Why 'quad-pumped' ?

[imgod2u]

Actually, no, everything *is* analog. There simply isn't a way to generate a "digital" wave as that's physically impossible. All you can do is take samples at discrete times (i.e. every 1 ns) and not worry about what happened in between. This creates the illusion that the wave you were reading was "digital". But it wasn't. Now, it's physically impossible to generate a pure square wave, those waves usually have variations (that's similar to those of a sine wave). By decreasing the discrete time in which you read samples (i.e. read every 0.5 ns instead of every 1 ns) and assuming your wave has enough variations (like a non-perfect square wave does, it has an increasing rise rate and decreasing rise rate, just like a sine wave), you can effectively read 4 different samples from a single wave.

And no, the P4's bus does *not* work like AGP 2x/4x. Please refer to http://www.interfacebus.com/Design_Connector_AGP.html
Agp 2x is a 133MHz, 64-bit data bus.
Agp 4x is a 266Mhz bus.
All that was done was an increase in clockspeed.

I wouldn't put too much trust in Tomshardware's "technical" explainations in the future.

 

[imgod2u]

Originally posted by: bobsmith1492
Digital circuits use square waves, however (on..off..on..off). What's the deal w/ sine waves?? If you use a sign wave in a digital circuit, you just have to use an op amp to change it over to a square wave so it is useable.

Sine waves are not exclusively generated by positive-feedback opamps. There is no such thing as a perfect square wave. All waves generated have variations similar to that of a sine wave. Some circuits *do* use a sine wave as that creates less EMI (less spikes in changes in the EM field) as well as occurs naturally (through crystals) vs square waves.

The problem on a square wave is... the time from low signal to high is supposed to be instantaneous (and winds up being as close as physically possible) so there is no time to do two different reads/writes, as the whole purpose of the clock is to space out the data.

When we're talking about taking discrete samples 0.5 ns in between, you *can* read different samples in between a rise and a fall. But you generally don't want to make your wave instantaneous (as there's no reason to). You're going to sampling at discrete times anyway, does it matter how long the wave has been high? As you increase clockspeed, obviously your rise/fall times will have to be lower, but then, your sampling rate goes higher as well.
Making your wave rise/fall extremely fast unneccessarily simply creates more EMI and uses more energy (not to mention difficult to generate as they do not occur naturally in crystals).

Ok... looking at one clock (I wrote it down on paper in front of me , starting at the top - left corner of a square wave, one full clock goes from that point to the top-left of the next wave. If your data is transferred at the initial point, then at the top-right (or bottom right, really) of the first square, those are the only two points on the wave where the level changes, and are probably the points used for DDR. For quad data, however..... who knows?? Mebbe they have some sort of modulated wave, or a half-phase shifted wave used in conjunction? Someone needs to find an Intel engineer methinks....

Or, you could realize that the waves being used aren't perfect square waves. Sine waves occur naturally in crystals, not extremely fast square waves.

 

[VIAN]

Agp 2x is a 133MHz, 64-bit data bus.
Agp 4x is a 266Mhz bus.
All that was done was an increase in clockspeed.

According to the PDF I linked, It's not an increase in clockspeed, but just tricks like DDR. The clockspeed is still 66MHz.

Actually, no, everything *is* analog. There simply isn't a way to generate a "digital" wave as that's physically impossible. All you can do is take samples at discrete times (i.e. every 1 ns) and not worry about what happened in between. This creates the illusion that the wave you were reading was "digital".

But there are circuits that turn an analog signal into a digital signal through a rectifier.

Now, it's physically impossible to generate a pure square wave, those waves usually have variations (that's similar to those of a sine wave).

The point to digital is that the Variations won't matter in a digital world.

I don't know about sample times though.

 

[bobsmith1492]

Ok... so if the bus for a quad-pumped P4 runs off a sine wave, is the central 'access point' at ground level? If it is, then the system has to use negative voltage. Or, maybe the bottom of the sine wave is ground, and there is 'something' that is activated by the half-sine-voltage?

Anyway, according to the AGP specs, they use square waves (of course, they can't be perfect squares) not sine waves. The page 79-80 diagrams show a kind of ramped square wave, which is probably what they would actually look like. The problem I see with a slow falling/rising wave like a sine wave is... when would the transistors actually be activated?? They are all slightly different, so they would switch on a different points along the wave, especially when it is on a slow slope like a sine wave. I don't know if this would be a problem or not.

 

[imgod2u]

Actually, no, everything *is* analog. There simply isn't a way to generate a "digital" wave as that's physically impossible. All you can do is take samples at discrete times (i.e. every 1 ns) and not worry about what happened in between. This creates the illusion that the wave you were reading was "digital".

But there are circuits that turn an analog signal into a digital signal through a rectifier.

What would you consider a "digital" signal? The best rectifiers in the world won't be able to create a purely flat square wave. And they do nothing for the rise/fall edges. Not to mention this is not very useful to do at all. Simply take a sample of the sine wave at set intervals and it'll "look" like it's a digital signal anyway. That's the whole point of digital vs analog, it's not what signals you use, it's how you interpret those signals. Analog just takes a continuous signal and translates it directly into another form while digital takes discrete samples encoded in on/off. They both use electrical wave signals. The only difference is, the digital wave can be more flexible and doesn't have to match the desired output exactly (5V or 4.8V is all counted as "1", in analog, 5V would be a difference signal than 4.8V).

Now, it's physically impossible to generate a pure square wave, those waves usually have variations (that's similar to those of a sine wave).

The point to digital is that the Variations won't matter in a digital world.

I don't know about sample times though.[/quote]

Variations *do* matter. It's how you tell a "1" from a "0". But that variation is in a discrete amount, not continuous. If you split that difference into smaller and smaller quantities (i.e. 1.5V at time 0 is a 1, but 1.5V at time 0.5ns is a 1, 3V at time 0 is a 1, etc.). Again, it's not the signal itself that matters, it's how you interpret the signal.

 

[imgod2u]

Originally posted by: bobsmith1492
Ok... so if the bus for a quad-pumped P4 runs off a sine wave, is the central 'access point' at ground level? If it is, then the system has to use negative voltage. Or, maybe the bottom of the sine wave is ground, and there is 'something' that is activated by the half-sine-voltage?

Anyway, according to the AGP specs, they use square waves (of course, they can't be perfect squares) not sine waves. The page 79-80 diagrams show a kind of ramped square wave, which is probably what they would actually look like. The problem I see with a slow falling/rising wave like a sine wave is... when would the transistors actually be activated?? They are all slightly different, so they would switch on a different points along the wave, especially when it is on a slow slope like a sine wave. I don't know if this would be a problem or not.

Same way you interpret when a square wave is on or off. You define a cutt-off voltage/current. Anything below that is off, anything above that is on. Digital or analog, the difference is in the interpretation of the signal, not neccessarily the signal itself.
Of course, it won't be perfect sine waves, but it will have variations that make it not a perfect square wave, and will also have enough variation in the rise and falling edges to interpret 2 different samples.

 

[sao123]

?Correct me if i'm wrong cuz i'm not sure...
Cant you turn a sin wave into a a digital representation just by modifying the sampeling rate??

 

[VIAN]

Variations *do* matter. It's how you tell a "1" from a "0". But that variation is in a discrete amount, not continuous. If you split that difference into smaller and smaller quantities (i.e. 1.5V at time 0 is a 1, but 1.5V at time 0.5ns is a 1, 3V at time 0 is a 1, etc.). Again, it's not the signal itself that matters, it's how you interpret the signal.

That's good. I learned something new. By variations, I meant small fractions of a volt, noise and interference on standard TTL. But I see what you mean. You just opened my mind to other things.

 

[imgod2u]

Originally posted by: VIAN

Variations *do* matter. It's how you tell a "1" from a "0". But that variation is in a discrete amount, not continuous. If you split that difference into smaller and smaller quantities (i.e. 1.5V at time 0 is a 1, but 1.5V at time 0.5ns is a 1, 3V at time 0 is a 1, etc.). Again, it's not the signal itself that matters, it's how you interpret the signal.

That's good. I learned something new. By variations, I meant small fractions of a volt, noise and interference on standard TTL. But I see what you mean. You just opened my mind to other things.

Those would be "unwanted" variations I suppose. Technically, analog is "faster" than digital as you can parse more information. Think of analog as "infinite-pumped" where you're taking infinite samples instead of 2 or 4 per clock. Each tiny, infinitely-small variation in the wave counts as a separate level. It won't be just 1's and 0's either, so you could have potentially an infinite amount of information in one wave. Of course, being able to generate and accurately read that would be.....near impossible. We'll get closer and closer as technology improves.