density but faster read latency NOR flash, the speed difference is even greater.

C. Capacity, Flash is right now is just a few GBs compared to nearly a TB on a HD.

Best to compare like to like. $150 currently buys an 8GB compactflash card, and a 400GB hard disk. Your point is correct, but the discrepancy isn't as large as you stated. 8GB is more than a "few". And if we compare against a 750GB drive - nearly a TB, they are retailing in the $450 range - which would buy you about 24GB of compactflash. Like, I said, I'm not disagreeing your with point, but with the magnitude of the difference. The ratio is about 50:1 (hard disk capacity vs. flash capacity) for the same amount of money whereas a "few GB" vs. "nearly a TB" would be more like 350:1.

Out of curiousity, how much RAM could fit within the space of two "core dies" using 0.65nm technology?

For SRAM, it would be approximately 30MB (very approximate - depends on core size, stepper limit, etc, but it's in the ballpark). For DRAM, I'm not sure. I'll try to do some research and see if I can calculate it.

 

[BrownTown]

Good luck making 65nm DRAM btw.

 

[pm]

Originally posted by: BrownTown
Good luck making 65nm DRAM btw.

I don't understand your point... Is it that DRAM manufacturer's process technology nodes are out of sync with logic manufacturer's process nodes? I thought that 65nm was a converged node (ie. one of those rare nodes where both logic and DRAM processes converge). Or is it that no one is shipping 65nm DRAM presently? Well, this whole thing is a thought experiment anyway.

65nm e-DRAM (logic process technology compatible DRAM, known as embedded DRAM) from major manufacturers:

TSMC:
http://www.tsmc.com/download/english/a0...df#search=%2265nm%20embedded%20dram%22
NEC:
http://www.necel.com/process/en/edram.html
Sony & Toshiba:
http://www.us.design-reuse.com/news/news4426.html
IBM:
http://www.research.ibm.com/journal/rd/492/iyer.html
AMD seems to be thinking about using Z-RAM:
http://www.eetimes.com/news/latest/showArticle.jhtml?articleID=177101749

 

[Ynyr]

The question is....

Its only a matter of time before Flash drives are used for the OS and other key applications. Flash is dropping in price at an insane rate. A lot of R&D is taking place in this area as there is a lot of money to be made.

I wonder how long it will be until MS distributes windows only on DRM protected Flash Drive that you stick into your computer? My guess is 10-15 years.

 

[CSMR]

I remember when my OS was stored on ROM. Now those were the days!

 

[Evadman]

Originally posted by: CSMR
I remember when my OS was stored on ROM. Now those were the days!

You can do that now, there just isn;t much of a point. AT reviewed one from gigabyte here buy it from newegg for $130

 

[CSMR]

ROM != RAM

 

[gsellis]

Originally posted by: CSMR
I remember when my OS was stored on ROM. Now those were the days!

HA! We stored the OS on the server. (use to RIPL OS/2 2.X) Not as old as ROM stored OS, but still cool.

 

[Calin]

Originally posted by: Genx87
I have heard this for 15 years, the problem I understand is

A. Flash doesnt have the write and read lifetimes of HDs. They lose the ability to save information way too fast and thus would have a lifetime of a few days or weeks, compared to years like a HD.
B. Flash while able to have high bandwidth has horrible latency when seeking compared to current disks.
C. Capacity, Flash is right now is just a few GBs compared to nearly a TB on a HD.

A. There are some "load leveling" algorithms that help with that - the idea is to switch heavily written "flash sectors" into less heavily written "flash sectors" (if you have a hard drive, the page file - which is the most read area of the disk) resides in the same place the entire lifetime of the disk. A flash drive will see that this area was heavily written, and alocate it in a different area (which wasn't written at all/was much less written). This load leveling can increase the expected life time tremendously
B. This is completely the reverse: hard drive sequential read/write bandwidth (varies in different areas of the disk, is bigger at start/on the outside) is many times greater than usual flash drive sequential read speed (RAID-ing flash drives can help with that). However, a hard drive is able to responds to just hundreds of different read/write requests per second, greatly limiting its bandwidth in case of fragmented reads. Flash, on the other side, is hardly penalized for non-sequential operations.