Sign in with
Sign up | Sign in
Your question

Merom to get Robson flash cache

Tags:
Last response: in CPUs
Share
February 8, 2006 10:41:34 PM

Yes, I know, it is one of those "according to sources" stories. (http://www.tgdaily.com/2006/02/09/intel_robson_technolo...)But we have heard the first credible details that Intel is making visible progress on bringing its Robson technology from presentation slides and research labs into products – and we felt it was well worth it to pass that news on to you.

Robson, as it appears, is still one year away, but could provide a nice jump in battery time and performance once it arrives. It's a no-brainer that most of us would be interested to see a Flash-supported motherboard and like to have more battery time in a notebook - but what do you expect from such a technology? Is it just for notebooks or does it make sense in desktops as well?

Looking forward to your thoughts!
February 8, 2006 11:00:51 PM

Quote:
If Flash can store all those small files that are permanently accesses on a computer, a hard drive would have to be activated less often - which, according to industry observers, could result in an additional 15-30 minutes of battery time in an average notebook.


I hate to say it, but once again Intel is leading the way in the mobile arena. If true, 15-30 minutes is pretty damn significant. And I tend to believe it. Take the iPod for example. It uses the hard drive to retrieve the MP3 file(s) and put them in the on board 32MB RAM. It's a cache. This is the reason why the battery life in iPods is so damn good (considering they're hard drive based). If the same thing can be done for laptops, i don't think it's unreasonable to believe that it would significantly increase battery life. (I personally believe laptop LCD manufacturers are the ones who should be leading the way in power consumption, but once again the job is being left up to Intel and AMD).

Quote:
It's a no-brainer that most of us would be interested to see a Flash-supported motherboard and like to have more battery time in a notebook - but what do you expect from such a technology? Is it just for notebooks or does it make sense in desktops as well?


If used in conjunction with this "SuperFetch" feature in Windows Vista, I'd say it's definitely worth extending to the desktop arena. Though I seriously doubt anything like this would show up in desktop processors. It seems to me it would make more sense to use a flash memory drive or similar component rather than soldering a bunch of flash chips on a motherboard. It'd probably be cheaper too. Putting 256MB or 512MB of fast flash memory on a hard drive would be pretty cool (I have no clue how well it would work, but there has to be a way...)

But hey, at the rate L2 caches are exploding- who knows? We might just have 16MB L2 caches by next year.

-mpjesse
February 9, 2006 1:24:49 AM

Flash the OS chip! This could be the end of hard drives someday.
Related resources
February 9, 2006 2:38:26 AM

I can definetly see it moving into the Desktop arena. Given current die size and the ability to produce more chips per wafer with already larger L2 caches (see Presler). I think this also sets them up for future expansion for these types of features.

Going OFF chip and ON bus for a flash HD will slow things down given the limitations of the EIDE, SATA I, SATA II interfaces (although it will still be quicker than accessing current mechanical HD tech). I would think that they would either stay on chip or make a daughter chip/memory controller that would handle the new flash. Maybe with a new interface with a faster/similar t bus speed (that would be nice).

Ram drives are nothing new. Using Non-Volitile memory for that ram drive is not new either. Using them as a boot element in a laptop/desktop is.
February 9, 2006 2:53:00 AM

WOW somebody really misunderstood what they were reading when they wrote this article.

The flash memory they are talking about is EFI otherwise known as Extensible Firmware Interface, intel has been pushing this for quite some time. It is a replacement for the old bios - basic input output system thats been in our machines for years. EFI is currently in the new intel based mac's and only supported by the 64bit version of windows xp for use with itaniums. Windows Vista will support it however a system with EFI wont be able to run legacy operating systems.
February 9, 2006 3:24:04 AM

PapaSmurf,

you may be right that Robson and EFI could be related. EFI has been pushed for quite a while and isn't new anymore, as you stated anyway. However, Robson is indicated to be significantly more than the EFI we know today for the Intel Macs.

As far as we know, Robson will surface togfether with ICH8 and will have a primary function to function as cache. In the context of EFI, NAND Flash however, will have the purpose to hold flashable applications (that also can be stored on the hard drive apparently) that control the features of the motherboard.

Interestingly, the NAND Flash approach will be able to not just cover the functions of what BIOS does today, but impact the boot time of the complete OS. The software showed in demonstrations last year that it also is capable to serve as cache when the OS is up - to accelerate the access to frequently used applications.

There is a clear difference in how Intel describes EFI and how it lays out the functionality of Robson - as far as we know.
February 9, 2006 12:30:17 PM

I am one of a group that is closely watching the development of phase change memory. (OUM =Ovonics Unified Memory= PRAM) Unless Intel is going to let Samsung beat them to market, Intel is going to have to introduce OUM into their products this year. It is possible that Viiv and AMT could be made to work using flash, both of these and the descriptions of where they are going with their processors fits with what OUM would enable.

Here is a good presentation on AMT:
http://www.intel.com/it/digital-enterprise/active-manag...
Figure 1 on page 5 is a good graphic.
_______________________________
December 28, 2005 – Ovonyx, Inc. (‘Ovonyx’) and Samsung Electronics Co, Ltd. (“Samsung”) today announced that they have entered into a long-term license agreement under Ovonyx’ intellectual property relating to Ovonic Universal Memory (OUM) thin-film semiconductor memory technology.

Oct. 19, 2005 –Ovonyx, Inc. (‘Ovonyx’) a semiconductor memory technology developer, announced today that Intel Capital has made an additional investment in the company. Financial terms were not disclosed.

Ovonyx memory technology uses a reversible phase-change memory process that has been previously commercialized worldwide in rewritable CD and DVD optical memory disks. The Ovonyx array-addressed semiconductor memory technology can be used in many applications benefiting from non-volatile low power memory such as mobile computing, communications, and information appliances, as well as embedded applications in many product areas such as microcontrollers and reconfigurable MOS logic.

In February 2000, Ovonyx announced that Intel Capital had made an investment in the company and that Ovonyx had licensed its phase-change memory technology to Intel. The two companies have been working together since then to develop and demonstrate the feasibility of high-density, non-volatile memory based on the technology.
_____________________________________
The primary patent base of OUM is held by Ovonyx
http://www.ovonyx.com/
_____________________________________
I have been trying to fing any techies that are 1-following Intel developments AND 2-know anything about PRAM. I haven't had much luck. (I am not a chip tech, I just like tech developments)

rbtbob
February 9, 2006 12:31:07 PM

Hello all,

What if, When all this comes together and everything is released at the same time and the processors and boards have all the technology implemented, that Vista loads All boot and prefetch files into the NAND built into the Motherboard and will actually adjust the "bios"(EFI) to what it needs to perform at it's best? That is probably the ultimate goal of EFI anyway, to be more flexible with operating systems and users.
February 9, 2006 1:06:21 PM

Wanting to extend flash memory to the desktop remindes me of those Gigabyte i-RAM cards. Most of the reviews I read were pretty positive, it being SATA1 instead of SATA2 and cost seemed to be the only gripes I read about it.

When mainstream memory pushes past the 500mhz mark, would it be possible to have a hard drive that communicated strictly through the PCI Express bus and circumvent "slow" HD interfaces? You'd probably need some kind of boot loader to access it...maybe put a small ROM on there that gets connected to the SATA just to get things running.
February 9, 2006 1:19:51 PM

Vista already uses a bootloader, no more boot.ini or sys files. It will work with the NAND and load on that. very fast indeed.
February 9, 2006 3:23:29 PM

Maybe I'm missing something, but what does Intel's trumped up version of hardware monitoring (AMT) have to do with this article? Seriously, I can't seem to make a connection here...

As for AMT- it's about time. There have been 3rd party solutions for hardware monitoring for years now. It's nice to see Intel finally integrating it. Only took '15 years or so.

-mpjesse
February 9, 2006 3:35:17 PM

It's quite the nice idea, but will it work in real life, not just in testbeds?

The problem with putting flash, or even DDR chips, on a hard drive or removable device is that CPUs still lack enough punch to deliver truly faster boot or load times, and interface bandwidth still suffers when compared to normal DDR bandwidth. In fact, PC66 SDRAM's 533 Mbytes/sec bandwidth from ten years ago still outperforms today's fastest SATA or SCSI connection.

In THG article "Can Gigabyte's i-RAM Replace Existing Hard Drives" from 7 Sept 05, benchmarks clearly showed the XP boot time was shaved from 35 seconds to 31 seconds. This is not what I would call a fast boot time. Loading Photoshop took six seconds down from nine. Marginal improvement, there. Saving a 309 MByte TIFF file also took six seconds, down from nine on the Raptor. Not a big improvement there, either.

The processor is responsible for many tasks during a Windows boot-up cycle. One can see that the hard drive isn't being hit 100% of the time during boot, and during a Photoshop load. The CPU has to think. We will never be able to have instant-on as long as Windows has a registry and has to load hundreds of libraries and fonts right at the beginning.

The solution: take a page out of the linux book, the Apple book, and the DOS book, and put them all together in the new Windows book.

Software is required to be zippy; after all, we don't want to be waiting more than two seconds for Word or Excel to open, or displaying a web page with non-standardized fonts. Windows should load only that code, those fonts and those libraries which are essential to presenting the user with a basic display at boot time (i.e., the desktop). Once the user is free to use his computer, then load the things that belong in the tray and in memory (i.e. virus scanners, malware scanners, extra fonts, extra libraries), but do it in low-priority so the user can still have zippy loads of his programs.

Along with this solution comes a transition from solid-state memory on interfaces that can't handle the high bandwidth (SATA, etc) to on the motherboard behind the memory controller in a separate area backed up by battery. In essence, build Gigabyte's i-RAM onto the motherboard. PC3200 DDR can provide a theoretical 6.4 GBytes/sec bandwidth, and in practice, half that. Who really needs that much transfer rate from storage that is so small as to be wholly transferred in just over one second? Even so, being there, the entire Windows data load can take place in one second.

The other solution is for Windows to save states similar to hibernation modes, where the state can be loaded from a flash, NAND or DDR device on demand. Where hibernation simply stores the entire contents of RAM to hard disk, then re-reads when re-started, a state would store the entire initial boot-up in similar fashion. The PC would be clean-booted (not hibernated) but start up in the blink of an eye because all the CPU processing has already been incorporated into the state, and all the required libraries, etc, are already there. All Windows has to do then is load the state it needs, and in two or three seconds, the user has full control with everything he needs loaded. The advantage of states would be the ability to create multiple states for multiple purposes, i.e. gaming, office, video, etc.

Clearly, the main bottleneck in the boot times of Windows is Windows itself. Vista doesn't seem to be all that zippy either, but it is still in early Beta, and I will reserve judgment on Vista until it is gold.
February 9, 2006 4:30:55 PM

If you read the Intel presentation you will see numerous references to non-volatile memory. AMT is also being planned to work whether or not the platform device is running. In other words, Intel is planning for all platforms attached to a network to have non-volatile memory that will enable the AMT. They use the term "platform" because they intend this to work with any network attached device, including those with a wireless network connection.

How is Intel going to make this work unless the inclusion of non-volatile memory is going to be standard across their entire processor line. PRAM can be embedded in the processor and requires only minimal added steps.

My theory is that sometime this year Intel will be embedding PRAM into their processors to enable AMT, Viiv "instant on", and a host of other functions that were not possible before. There is some possibility that they have already done so.

In addition to PRAM, phase change cells can be added to allow laser repair of chips with a manufacturing defect or to make all Intel processors PLD's. There is a recent patent that points to this use.

Is that enough to make it pertinent?
February 9, 2006 4:38:13 PM

OK, now I get the connection. Thanks,

-mpjesse
February 9, 2006 5:22:55 PM

Here are my thoughts.

1) Don't flash systems have a limit to the number of flashes before they start getting errors? So that's got to restrict what kinds of data you store there.

2) Give anyone the ability to flash something and they'll manage to screw it up somehow. This includes simple read/write errors produced by hardware, software, and strange cosmic events. So whatever gets implemented, it definately needs a way to wipe and start over from scratch when the inevitable happens.

But really, other than that, meh. I don't care one way or the other about it for laptops or desktops. It just doesn't really impress me.

I'd rather see a way to mass produce some low-cost very-high-density but slow (by modern PC standards) RAM that could be used to max out the IDE/SATA bandwidth in, say, a 100GB device. Give that device a battery that'll let it run long enough to ensure a backup from RAM to an internal 2.5" (or smaller) hard drive. Make the device interface through a logic unit that pretends it's just any old hard drive, and that automatically transfers between RAM and hard drive behind the scene whenever RAM changes or gets wiped by a power loss. Have a connection to a +5V SB line from the power supply so that the RAM only loses data (and thus has to reload from the HD) when power is actually lost to the PC. Put it in a standard 3.5" hard drive's packaging, and stun the world. :lol:  :lol:  :lol: 
February 11, 2006 9:14:10 PM

Quote:
Here are my thoughts.

1) Don't flash systems have a limit to the number of flashes before they start getting errors? So that's got to restrict what kinds of data you store there.

2) Give anyone the ability to flash something and they'll manage to screw it up somehow. This includes simple read/write errors produced by hardware, software, and strange cosmic events. So whatever gets implemented, it definately needs a way to wipe and start over from scratch when the inevitable happens.

But really, other than that, meh. I don't care one way or the other about it for laptops or desktops. It just doesn't really impress me.

I'd rather see a way to mass produce some low-cost very-high-density but slow (by modern PC standards) RAM that could be used to max out the IDE/SATA bandwidth in, say, a 100GB device. Give that device a battery that'll let it run long enough to ensure a backup from RAM to an internal 2.5" (or smaller) hard drive. Make the device interface through a logic unit that pretends it's just any old hard drive, and that automatically transfers between RAM and hard drive behind the scene whenever RAM changes or gets wiped by a power loss. Have a connection to a +5V SB line from the power supply so that the RAM only loses data (and thus has to reload from the HD) when power is actually lost to the PC. Put it in a standard 3.5" hard drive's packaging, and stun the world. :lol:  :lol:  :lol: 



There IS only a certain number of times one can write to flash before it wears out, I've experianced this with cf cards, they go belly up after a while. Personally I like my hardware to be able to last for years, I still make use of my old pentium pros I'm not likely to be an early adopter of this if indeed it is what they mean, I believe it only to be efi. Why would intel need a new processor to implement this technology? they dont. they may need a new operating system however.

There was an article not too long ago on thg about a memory manufacturer that is going to produce memory sticks using carbon nano tubes. These would last longer, be faster, take less power and generally beat the hell out of any flash based solution intel could come up with to enhance boot speeds. I'll wait for these :) 
!