FSB Wall question

[nightscope]

I have a question regarding processor fsb walls. Why do they occur? Not factoring in heat dissipation, why are there FSB walls to processors? It seems that these FSB walls only showed up with the Core 2 Duos...I remember with a pentium 4 and some amazing cooling, you can go to about 6 Ghz easily. So, why are there fsb walls to processors?

 

[Xpyrofuryx]

I've wondered this too

 

[Evilonigiri]

Well see here, the netburst architecture is designed for high frequencies. There's probably a wall for them too, just much higher. As for why, I have no clue. Does anybody know? I'm curious now...

 

[imrul]

italy oc'd a pentium 4 to 8ghz...

 

[starcraftfanatic]

I believe that the motherboard controllers for the FSB have a tough time keeping up with the high speeds and relaying information like it's supposed to, like the problems with proccesors and voltage leak.

 

[nightscope]

So it's not the processor, it's the motherboard that provides the FSB walls? So CPU frequencies are limitless without factoring in heat?

 

[cnumartyr]

So it's not the processor, it's the motherboard that provides the FSB walls? So CPU frequencies are limitless without factoring in heat?

Everything will have it's limits.

I don't think chips have FSB walls, the motherboards they are in do. As new motherboards come out the FSBs get higher and higher.

 

[nightscope]

Everything will have it's limits.

I don't think chips have FSB walls, the motherboards they are in do. As new motherboards come out the FSBs get higher and higher.

Well why do motherboards have fsb walls? It just gets too fast for it to handle? Or too hot?

Has this ever been tested to prove that chips have no fsb walls?

 

[cnumartyr]

What motherboard would you test it on?

I consider it like this because noone thought we'd see 400+ MHz FSB quads. It happened on some newer MBs. Now 450-500 MHz quads are slowly starting to appear (same kinds of chips).

Clock generators on the motherboard would be my guess.

 

[nightscope]

Well someone needs to test this! It just doesn't make sense to me. Higher frequencies result in a higher heat output, but if that is taken care of, then what's stopping the reach of a higher frequency? Blah...blah...blah...

 

[MarkG]

Well someone needs to test this! It just doesn't make sense to me. Higher frequencies result in a higher heat output, but if that is taken care of, then what's stopping the reach of a higher frequency? Blah...blah...blah...

You mean the actual frequency the FSB can run at? The FSB has shedloads of wires on which the signals all have to get to the far end at the same time, plus or minus a small fraction of a clock cycle. Laying them out is hard, and the higher the clock speed the lower the tolerance for variation between signals.

Also, the speed of light itself is only a foot per nanosecond, so that introduces limits on how far signals can travel at high clock speeds.

This is why the PC world switched from the parallel AGP bus to serial PCI-Express bus, for example; PCI Express has less wires so it's easier to lay out. More precisely, a 16x PCI-Express bus has lots of wires, but each lane has only four wires that have to be in sync.

 

[epsilon84]

Firstly, there is the FSB-strap of the chip in question. There are 3 levels of FSB straps, 800/1066/1333. The lower straps have more aggressive memory latencies than the higher straps. This can cause a 'FSB wall' on chips such as the E2x00 and E4x00 which default to the 800FSB strap. When you push the FSB to extreme levels, the chipset is still trying to communicate with the CPU via the 800FSB strap parameters, which would be too aggressive, and you have a 'FSB wall'. I've seen a mod where an E4300 modded to the 1066FSB strap attained a 70MHz higher FSB than the 800FSB strap.

Secondly, there is chipset limitation as cnumartyr said. This is especially true in quad cores compared to dual cores. You have (potentially) double the amount of data pumping through the FSB, which puts the chipset under a lot of stress.

 

[Evilonigiri]

Wait...now that I think of it, how about this?

The wires in the motherboard has a certain size, therefore only so much current can go through it. The thicker the wires, the more current can pass. Basically when you get in the high frequencies the current goes up (because voltage goes up), but it's being limited by the size of the wires in the motherboard. After all, only so much current can get through...OH YEAH and with more current, more resistance is present, thus making it exponentially harder to go faster! It's just like wind resistance ya?

So who agrees with me? Maybe I'm totally wrong?

 

[easyg]

Well why do motherboards have fsb walls? It just gets too fast for it to handle? Or too hot?

Has this ever been tested to prove that chips have no fsb walls?

You can test it yourself if you have an enthusiast class motherboard.

Suppose, for instance, you have a Q6600. Go into your BIOS and lower CPU multi on your Q6600 to 4 (enthusiast mb's support the CPU multi unlock feature). With the front side bus running at deault 266 MHz (1066), the CPU will be massively underclocked running at less than half its default frequency (266 x 4 = 1066 MHz). By the way, if you really try this, make sure to take the divider off your memory modules (running them 1:1 would be best methinks).

Now begin to raise the FSB in increments of 20. With a vanilla board (eg 975X), I guess you'll hit the "FSB wall" maybe around 500 MHz? Whatever, let's assume the system starts to cr@p out at 500.

Obviously, if your system (with CPU multi =4) is conking out with FSB running at 500 MHz, it's not the Q6600 that is overheating, because the CPU is still underclocked and cool at 2.0GHz (500 x 4 = 2000 MHz). Rather, it's the system bus that is running hot and unstable at 88% overclock (266 vs. 500).

 

[cnumartyr]

You can test it yourself if you have an enthusiast class motherboard.

Suppose, for instance, you have a Q6600. Go into your BIOS and lower CPU multi on your Q6600 to 4 (enthusiast mb's support the CPU multi unlock feature). With the front side bus running at deault 266 MHz (1066), the CPU will be massively underclocked running at less than half its default frequency (266 x 4 = 1066 MHz). By the way, if you really try this, make sure to take the divider off your memory modules (running them 1:1 would be best methinks).

Now begin to raise the FSB in increments of 20. With a vanilla board (eg 975X), I guess you'll hit the "FSB wall" maybe around 500 MHz? Whatever, let's assume the system starts to cr@p out at 500.

Obviously, if your system (with CPU multi =4) is conking out with FSB running at 500 MHz, it's not the Q6600 that is overheating, because the CPU is still underclocked and cool at 2.0GHz (500 x 4 = 2000 MHz). Rather, it's the system bus that is running hot and unstable at 88% overclock (266 vs. 500).

It'll happen a lot sooner than 500 MHz on a Q6600. For the record though, PLL/GTL Ref Voltages will help in raising the wall to a point.

 

[epsilon84]

easyg, the lowest multi available is 6x, not 4x, but your point still stands.

I also doubt a 975X would get anywhere NEAR 500MHz with a Q6600... 400MHz maybe... and that is a BIG maybe...

 

[systemlord]

What you are saying makes sense, because you have higher heat when the NB is trying to comunicate with the CPU and at the same time you trying to push more current. There comes a point when the heat is just too much for the mainboard circuit pathways to handle and at anywhere near 500 +FSB you must have lots of fan pointed directly at the motherboard.

Now with the processer you are dealing with how well it conducts electricity and most CPU chips behave differently even ones from the same waffle or batch. The higher bin processers are granted the names Extreme Editions because they conduct electricity so well thats why at their stock speeds, example; Quad Cores that only need 1.24 volts to run at 3.0GHz while still at stock volts as deemed by Intel.

The lower bin CPU chips are pushed down to models like the my processer (E6600) that needs more than stock voltages to reach the same speeds as an Extreme Edition, they have gotten much better through revisions and thats why it Soo much smarter to buy a Q6600 instead and overclock it to Extreme Edition levels to save you some major cash!

 

[easyg]

easyg, the lowest multi available is 6x, not 4x, but your point still stands.

I also doubt a 975X would get anywhere NEAR 500MHz with a Q6600... 400MHz maybe... and that is a BIG maybe...

Oops, sorry, my bad. Thanks to cnumartyr too, for pointing out my mistakes.

I was just trying to make an example and got a little careless.

 

[ffffff]

Holy crap. Those are some of the worst explanations I have ever heard.

First of all, the reason for FSB wall is the same reason for any other digital circuit to have a 'wall'. Digital circuits consist of datapath and control. Datapath is stuff like multipliers, adders, etc. Control is the state machine for controlling the dataflow and is implemented using registers. The maximum frequency or the minimum cycle period is equal to the Tsu (flip-flop set up time) + Tprop(propagation time of the datapath) + Tcq(clock to Q of the flip flop). As temperatures decrease, transistors are able to switch faster and all these values decrease. The relationship is very complex because many parameters scale different with temperature. In the end, regardless of how low the temperature is, Tcycle will no longer decrease and the circuit can no longer be clocked lower.

In addition, some digital circuit elements have a probability of becoming unstable (crossing clock domains). Increasing the frequency results in a higher probability of a circuit instability. So, if it's running fine, it still has a larger probability of failure, etc.

Oh yeah, there are lots of other factors associated with increasing voltage (more tunneling and leakage) and higher parasitic capacitance

 

[systemlord]

I wanted to make it simple so the beginner could have an easy way to understanding. I did not want to write an SA about capacitors, resistors and transistors. Making yourself feel better by insulting others is wrong and by doing so makes you look bad because you have to put down others to make yourself feel better. Oh I just love your name it shows intelligent.

 

[nightscope]

Yeah, that makes a bit more sense, but why does instability occur at higher frequencies? So too much stuff is going on for all the components to handle? How can this be fixed? Less wiring? How do newer chipsets offer higher frequencies? Better NB/SB heatsinks?

So basically, the motherboard is the bottleneck for the CPU?

 

[3Ball]

I thought I would point out the the wall for the C2D is around the same it was for the pentium 4's. Remember that the 8 ghz pentium 4 was running somewhere near I believe a 17 or 18x multiplier, which means. That they only need to reach between 440 and 470 fsb depending on the multiplier used. So with respect to that it appears the walls are similar placed at around 500 fsb (obviously depending on the chip, mainboard, system config, and cooling setup). This is no real technical data that I am presenting, but a mere observation...I just believe saying that the pentium 4's (or rather their chipsets) had a much higher FSB "wall" is slightly inaccurate. Hope this helps!

Best,

3Ball

 

[nightscope]

I thought I would point out the the wall for the C2D is around the same it was for the pentium 4's. Remember that the 8 ghz pentium 4 was running somewhere near I believe a 17 or 18x multiplier, which means. That they only need to reach between 440 and 470 fsb depending on the multiplier used. So with respect to that it appears the walls are similar placed at around 500 fsb. This is no real technical data that I am presenting, but a mere observation...I just believe saying that the pentium 4's had a much higher FSB "wall" is slightly inaccurate. Hope this helps!

Best,

3Ball

Oh 3ball you're so technical

 

[3Ball]

Oh 3ball you're so technical

Yea its just how I roll...lol

Best,

3Ball

 

[jitpublisher]

Yeah, that makes a bit more sense, but why does instability occur at higher frequencies? So too much stuff is going on for all the components to handle? How can this be fixed? Less wiring? How do newer chipsets offer higher frequencies? Better NB/SB heatsinks?

So basically, the motherboard is the bottleneck for the CPU?

In a sense, yes. But look at the die size of a CPU. They get to their high frequencies by utilizing die shrinks.
A motherboard is a complicated, BIG piece of architecture that must work with multiple components all running at different speeds, accross very, very long traces and paths by comparison. Some day, everything inside your PC will fit basically into a single integrated chip. The only reason for a "motherboard" will be so you have a place to plug in your input devices, if those are even still needed.