What effect does FSB have on the CPU?

[BlackLoTuS]

I know that less of a divider you use the better, but what are the limits that you can push the FSB and why is that so?

I remember back in the days of the Pentium MMX that I could overclock FSB on the TX chipset to 75 and even 83 Mhz and that would increase performance substantially (especially because L2 was run at that speed as well). Then when the Super Socket 7 platforms came out they had up to 100 mhz FSB. However I remember Tom and others not being able to get the Pentium MMX to work at 100 Mhz FSB even when run within the speed the chip was capable of. The K6's would work however, because they were designed to work at that FSB.

I always thought that you could use any FSB you wanted as long as the CPU could handle the speed it was set at. Why couldn't a PIII 800 be run at 100x8, 133x6 or 266x3 (assuming the chip is unlocked)? What is it that is stressed on the CPU when the FSB is cranked up? Does anybody know the FSB limitations of todays chips?

 

[Guest]

For quite a while now, Intel & AMD have been multiplier-locking their chips. For example, if you have a PIII 800EB (using a 133MHz bus), the multipier is locked at x6.0, preventing you from upping the bus speed without overclocking the actual speed of the chip.

Similarly for Athlons, a 1GHz Athlon is multiplier-locked at 10.0x (yes, that's 10.0x 100MHz, although the bus is double-pumped, the multiplier still relates to the basic bus speed), and the newer 1.2GHz Athlons are multiplier locked at 9.0x, or 9.0x 133MHz = 1200MHz. P4's are also clock locked, e.g. a 1400MHz P4 is clock-locked at 14.0x, bearing in mind the 100MHz bus is quad-pumped.

It is possible to overclock a chip by upping the FSB, for example, an 800MHz Athlon can run at 864MHz by upping the FSB to 108MHz, retaining the 8x multiplier. Celerons can be overclocked easily as well - they only have a 66MHz FSB, so a Celeron 600, running at 9.0x 66MHz, will run happily at 9.0x 83MHz = 750MHz, or even 9.0x 100MHz = 900MHz, providing you've got good cooling!

The only chips still (barely) available that are not multiplier-locked are AMD's K6-2 & K6-III, which both normally run with about 100MHz bus speed. This is mainly so older boards with only 66MHz or 83MHz can accept the chip, by upping the multiplier. E.g. 500MHz K6-2 normally runs at 5.0x 100MHz, but can run at 6.0x 83MHz, for older motherboards.

 

[yoda271828]

While that is true that modern CPUs are multiplier locked, AMD chips can be unlocked <b>very</b> easily. You just close all the L1 bridges on top of the CPU with a pencils' graphite and voila! An unlocked processor.

 

[BlackLoTuS]

I was not asking HOW to OC CPU's. I already know that and I know that all Intel CPU's hav ben locked for a while. I was asking why there would be a limitation on FSB speed? I mean say if I had a Coppermine with a multiplier of 3 (this is hypothetical), what would prevent me from running it at 266x3=800 Mhz providing that the chipset & BIOS would support the proper multipliers etc? I'm betting that you probably couldn't, due to some limitation of the FSB, but I don't know why that limitation exists.

Back when the Pentium MMX was unlocked and the super 7 platform was out, I could run a Pentium MMX 233 at 66x3(200), 83x2.5(208) but you could NOT run it at 100x2 which would only be 200Mhz. Obviously the CPU could handle 200 Mhz no problem so what was preventing it from using the 100Mhz FSB?

Likewise today I am wondering what the FSB limitations are for current processors (133,200,266,500?).

 

[girish]

hi!
the FSB of a CPU is Front Side Bus, which interacts with the chipset and other hardware. As a CPU cannot handle very high internal speeds, it cannot handle very high external (front side) speeds. Its silicon isnt made for that.

The FSB is dependent mostly on external conditions, these signals travel to the chipset over the motherboard, so they have natural physical limitations imposed by the imductance and capacitance. The same is true for onchip electronics, only that you can make finer and shorter conductor tracks on the silicon, the innards of the CPU can go this further. Since it is not possible to run the external hardware at this speed, why make the chip's interface to it that fast when it isnt being used at all? That is why the concept of FSB and its multiplier came around.

Designing for higher speed requires careful high quality design efforts which would increase its cost and its development time. Intel could have made a P-III 800 with a "physical" FSB of 400 MHz and locked its mutiplier to 2. But could the other circuitry and the signal lines on the motherboard PCB sustain this high frequency?

So as far as the frequency is within its operating limits, the CPU can very well work on it, regardless of whether it came from a 66 MHz clock or a 100 MHz clock. If its I/O pins that interact with the chipset and other hardware could work at higher frequency they will very well work. If you use an older motherboard or CPU to run at higher FSB, the CPU might not support it because it wasnt designed for it. Its external circuits cannot handle this high speeds although its inner core can.

girish

 

[girish]

Efforts are going on to increase the FSB, altough there are too many physical limitations. Apart from track inductance and board capacitance (which can be reduced by making smaller, shorter, finer tracks - that carry lesser current at lower voltage) and more important the radio emmision (imagine a motherboard full of bare connectors carrying 100 MHz clocks, they are little antennas transmitting the 100 MHz radio signals). What do you do about these? Besides there is possibility of cross-talk between signal lines (a data line picking up the signal on te adjacent line) this can be managed by providing a lot of grounding on the PCB.

This is why now they are working on organisational improvements rather than brute force methods of increasing clock frequency, the FSB. These include widening the bus, using both the edges of the clock (the DDR) or increasing memory "channels" so that there is more room, more effective throughput of the memory subsystem. Purpose is to take in as many number of bytes as possible in a given time in a set of cycles!

Today 133 Mhz is the highest official physical FSB although some have worked it around 150-166 MHz in controlled conditions. I expect the FSB going not much beyond 233 MHz next year. And not above 266 anytime. Let's hope they prove me wrong.

Hope this explains enough. Any additions on this?

girish

 

[Guest]

i see the effects of FSB mainly in two aspects:

1. out-of-spec FSB really stresses many subsystems, such as the PCI bus, AGP bus, RAM, and even IDE equipments, because all these subsystems are running at a set proportion of the FSB frequency.

2. out-of-spec FSB feeds the CPU with too much data than what the CPU is able to handle. a PIIIeb features a 64bit system bus interface running at 133MHz, which means it can handle 1Gbytes/s data throughput stably and steadily. at 133MHz, the systam bus talks to the Northbridge with ease. of course, this doesn't mean that there is no more headroom beyond the 1Gbytes/s spec, actually some have overclocked PIII's FSB up to 200MHz, i.e. 1.6Gbytes/s throughput, with some <i>extra cooling</i>. 1Gbytes/s is the spec'ed limit that a <b>stable</b> (unstrained and unoverheated) PIIIeb can deal with.

then how do we spec a certain CPU? i guess the first thing we should look into is not the CPU main clock frequency (667MHz, 800MHz, 1GHz, etc.) but the FSB. firstly, we need to find out the FSB limit for the current process/stepping/design. e.g, presume that we have a CPU die that can stand 200MHz FSB at max and runs flawlessly stable at 133MHz FSB, so we decide to spec the CPU at 133MHz FSB. secondly, we add in the multiplier factor to determine the main clock frequency of the CPU. e.g. we find the above CPU die runs flawlessly stable at 800MHz, then we lock its multiplier at that point, i.e. 6x. well, it's done: a new 133MHz FSB PIIIeb that can run at 133x6=800MHz.

so i guess the whole CPU spec determining process starts from the major limiting factor, the FSB, to the multiplier and the CPU main clock frequency. the other way around is wrong: we point at a CPU and say it is a 800MHz CPU and then simply divide the number into 8x100, 6x133, 4x200 or even 1x800. the major limiting factor is the FSB, which gates the data flow that the CPU architecture is capable of dealing with <b>stably</b>.

just my $0.02.


-----------------------------
Some are ignorantly happy,
While some, happily ignorant.

 

[phsstpok]

If it were just the CPU you would not have a problem as long as the resultant speed was reasonable. However, it is
the FSB speed that determines the speed at which the CPU talks to other components. With an Athlon/Duron system the PCI clock is 1/3 the FSB. On my KT7 motherboard memory can be, basically, FSB clock + PCI clock. Normally, that would be 100 Mhz + 33 Mhz (100/3) or 133 Mhz. If you boost the FSB to 110 then the CPU is trying to talk to the memory at 110 Mhz + 37 Mhz (110/3) or 147 Mhz. This is beyond the memory spec of PC-133.

That was just one example. There are other components that can't handle big speed increases, either. Some of them are integral to the motherboard.

You can buy faster memory but you can't change parts of motherboard.

Incidentally, on the KT7 you can also, essentially lock the the PCI bus at 33 Mhz and boost the FSB clock independently but you would eventually run into the same problems.

 

[jg38141]

True- Though I've had my pc133 running at 147, but it's ficle on boot up. Also some pci/agp cards may not work when the fsb is upped. I've left mine at 107 runs my ancient athlon 500 at 538(?) and my ram at 140 something. It runs very stable, no agp pci problems etc.

"Are you saying that I can dodge bullets?"

 

[Guest]

hoho~~he is asking about the effect of FSB <b>on the CPU</b>, not that on PCI or AGP or whatever.

-----------------------------
Some are ignorantly happy,
While some, happily ignorant.

 

[Guest]

Well, after reading you guys opinions. I have a question, well more like a problem. I have a TB 750@900 STABLE. (solid) can run up to 950 but glitch sometime. anyway, here is my config. (I'm trying to optimize my system to open up more bandwith to the memory.)

Abit KT7
TB 750 set at 9X100mhz
Kingmax PC133 256ram,
TNT2 32mb
SB live value
IBM HD, ata100 7200rpm

ANYway, when i try to use the FSB to OC, it just simply instable. I tried 8.5X105 and 107FSB. but after about 10mins it would come up error, as Kernel bla bla bla. and sometime crashes. I don't think it's the temp either, the CPU is running cool, and the VIA board, somewhere around 24C in MBM5. so that seems OK.

the thing i don't get is I can run stable 900 all week long w/o restart at 9x100 but i cannot run 8.5x105-107(892-910)
It's certainly not my MEMORY cause it been seen to run at 148+. I just don't get it!

 

[phsstpok]

The question I was answering was in the main body of BlackLoTuS's message.

"I know that less of a divider you use the better, but what are the limits that you can push the FSB and why is that so?"

As for what effect does the FSB have on the CPU?

The effect the FSB has on the CPU, in terms of speed is the CPU runs at a speed equal to the FSB clock times the multiplier. It's as simple as that.

 

[jg38141]

I understood his question and as the post above this one mentions the question has been pretty much answered, I was mearly adding to the fsb points a few other things that can become unstable even if the chip can handle the increase. Thanks for paying attention though.

"Are you saying that I can dodge bullets?"

 

[BlackLoTuS]

Thank you for understanding my question. That is what I needed to know. I understand now about the traces & crosstalk issue and that makes a lot of sense.