Slashdot Mirror
An Overview of Quad Band Memory
tedgyz writes "AnandTech has a short article on a new memory technology from Via, called Quad Band Memory (QBM). Rather than using dual-channel DDR to increase bandwidth, they use phase-shifting inside the memory modules to accomplish the same goal. The end result is simpler (and presumably cheaper) motherboard designs that are backwards compatible with current DDR modules. The downside? It is currently only going to available in a P4 chipset that Intel has not authorized."
The downside? It is currently only going to available in a P4 chipset that Intel has not authorized."
Keyword: currently. I'm sure the technology will be available soon for plenty of other motherboards. I don't consider this much of a downside (feel free to set me straight if I'm wrong).
Slashdot: Where people pretend to be twice as smart as they really are by behaving like children.
Given the memory manufacturers' resistance to DDR400 and the achingly slow progress that DDR2 is taking (the module standard isn't even final yet), this technology has a pretty good potential to reach production.
-h-
...how long will it take for the major chipset makers (Via, SiS) to adopt this technology? It'd be great to see this available on the Athlon platform in the near future.
If you can't beat them, arrange to have them beaten. -George Carlin
From the article:
*snip*
Here's where the difference between QBM and conventional modules comes into play; QBM modules will have a set of 8 registers (QBM-10) as well as a phase-locked loop (PLL). The purpose of the PLL is to take the incoming clock signal from the chipset and shift it by 90 degrees; this shifted signal is then fed to the second bank of the DIMM, while the first bank receives the unaltered clock directly from the chipset.
The 8 registers then switch between which bank gets to transfer data every clock; because of the 90 degree phase shift, there is a slight delay in transferring data from the second bank but both transfers actually end up happening within a single clock cycle. The end result is that you get two DDR transfers per clock, or 4 bits of data are sampled per clock thus doubling the throughput of DDR (hence the name Quad Band Memory).
*snip*
QBM modules will obviously be more expensive than regular DDR modules, the question of how much remains to be answered however.
Let's see, one PLL... damn, I don't know if I can afford the extra six cents!
(That extra six cents though doesn't detract from fact that this idea is just pure genius... with about 30,000 folks slapping their forheads for not thinking of it first!)
If it's *that* simple to double the data rate of memory, why don't they, for example, divide the memory architecture into eight sectors and have each bit of a byte on a different sector, making 16x memory? It seems that this philosophy has no limit, as long as you have lots of sectors. What's preventing people from doing that?
:).
Sorta like a beowulf cluster of chips, really
In essence, they're putting additional capacity in quadrature. How clever! Memory that shares technology with NTSC color television....
"Technical articles" basically consist of, "okay, here's this new chipset, let's compare it to an older chipset, (CLICK TO THE NEXT PAGE TO SEE MORE BANNER ADS). Wow, the new chipset (which just happens to be made by one of our advertisers) performs faster than the old one, as evidenced by this graph where the red line is 1/10 of an inch longer than the blue line."
I quit reading that tripe years ago-the incestuous relationships between those sites and their advertisers is even worse than AOL ads on TNT. Please, don't click through on this article and give these fuckers more undeserved money.
(-1, Raw and Uncut is the only way to read)
they use phase-shifting inside the memory modules to accomplish the same goal.
Isn't phase-shifting what happened to Jordi and Ensign Ro causing the crew of the Enterprise to think that they had died in a transporter accident?
Or is phase-shifting what the Traveler used to send the Enterprise to the center of the galaxy?
In any case, I don't ever recall Star Trek using phase-shifting to increase memory bandwith. Something's amiss here.
It probably will cost a buttload of money
There are only 10 kinds of people in this world... those who understand binary and those who don't
Actually AMD will be implementing the so-called DRM technologies in their processors as well. Read carefully in future please.
This is a link to kentron's FAQ about Quad Band Memory.
> Memory as well - how many of you TRULY saw a difference between PC100 and PC133 DRAM?
Yeah, the benchmark numbers don't lie - but again, those are JUST benchmarks.
You may call it JUST a benchmark, but the extra 50 fps I got in Quake 3 enabled me to crank the resolution up, along with the detail quality. This was from PC100 cas 3 to PC133 cas 2 on a 1.2 T-Bird.
Isn't this a technology that could be combined with dual bank motherboards? This would then provide 4x the bandwidth of standard single channel DDR.
I'm not thinking so much for main memory, but for graphics. Dual Channel DDR2 + QBM would be a very very good thing. Especially on something like Nv30.
Anyone know if it'd be possible to combine the technologies?
Even if Intel doesn't embrace it, AMD should. Fast memory is irrelevant for Athlon, but Opterons (especially multi-processor Opterons) could seriously take advantage of this.
Make a reference board, and others will follow suit.
/^[A-Z0-9._%+-]+@[A-Z0-9.-]+\.[A-Z]{2,4}$/i
The same held true for DDR boards last year, where you were pretty much only guaranteed to get registered, buffered memory modules to work. Now, pretty much any recent motherboard will accept pretty much any DDR module.
Compatibility and compliance always suffer at the beginning of a new product release. That's why technology and product reviews are so helpful.
Karma: Dyn-o-mite!(mostly affected by Jimmy Walker reading your comments)
The downside? It is currently only going to available in a P4 chipset that Intel has not authorized."
Why is this a downside? Why should I give a rat's ass what Intel "authorizes".
Intel sure as hell didn't authorize my Athlon on it's Abit mobo with a Via chipset.
Is there an actual downside to not getting Intel's blessing (downside for consumers, not the company making the mobo)?
What's next, CPUs that use dilithium crystals?
!#@%*)anks for hanging up the phone, dear.
Um, disk isn't a bottleneck on my system. I do lots of C++ programming, and everything gets cached in RAM after the first build. Thus, my bottleneck becomes gcc ...err... CPU and memory bandwidth :) Same thing when I'm doing 3D rendering (which had better fit in RAM or else) or playing 3D games and whatnot. I thought moving to a 4200 RPM laptop hard drive was going to be bad, after my nice 7200 RPM IDE RAID. In truth, thanks to the Linux VM, I don't notice the difference after the first half hour of using the system. However, I did notice the big boost that came with moving from a PC100 memory system to a PC266, even just palying around with GUI wigets (resizing and whatnot).
A deep unwavering belief is a sure sign you're missing something...
The bottleneck is most definately not the disk for most games anymore (the only thing I need a high performance PC for). They basically load the level/area into memory and you play, for hours sometimes in the same area, very little disk access until you load the next area or if you have too little memory.
Most server applications are definately bottlenecked by the disk since you serve more data than can fit into memory.
Q.
I always love morons who criticize things when they aren't part of their intended audience. Unless you're a gamer, you have no reason to be at those sites. If you are, those sites are invaluable. Quick, what's faster, the Radeon 8500 or the GeForce4 MX? I don't really care, my gaming is limited to some RPGs and the occasional CounterStrike match. However when I wanted to buy my brother a new graphics card, you bet he was happy that I did my research at those sites first. I used to frequent those sites a lot, back when I did a lot of 3D programming and gaming. I honestly needed the performance, and those sights provided a lot of good guidence, as well as useful ways to get cheap hardware to do cool things (overclocking Celeron 300A's and running them in SMP for example, or joining the bridges on Athlon XPs to get them running in SMP). As for the bit about advertisers, that's bull. The internet keeps these sites honest. A GeForce 4 TI 4600 really is a lot faster than a Radeon 8500. If a site says otherwise because of advertising, a second opinion (or third or fourth) is just a click away. In the future, it would be wise to provide specific proof before criticizing people of lacking journalistic integrity.
A deep unwavering belief is a sure sign you're missing something...
Via has not explicitly licensed the P4 bus. Via insists it has rights to the necessary patents through the purchase of Cyrix. If Intel officially approves this arrangement then they may lose some licensing sales in the future by setting a precedent.
The whole thing is kind of silly unless Intel is making money hand over foot in the chipset market. I wonder if their motivation to discourage 3rd party chipset development is to lock down control over various platform technologies? Sis currently makes P4 chipsets but they have a poor reputation for compatibility. Via has improved their rep by dominating the Athlon market. They might have the necessary market share to take the P4 platform in directions that Intel doesn't want to go.
How could you forget inverting the polarity? That seems to be the cure-all of the 25th century.
I guess it wouldn't do to see Jordi running around with a roll of duct tape and giving malfunctioning gadgets a thump to get them working again.
If Intel doesn't bless a chipset from Via and a large MB manufacturer like Asus declines to make a MB based on this chipset to avoid retaliation from Intel then I would say there is a downside for the consumer.
For example, the expected soft-fail rate of a computer memory system in Denver, Colorado is about 4 times greater than the rate expected at a city it sea level (such as New York City). Even in Leadville, Colorado (which is located at 10,151 feet) the expected failure rate is only about 13 times greater than in NYC. No location in Colorado even approaches 100x.
For more information, see the following paper:It can be found online here.
It is pretty pointless to read those sites every day. You might as well sit out on your lawn and watch grass grow. But if you are about to buy or build a new PC then they are quite handy.
Maybe you prefer to just throw your money at Dell and hope for the best, or you prefer the unbiased reviews of CNET (where expensive is always better). I like having more information at my disposal than that.
I like reading a few reviews from around the net. They each have different sponsors and biases, so you can balance them against each other to get a better idea of the truth.
How could you forget inverting the polarity?
Unlike the first generation SDRAM, which sent a word whenever the clock signal went from low to high, DDR (standing for double data rate, or Dance Dance Revolution, or East Germany) sends a word whenever "phi" (the clock signal) goes low to high or high to low, that is, whenever it inverts. QBM adds a 90 degree phase shift, which lets it send a word 1/4 of a clock after phi rises or falls.
Will I retire or break 10K?
Why should I give a rat's ass what Intel "authorizes".
Because you can't get warranty service from a company that's been sued out of existence for patent infringement.
Will I retire or break 10K?
Modern RAM ramps up at such a high voltage that it causes side effects that could be interpreted as signal when combined with other RAMs in phase.
Basically, rather than multiplexing, RAM has just gotten faster.
And as far as being analog devices, and not digital, that's not a very good accessment. While the signal that they produce is analog, they look more like sinc functions in the analog domain than sinusoids. Those kind of functions are good for DIGITAL systems. They are defined in such a way that part of the wave form is considered unusable, and that nothing should interfere with that part of the signal.
Compare that to modem signals, which look a good deal more like sinusoids - nice, slow, smooth curves, by comparison.
I suppose they could be made to phase shift, and do all that sort of thing, but computers as we know it would have to be redesigned to interpret signals much, much differently. We'd have to have wavelet processors, or something like that, and everything would have to go slower.
A good question is whether or not such slowdown would be worth it. Considering how well analog computers have done, perhaps not.
Mod me down and I will become more powerful than you can possibly imagine!
``only going to available in a P4 chipset that Intel has not authorized.''
Ladies and gentleman, here's one fine instance of shooting oneself in the foot. It just proves how stupid Intel is, that they don't want faster memory. I mean, it goes without saying that they aren't _going_ to authorize it...
Please correct me if I got my facts wrong.
That, I believe is an issue with how half-life was created, not the system. some of the half-life mods use much larger maps and never have the "loading" issue that the original half-life has. and these are more more detailed maps than the original. but I really now nothing about this. maybe someone in the know can add more?
---
I post links to stuff here
What if this could also be applied to DDR II (QDR) memory? THAT would give some REALLY impressive bandwidth.
The crappy side is that even if it can be applied, it's virtually guaranteed that the memory industry will take the wimp's way - first introduce DDR II, then wait a few years to introduce the dual-band DDR II. No sense in skipping a generation, that would just mean less revenue, right?
steve
Oh, you're not stuck, you're just unable to let go of the onion rings.
Wouldn't work. When one bank is driving the shared bus, the other has to be turned off, and standard DDR modules won't have the correct timing for that.
-- ;-)
Kuro5hin.org: where the good times never end.
When I first glanced at the title of this story, I thought of memories being shifted four times.
The first look at this Slashdot item prompted the mental image of thoughts regarding past events taking on four interpretations.
When I first read the text above, I imagined remembering something, then remembering it again, and again, and one more time, each time different.
The interpretation which before all others was formed in my mind after I read the information about which this post comments regarded bits of stored information being repetitively read and interpreted in different arrangements, more than three times, but less than five.
Why, no, my name isn't Mojo Jojo, why do you ask?
Good judgment comes from experience.
Experience comes from bad judgment.