25.6 GB/s of video bandwidth?

[Raystonn]

It looks like we're finally going to be squeezing more than a single bit at a time from each wire. It's about time!

<A HREF="http://www.hardwarecentral.com/hardwarecentral/reports/1973/" target="_new">http://www.hardwarecentral.com/hardwarecentral/reports/1973/</A>

Almost all component interconnect technology today transfers data using 0's and 1's represented by voltage levels. Traditionally a "1" was presented by 5 Volts and a "0" by 0 Volts. RSL used a reference voltage of 1.4 Volts, wherein a "1" was presented by 1.0 Volts and "0" by 1.8 Volts. QRSL uses four voltage levels to represent two bits of information: Logic 10, 11, 01 and 00.

This multi-level signaling technology allows higher bandwidth, double that of RSL, without increasing clockspeed. In this case, a single memory device is capable of providing up to 3.2 GB/s of memory bandwidth, using only a 16-bit databus. If scaled upwards to a 64 or even a 128-bit databus as often used in today’s videocards, this comes to:

64-bit / 16-bit = 4 : 4 x 3.2 = 12.8 GB/s for a 64-bit databus
128-bit / 16-bit = 8 : 8 x 3.2 = 25.6 GB/s for a 128-bit databus

--------

I can imagine we will end up with at least 8 distinct voltage levels representing a full byte per wire in the not-so-distant future. That will be an 8-fold increase over today's bandwidth without taking into account adding more pins (wires) or increasing clockspeeds. In my opinion what we need is more of these new algorithms and technologies. They are much more effective than simply pushing the old stuff with higher clockspeeds and increased pins.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[74merc]

seems there would be a lot of low level reprogramming involved there...
maybe not so much in the actual software, but the HAL or whatever from the BIOS/hardware to the OS.

Octo-datarate motherboard running at 100Mhz... lol
DDR800 without the added latencies...

----------------------
Independant thought is good.
It won't hurt for long.

 

[Raystonn]

It would just require reworking of the busses and the areas in the components that communicate with the busses. That's not actually as large a change as it seems. The system would look pretty much identical, with just about the same hardware. The components would just interpret the voltage levels differently when they read or write on a bus.

This new technology can be used on every bus: frontside (memory) bus, AGP bus, PCI bus, etc. Imagine keeping the same 32-bit PCI bus, but all of a sudden data is being transmitted 8 times faster over it because each wire indicates an 8-bit value rather than a 1-bit value. This would actually require 256 voltage levels, not 8 as I previously stated. We would require much greater precision in voltage across the system if we were to use the same 5 volts as we currently use. 0.02 volts per step uses about 5 volts for 256 values. 0.047 volts per step uses about 12 volts for 256 values. It can be done. Technology is such a wonderful thing.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[Raystonn]

"Octo-datarate motherboard running at 100Mhz... lol
DDR800 without the added latencies..."

More like the equivolent of 1600MHz per channel on a 100MHz frontside bus when used as double-data-rate SDRAM. With video cards using 128-bits we can have 8 channels of RDRAM internally on the video card and still use the same number of pins as current video cards use. That's a heck of a lot of bandwidth.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[somerandomguy]

Sounds nice, but how complicated is it to do in practice? will it increase latency? and how much are Rambus going to charge us for it?

"Ignorance is bliss, but I tend to get screwed over."

 

[Raystonn]

"how complicated is it to do in practice?"

It will require no big changes in the way we manufacturer components. Thus, the costs are negligable. There will be no requirement of increased clockspeeds or extra pins/wires. Thus extra performance without extra physical parts. We're just using the same hardware in a more efficient manner.


"will it increase latency?"

No. Latency per pin/wire will remain the same as it has been. We just get more data per pin/wire now.


"how much are Rambus going to charge us for it?"

You always pay for new development. R&D costs money and all companies, even AMD, exist to make money. I would imagine the royalties on something like this would be pretty much in line with royalties throughout the industry. Royalties on RDRAM are only around 2-2.5%. They're mostly negligable. A $50 stick of RDRAM has $1 of royalties built into the price. As far as an impact on consumers goes, that's nothing.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[Guest]

So
Rayston technologies will lead the way... if every one knows this why don’t they do it?
Bottlenecks on Joe averages?

Royalties.ha
Not worth the paper it’s written on.



~I want a vapochill inside, but who gives 2 [peeps]~

 

[Raystonn]

"if every one knows this why don’t they do it?"

If everyone knows what why don't they do what?


"Royalties.ha Not worth the paper it’s written on."

What do you mean?


-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[HolyGrenade]

I wonder who they stole this piece of technology from?


<font color=red><i>Tomorrow I will live, the fool does say
today itself's too late; the wise lived yesterday

 

[Raystonn]

I doubt they stole this from anyone. They have some great engineers. If you'd like to rake someone over the coals, go for their lawyers.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[flamethrower205]

Quite amazing. Imagine that w/ quantum computing....oops I drooled on myself.

When I rule the world, Apple will only mean the fruit.

 

[HolyGrenade]

You must've read the report on how they stole and patented the sdram stuff.


<font color=red><i>Tomorrow I will live, the fool does say
today itself's too late; the wise lived yesterday

 

[Raystonn]

Welp, they recently managed to create a tiny piece of quantum memory, totalling.. *drumroll* 2 bits. They'll have to drastically pick up the pace on that one for us to see anything on this within the next 15 years. I don't think we're really that close to the end of the line on silicon. We are only now beginning to realize the benefits of parallel concepts, from desktop SMP systems to parallel execution units. What we have now is basically the computing power of a really large fast truck on a single lane highway that can carry our loads. (The cargo is our work to be done or the data to be transferred.) Wait until we add more lanes and turn that into a multi-lane freeway.

Increasing the clockspeed (for memory) and moving to DDR/RDRAM technology is the equivolent of installing a better engine in that truck to allow for higher top speeds. Increasing acceleration is akin to reducing latency. This new technology is akin to at least doubling the cargo capacity of our truck without slowing it down at all. It took long enough to start looking at the cargo sizes, but I'm glad they finally did.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[Raystonn]

That is still making its way through the courts, but it does look bad for them. On the other hand, RDRAM (and this new technology) are their own creations. Just because they have evil lawyers doesn't mean they lack adept engineers.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[Guest]

To slightly paraphase my favorite line from "Escape from New York..."

RAMBUS???? I thought you was DEAD!!!!

 

[rcf84]

Well if i believe the Radeon 8500 Maxx should be around 15gb's to 20gb's of bandwith when released. Image the Fill rate.

Nice Nvidia and ATi users get a Cookie.... Yummy

 

[juin]

That not yellowstones ????????

 

[Raystonn]

Try <A HREF="http://babelfish.altavista.com" target="_new">here</A>. (That is, unless you are talking about the national park, in which case I must commend you for your non-sequitor.)

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[Matisaro]

LoL, juin what the hell is the matter man, you have gone from hard to understand, to impossible to understand to speaking total gibberish!

~Matisaro~
"Friends don't let friends buy Pentiums"
~Tbird1.3@1.55~

 

[HolyGrenade]

Its not the lawyers. They're just doing their job. The whole company is corrupt. Read <A HREF="http://www.fortune.com/indexw.jhtml;jsessionid=YXWMI5Y2ICBM0QAMEHTSFFKABQQ4KIV2?channel=artcol.jhtml&doc_id=203141&page=1&_DARGS=/artcol.jhtml.3_A&_DAV=artcol.jhtml" target="_new">this story</A> on Rambus. It is an extremely good article. Raysonn, this will show you what they are really like.

Lawyers are universally evil. Pointing out a lawyer as evil is just like saying satan might be evil. The thing is rambus are like that, if they see anything they like, and it isn't patented, they'll be running to the patent office.


<A HREF="http://www.fortune.com/indexw.jhtml;jsessionid=YXWMI5Y2ICBM0QAMEHTSFFKABQQ4KIV2?channel=artcol.jhtml&doc_id=203141&page=1&_DARGS=/artcol.jhtml.3_A&_DAV=artcol.jhtml" target="_new">http://www.fortune.com/indexw.jhtml;jsessionid=YXWMI5Y2ICBM0QAMEHTSFFKABQQ4KIV2?channel=artcol.jhtml&doc_id=203141&page=1&_DARGS=/artcol.jhtml.3_A&_DAV=artcol.jhtml</A>


<font color=red><i>Tomorrow I will live, the fool does say
today itself's too late; the wise lived yesterday</i></font color=red><P ID="edit"><FONT SIZE=-1><EM>Edited by HolyGrenade on 09/06/01 06:52 PM.</EM></FONT></P>

 

[HolyGrenade]

I think it would be completely valid of you to call rambus as snake. But snake plissken? thats just err...



<font color=red><i>Tomorrow I will live, the fool does say
today itself's too late; the wise lived yesterday

 

[FatBurger]

Good call, Raystonn. Thanks for doing us all a favor.



<font color=blue>Quarter pounder inside</font color=blue>
<font color=red>Change the Sig of the Week!!!</font color=red>

 

[Raystonn]

From reading that article I got the distinct impression that only 2 people plus the team of lawyers are responsible for what has occurred. Those would be Tate and Crisp. I would hardly call that the whole company. Their engineers have done some nice work on RDRAM and this new technology. I'm not going to hold them responsible for what these others have done. I will only hold Tate and Crisp responsible for the Patent fiasco.

By the way, did you miss the following from the article:

"In 1996 the FTC charged Dell with hiding patents from a standards body and ordered the company to give up the patents in question. The FTC is also investigating Sun, which settled a suit earlier this year involving similar allegations."

It would seem Rambus is not alone in this behavior.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[Guest]

"I can imagine we will end up with at least 8 distinct voltage levels representing a full byte per wire in the not-so-distant future "

Last time I looked 8 states (voltage levels) only allowed you to represent 3 bits (not an 8 bit byte) or is this the new math?
It is still interesting though

 

[Arbee]

It must be new math .

Anyway, this resembles QDR technology.

How terrible is wisdom when it brings no profit to the wise