backtop


Print E-mail del.icio.us 25 comment(s) - last by eye smite.. on Feb 26 at 11:50 PM


MetaRAM-based 8GB Dual-Rank DDR2 RDIMMs  (Source: SMART)
MetaRAM technology to cut server and workstation memory costs by up to 90 percent

Fabless semiconductor company MetaRAM today launched a new memory technology, called DDR2 MetaSDRAM that promises to double or quadruple memory capacity while maintaining compatibility with existing DIMMs.

MetaRAM achieves this feat by using 3D chip stacking methods to fit more capacity into a single memory chip. Such a method alone would have caused issues with compatibility, but MetaRAM developed a custom chipset that sits between the chips that makes the multiple DRAMs look like a larger capacity DRAM to the memory controller.

Also designed to maintain compatibility with existing systems is MetaRAM’s WakeOnUse power management technology. WakeOnUse, as the name implies, enables MetaSDRAM to remain in a sleep state until needed, allowing the memory to fit within current constraints. The company claims that its advancement has accelerated memory technology development by two to four years.

Without any system modifications, MetaRAM says that its memory technology will work with current AMD and Intel server and workstations.

“I've spent my career focused on building balanced computer systems and providing compatible and evolutionary innovations. With the emergence of multi-core and multi-threaded 64 bit CPUs, I realized that the memory system is once again the biggest bottleneck in systems and so set out to address this problem,” said Fred Weber, CEO of MetaRAM, and once the CTO at AMD. “MetaRAM's new MetaSDRAM does just that by bringing breakthrough main memory capacity to mainstream servers at unprecedented price points, without requiring any changes to existing CPUs, chipsets, motherboards, BIOS or software.”

Using DD2 MetaSDRAM, MetaRAM estimates that a four-processor server with 16 cores and a quarter-terabyte of memory would cost under $50,000 – a figure the company says is a 90 percent reduction in current system cost.

Perhaps for the reason of lowered server hardware costs, smaller firms and industries will suddenly find themselves able to afford high-end hardware, thus driving the rest of the industry. One such interested party in increasing chip sales is Intel, an investor of MetaRAM.

“The rapid adoption of Quad-Core Intel Xeon processors and platform virtualization, combined with the growth of data intensive applications, is driving demand for increased server memory capacity,” said Bryan Wolf, managing director, Enterprise Platforms, Intel Capital. “MetaRAM's technology presented an opportunity for Intel to participate as both an investor and a strategic technology collaborator to deliver a compatible solution that enhances system performance.”

Launching today are two chipsets based on the MetaRAM technology. The MetaSDRAM MR08G2 chipset enables 2-rank 8GB DIMMs and is available at $200 each in 1,000 kit quantities. For those who crave more, the MetaSDRAM MR16G2 chipset enables 2-rank 16GB DIMMs and is priced at $450 each in 1,000 kit quantities. Both chipsets are capable of functioning at speeds up to 667MT/s.

As MetaRAM is a fabless company, it partners Hynix Semiconductor and SMART Modular Technologies will be the first to bring to market 8GB dual-rank PC2-4200 registered DIMMs.

Besides increased memory space, moving to the new 8GB dual-rank modules will also cut down on power requirements. SMART believes that, by replacing four traditional 4GB dual-rank RDIMMs with two of 8GB dual-rank DDR2 RDIMMs, designers can reduce power consumption by up to 20 percent.



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Latencies
By SilentSin on 2/26/2008 9:40:31 AM , Rating: 5
But what about latencies? Every time we hear about something like this it *always* increases memory access latencies across the board. Especially at those low bus speeds. Not to say that this technology won't find its uses, but that severely limits its usability in applications where every nanosecond counts or bandwidth is a limiting factor.

I really don't think trying to reverse engineer a new memory technology onto existing chipsets is the way to go. 3D memory technology is a fantastic way to increase density per DIMM but to really make use of it they need to develop a native platform meant to leverage the advantages of it without the need for special chips on every stick.

I guess this is more of a proof of concept than anything, hopefully we'll see some large advancements with this tech in the next year or so.




RE: Latencies
By wrekd on 2/26/2008 9:52:39 AM , Rating: 2
The first ones are registered, which do have higher latencies


RE: Latencies
By Micronite on 2/26/2008 10:12:19 AM , Rating: 4
They won't be able to make an unbuffered (non registered) version of this. This technology is actually quite simple.
There are a few problems, however...

1) Latency - Not that big of an issue. There shouldn't be much more (if any more) latency than a standard registered DIMM. And much less than FBDIMM.

2) Addressing - Just because you have a 256GB module does not mean you can address it. Unless someone is willing to redesign their memory controller for DDR2, 16GB is as high as it gets (and in some cases 8GB). This will also depend on the board you're using. 8GB modules are already available, so what's the big deal?

3) Price - Stacking technology has been used for a while, so there's really not that much new about it. The problem is that it's still going to need the DRAM. Since it still uses DRAM, it is still going to be the same price per bit plus any premium for a very large density module.

4) Competition - OK, so I work for a large memory manufacturer, but looking at this realisticly, the market was already moving in a way that would allow much higher density modules. By the time this technology gets anywhere, they will be beaten out by the rest of the industry.

One thumb up for a good idea, two thumbs down for it being about two years too late. Now if you'll excuse me, I need to go amputate my third thumb.


RE: Latencies
By mcnabney on 2/26/2008 10:40:51 AM , Rating: 2
Have you priced 8GB FBDIMMS recently? I think the suggestion was that these DIMMs would have per GB pricing more on the line of 1 or 2 GB sticks.


RE: Latencies
By Anonymous Freak on 2/26/2008 12:46:34 PM , Rating: 1
Except that is rather ridiculous. They still need 8 GB of chips. THAT is the expensive part. Yeah, maybe they have some revolutionary new method for assembly that requires more, lower-cost chips, but that would still not make it significantly cheaper.

In addition, these aren't FB-DIMMs, so they don't suffer from FB-DIMM price inflation.


RE: Latencies
By NullSubroutine on 2/26/2008 12:59:21 PM , Rating: 4
That is because some DIMMs can only have x amount of memory chips installed on them. In order to achieve more memory per ram dimm right now you have to use higher density memory chips, which are more expensive. In this case you can use lower density memory chips, but have more of them, this in theory should reduce the cost of the Ram DIMM.

For concrete example, let's say for a moment that we are restricted to 8 chips on a single Dimm. To achieve 8GB we must put 1GB memory chips on there to equal 8GB of memory total. Let's also say the 1GB chips costs $20 dollars each.

Now lets say you use this technology can it allows you to have 16 chips on a single Dimm, so to create 8GB dimm you take 16 512MB memory chips. Let's say that 512MB chips cost $7 dollars each.

8GB - with 1GB chips = $160
8GB - with 512MB chips = $112
savings = $48 or 30%

(these numbers are all hypothetical)


RE: Latencies
By masher2 (blog) on 2/26/2008 10:47:41 AM , Rating: 1
> "Unless someone is willing to redesign their memory controller for DDR2, 16GB is as high as it gets "

We have plenty of systems in my lab with 64GB of RAM.


RE: Latencies
By murphyslabrat on 2/26/2008 12:20:15 PM , Rating: 4
He means the addressing of individual channels. There are limits on individual slots/channels as well as total RAM. On most systems with total addressing limitations at 8GB, the addressing limitation per-DIMM is at 2GB.


RE: Latencies
By Micronite on 2/26/2008 4:54:46 PM , Rating: 2
Yes, thank you. The AMD Opteron can address a quad-rank 2GBit-based DIMM for a maximum of 16GB in a DRAM x4 configuration.
Intel's FBDIMMs will also be able to handle 16GB DIMMs when the AMB's and DRAM vendors come out with the appropriate modules.

2GBit * 16 chips * 4 die per chip = 128GBits = 16GB


RE: Latencies
By eye smite on 2/26/2008 11:50:11 PM , Rating: 4
Yes I'm sure there will be latencies. I'm also sure that like any product it will be matured and refined to an acceptable level and should do fine in the mainstream.


RE: Latencies
By Sulphademus on 2/26/2008 10:05:18 AM , Rating: 2
More useful if they can produce them with ECC for servers. That is where this density will pay off first and speed is secondary to stability.

Then again, they do like to use the home market as guinea pigs for new memory to test that stability.


RE: Latencies
By EidolWays on 2/26/2008 12:33:55 PM , Rating: 2
There's actually an argument out there that ECC is no longer needed. One of the chief worries with any tech is alpha strikes affecting transistor operation and causing erroneous values. Memory cells are so small, now, and their capacitance still so large that the probability of an alpha strike (which particles can also come from chip packaging and not just outer space) adversely affecting anything is extremely low.

So the only way you should be getting errors is if you have a malfunctioning memory chip, PCB, or connection to the rest of the system.


RE: Latencies
By InternetGeek on 2/26/2008 6:38:15 PM , Rating: 4
Even an [i]extremely low[i] error possibility makes the technology unworkable. You need to have a 0 chance error and the only way to assure that is using a error checking/correction logic such as ECC.


Stacking.
By Mitch101 on 2/26/2008 9:25:29 AM , Rating: 5
Back in the Atari 520ST days in order to upgrade the memory you soldered the memory chip on top of the existing memory chips and ran a few jumper wires. You kids have it so easy with your sticks of ram.

It could also be that they finally were able to reverse engineer that Terminator they deny having parts of.




RE: Stacking.
By Samus on 2/26/2008 9:48:27 AM , Rating: 2
Hey, I was going to call a Terminator reference too ;)

Module stacking has been around since catapillar-style PROM's, but this is a little more complex than just 'doubling up.' The problem with stacking was always compatiblity. You must be dealing with a compatible system, as stacking draws more power and unknowing systems have conflicts of memory space.


RE: Stacking.
By Trisagion on 2/26/2008 9:53:49 AM , Rating: 2
Actually even today, if you want to increase the RAM on an embedded board, you have to solder additional SRAM modules over the existing ones. But thankfully that's not something we have to do everyday on our desktops and laptops.


RE: Stacking.
By 3kliksphilip on 2/26/2008 12:16:03 PM , Rating: 2
My Dad has an almighty 4 mb version up in the loft. I was stuck with half a meg in my room, which I played on constantly. Did additional RAM actually make much of a difference? (STOS Basic, Bubble Bobble, Turrican 2, Rick Dangerous, Lemmings, North and South, 3D Robocop, Robocod (love that game), Hyper Paint, Hyper Draw, (I made a 3D rollercoaster ride on it, drawn step by step!)... Those were the days.)

I had an Atari until the age of 12 because my parents said that I would never use a PC if I had one. They were sooooo wrong.


Cache
By wrekd on 2/26/2008 9:35:24 AM , Rating: 2
Time for another Gigabyte I-RAM model. Me hopes anyways.




RE: Cache
By Rugar on 2/26/2008 10:00:42 AM , Rating: 2
I'd just like to see the DDR2 version that was shown around 2 years ago.


RE: Cache
By drank12quartsstrohsbeer on 2/26/2008 10:15:21 AM , Rating: 2
Same here.


RE: Cache
By superkdogg on 2/26/2008 12:37:36 PM , Rating: 2
How about just building on this to make SSD's? Seems that all it would need is a failsafe battery backup and could supplant flash (granted-I'm not an engineer or hardware R&D) as a viable means of storage and could be faster with more write cycles before it spontaneously breaks.

But yeah, the DDR2 version that was promised would be great.....


wow
By RyanM on 2/26/2008 9:29:12 AM , Rating: 4
This is a pretty promising tech. The article fails to address what, if any, latency hit is incurred by the special controller.




RE: wow
By erikejw on 2/26/2008 9:48:32 AM , Rating: 1
It probably has a latency hit but a hd has 1000 times higher latecny anyway so the choice is easy if you have a memory need of tens or hundreds of Gbs. I guess the system won't be used for gaming anyway :)


RE: wow
By Oobu on 2/26/2008 4:19:19 PM , Rating: 2
Why would you need so much RAM for current games anyways? I'm using 4GB and running some of the most modern games available I rarely see it reach 3GB. That's not mentioning any other programs I have running in the background. For a workstation needing loads of RAM though, this seems like it would be incredibly useful.


what about DDR3?
By Screwballl on 2/26/2008 1:27:47 PM , Rating: 2
With many new systems offering only DDR3, I would assume it would be very easy to modify it to work with DDR3 as well since they share the same 240pin setup but at higher speeds and latencies. I am just curious why only DDR2 is mentioned... I assume that DRR3 in servers and workstations is not available or taking hold yet???




"When an individual makes a copy of a song for himself, I suppose we can say he stole a song." -- Sony BMG attorney Jennifer Pariser

DailyTech Poll
Do you use copy/paste on your smartphone? 




14 Comments









botimage
Copyright 2010 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki