clock multiplier technical questions

[redzoneos]

hey guys not sure if this is the appropriate section for this, but please move it if needed!

So I just want to learn more by means of a free education. I'm trying to understand all the different stuff about cpus/motherboards. if someone can elaborate an explanation about what the various numbers mean I would appreciate it. I know how to calculate the speeds, I just don't understand the WHY behind it.

so my motherboard has an FSB of 1600. what does that mean exactly?
my processor (e2160) is rated at 800mhz with a speed of 1.8Ghz, and a multiplier of 9 (i think)... what does all this stuff mean? again, I understand how to calculate the 1.8Ghz, but i'm confused about the need of a multiplier, what purpose does it serve?

From what I've read, the multiplier means that the cpu performs 9 (in my case) cycles for every 1 cycle of the motherboard. Why? Does that mean that all modern motherboards have a bottleneck in FSB, and the processor is working 9 times less efficiently?

Theoretically speaking, assuming my previous statement about the FSB bottleneck is correct, the best performance from a system can be expected when there is no cpu multiplier? So in this case, the cpu would complete 1 cycle for every 1 cycle of the motherboard??? I have to assume that I'm wrong about all of this, but if I'm right why is FSB technology so far behind? I mean we've had cpu multipliers since 486...

Thanks!

 

[endorphines]

i'm no expert... but as i understand it, it only goes through the fsb to communicate with other system components. so if it's calculating stuff on information that's already stored in processor cache it'll have work it can do without ever needing to go through the fsb, which is why it doesn't matter that it might be at a lower clock speed. which is also why ram that runs a clock speed twice as fast won't actually offer twice the performance because the processor's not always relying on the ram to get it's work.
as i said i'm no expert, so correct me if i'm wrong since i'd like to know how it really works.

 

[Phanboy]

FSB 1600 means that you have a 400MHz system clock, and data can be transfered 4 times per clock cycle (called "quad-pumped").

FSB is a parallel bus, connecting multiple agents over a relatively long physical distance. Due to these constraints, the bus cannot run as fast as a CPU clock can run. Due to this, CPUs are designed to run as fast as possible, and not be bottlenecked by the system bus. This is where the multiplier comes in. Also, there is no need for the system bus to run that much faster than the memory will allow.

As for your last paragraph, you can't separate the chipset from the CPU. The fastest system consists of running the constituent components as fast as possible. Only the CPU completes cycles-- the chipset acts as an system arbiter and memory controller.

 

[jsc]

hey guys not sure if this is the appropriate section for this, but please move it if needed!

Theoretically speaking, assuming my previous statement about the FSB bottleneck is correct, the best performance from a system can be expected when there is no cpu multiplier? So in this case, the cpu would complete 1 cycle for every 1 cycle of the motherboard??? I have to assume that I'm wrong about all of this, but if I'm right why is FSB technology so far behind? I mean we've had cpu multipliers since 486...

Thanks!

True. The FSB is the bottleneck. Unfortunately, there's a practical limit to how fast you can move data around on a PCB. That limit seems to be around 500 - 600 MHz now. And motherboards that can do that are very expensive.


As far as frequencies and clocks are concerned, the problem is one of terminology. You have processor speed, internal multipliers, bus frequencies, bus clocks, and memory clocks (or whatever your motherboard manual calls them).

Using an Intel E6600 for an example:
It has a fixed internal multiplier of 9. (Internal multipliers vary with the model of the CPU.)
It has a bus frequency of 266 MHz.
Processor speed = multiplier X bus frequency = 2.4 GHz.

Now it gets a little confusing. As Phanboy stated, the Intel bus is quad-pumped. That means that the CPU transfers 4 chunks of data each cycle of the bus frequency.

So ... bus frequency X 4 = bus clock: 266 MHz X 4 = 1066MHz.

DDR2 RAM transfers 2 chunks of data each cycle of the bus frequency.
So memory clock = bus frequency X 2 = 533MHz (DDR2-533 memory).

Now, if I do a moderate overclock to 3.0GHz:
Bus freq = 333 MHz X 9 = 3.0 GHz.
Bus clock = 333 X 4 = 1333 MHz.
Mem clock = 333 X 2 = 666 Mhz (DDR2-667 memory needed).

This is pretty much of a freebie. There's not much price difference between DDR2-533 and DDR2-667 memory. And the stock Intel HSF will work nicely at this frequency.

Now, if I push as hard as I can (at this point, ymmv):
Bus freq = 400 MHz X 9 = 3.6 GHz.
Bus clock = 400 X 4 = 1600 MHz.
Mem clock = 400 X 2 = 800 Mhz (DDR2-800 memory needed).

This is no longer a freebie. Memory costs more. You need either a very good HSF or water cooling. In my case, that last 600 MHz cost me about $120 last year.

 

[dragonsprayer]

false the fsb is not the bottleneck - if the cpu is running at 100% where is the bottleneck?

the bus carries data from the ram to cpu - if the bus was bottleneck the cpu would not run at 100%

all this bottlececk stuff - all you care about is fps not fbs

max fps is achieved when the gpu is unhindered and matched and the memory bandwidth is maximized and cpu speed is optimized.

or pick the right cpu for the gpu and then set the cpu up correctly buy the fastest low latency ram and match it too mobo and fsb

if you have bottle pop and you break the top off the pop comes out faster but do you get more pop? no! if the cpu is at 100% there is not wasted cpu cycles thinking too itself "come on fsb take some calculations already" in its free time while fsb is bottlenecking!

 

[Grimmy]

so my motherboard has an FSB of 1600. what does that mean exactly?

This basically means you can run a CPU that has a rated FSB up to 1600. So you can run slower rated FSB CPUs, like your E2160 (800FSB).

my processor (e2160) is rated at 800mhz with a speed of 1.8Ghz, and a multiplier of 9 (i think)... what does all this stuff mean? again, I understand how to calculate the 1.8Ghz, but i'm confused about the need of a multiplier, what purpose does it serve?

In this case, Intel has binned CPUs to run at certain stock speeds for the best stablity of that particular chip. Basically you can figure out the multiplier without looking for spec's. As long has you know the rated FSB, you can figure out its stock multiplier:

800 (rated FSB) / 4 (quad pumped) = 200 Base Speed
1800mhz (CPU speed) / 200 (base speed) = 9

It may be more tricky to figure out the rated FSB speeds of 1333, since they are now using multipliers like 8.5, 9.5, ect.

Its still doable though.. Example (E8500):

1333 / 4 = 333.25
3160mhz / 333.25 = 9.482370592648162040510127531883 or 9.5 rounded.

From what I've read, the multiplier means that the cpu performs 9 (in my case) cycles for every 1 cycle of the motherboard. Why? Does that mean that all modern motherboards have a bottleneck in FSB, and the processor is working 9 times less efficiently?

The multiplier is what determines its speed, in relation of its rated FSB speed.

Theoretically speaking, assuming my previous statement about the FSB bottleneck is correct, the best performance from a system can be expected when there is no cpu multiplier? So in this case, the cpu would complete 1 cycle for every 1 cycle of the motherboard??? I have to assume that I'm wrong about all of this, but if I'm right why is FSB technology so far behind? I mean we've had cpu multipliers since 486...

Thanks!

The rated FSB of the CPU is the speed that your CPU will process data. So basically your statement about the FSB bottleneck is incorrect.

 

[redzoneos]

thanks to all who replied I really appreciate it! I still find myself confused however, as some of you said there is an FSB bottleneck, while others said there isn't.

@grimmy:
thanks for describing all the math but I already knew all that. What I really wanted to know is why use multipliers at all? I mean if my example processor (E2160) is said to be 800mhz, why worry about dividing by 4 to get 200mhz, and then multiplying that by 9?? This is what I don't understand. What is the true function of the multiplier... There has to be a rational explanation for why it was created aside from the math, because ultimately math gives you a number, but the number means nothing without having a relationship. 1+1=2 but what is the 2? it could be 2 apples, oranges, male/female couple, etc. etc. what does the 9 for my processor represent? Is it 9cycles for cpu vs. 1 cycle of the motherboard as I stated???

@dragonsprayer:
that's exactly my point. If the cpu is running at 100% but it's completing more cycles then the motherboard FSB, then clearly there IS a bottleneck, because those extra clock cycles of the CPU are going to waste... That's what I'm trying to figure out.

@jsc:
you're more on the right track about what i'm asking I've never OC'd my PC, but I have OC'd a couple of my friends. Maybe I'm not experienced, but from my understanding, you achieve much better overclocks by raising the FSB of the motherboard while lowering the cpu multiplier. If I'm correct, this is an obvious indication of the FSB bottlenecking the processor.... maybe bottlenecking is not the right word, but it's certainly slowing down performance of cpus.

@phanboy:
If anything, I would think that your explanation would support my initial argument of FSB bottlenecking the cpu even more. I know that the FSB handles more processes then does the cpu, so if anything, more bandwidth is being used up by other hardware, and thus slowing down the cpu even more. but I thought recent motherboards separated FSB bandwidth to specific hardware (so cpu would have it's own bandwidth, ram would also, etc. based on a diagram I saw at wikipedia...not sure of accuracy)

@endorphines:
hah, join the club man, I've been doin' more and more research on this stuff since I've been wanting to buy a motherboard to OC the hell out of my E2160 but still upgrade to Quad 45nm at some point before december! Anyway, everything I research, leads to more research, which leads to more research... but this was the topic I was most confused about.

Ok, so I think I addressed everyone that provided me input. I just don't understand why they keep improving cpu's when the FSB has been slowing things down for YEARS!! (well... assuming I'm right about that... hah) Although, intel is coming out with directconnect or something like that soon... I'm not about to research that until I figure out FSB though!

 

[Grimmy]

@grimmy:
thanks for describing all the math but I already knew all that. What I really wanted to know is why use multipliers at all? I mean if my example processor (E2160) is said to be 800mhz, why worry about dividing by 4 to get 200mhz, and then multiplying that by 9?? This is what I don't understand. What is the true function of the multiplier... There has to be a rational explanation for why it was created aside from the math, because ultimately math gives you a number, but the number means nothing without having a relationship. 1+1=2 but what is the 2? it could be 2 apples, oranges, male/female couple, etc. etc. what does the 9 for my processor represent? Is it 9cycles for cpu vs. 1 cycle of the motherboard as I stated???

Well... if you already knew the math, I'm not understanding why the answer is eluding you.

CPU multiplier:

All Computers recognize the hardware in which they consist of prior to starting the operating system, the computer BIOS allows this function. The Basic Input Output System also allows its user to modify the recognized hardware to run faster or slower than it was originally intended. One such option is the CPU Multiplier.

The CPU multiplier is one way for processors to run much faster than the clock speed of the motherboard or RAM allows. For every tick of the front side bus (FSB) clock, a frequency multiplier causes the CPU to perform x cycles, where x is the multiplier.

For example, if the FSB has a clock speed of 133 MHz and the CPU multiplier is 10x, then the processor would run at 1.33GHz. This means a processor can run much faster without having to increase RAM speeds or motherboard speeds and worry about desynchronizations, or other speed-related issues.

The CPU multiplier is frequently used by PC enthusiasts to overclock a CPU to get somewhat higher performance from it.

One downside of the multiplier is that it only increases CPU speed. In the previous example there is a multiplier of 10x, but RAM still runs at 133 MHz, so the computer can only access memory at 1/10th of the processor speed, 133 MHz, the speed of the FSB. Because of this, many overclockers prefer to have lower multipliers with higher FSBs.

The quote above just says the same thing I explained to ya. That was from an older discription of what the CPU multipliers purpose is, although it talks about older hardware, like 133mhz ram. Now we have DDR2 memory, that is double rated, and so we have memory that can run faster then the CPU runs. The old 486 CPU you refer to, can execute 1 instruction per cycle. The Core 2 Duo can execute 4.

 

[jsc]

false the fsb is not the bottleneck - if the cpu is running at 100% where is the bottleneck?

the bus carries data from the ram to cpu - if the bus was bottleneck the cpu would not run at 100%

all this bottlececk stuff - all you care about is fps not fbs

max fps is achieved when the gpu is unhindered and matched and the memory bandwidth is maximized and cpu speed is optimized.

or pick the right cpu for the gpu and then set the cpu up correctly buy the fastest low latency ram and match it too mobo and fsb

if you have bottle pop and you break the top off the pop comes out faster but do you get more pop? no! if the cpu is at 100% there is not wasted cpu cycles thinking too itself "come on fsb take some calculations already" in its free time while fsb is bottlenecking!

I have to disagree with you here. First, for the sake of THIS discussion, any gpu considerations are irrelevant. When we are talking about fsb, we are talking about the processor core connection to the outside world.

Second, if the processor's bus frequency were not a bottleneck for the CPU core, why the drive to increase bus frequency? In less than two years, stock bus frequencies have gone from 266 MHz to 400 MHz. Memory technology has gone from DRAM to DDR2 to DDR3. Why? To increase memory I/O to the processor core.

Why? To minimize the time the cpu core has to wait for data transfers. A 3.6 GHz CPU operating at 450 MHz X 8 mult will yield a higher memory I/O than the same CPU operating at 400MHz X 9 because the higher FSB gets data into and out of the CPU core a little more quickly. In my case, I had two problems. I didn't see the expected 11% increase, only about 7%. And operating my MCP at 450 MHz with necessary voltage increase drove the MCP temp up past 85 deg. C.

Operating a motherboard in dual channel mode increases memory I/O. A quad-pumped bus transfers data to and from the processor core faster. And a large L2 cache operating at CPU core speed increases performance because the CPU core does not have to waste processor cycles waiting for access to main memory.

"if you have bottle pop and you break the top off the pop comes out faster but do you get more pop? ..."
Bad analogy. A better analogy would be the increased flow that a larger hose gives compared to a smaller hose operating at the same pressure.

The proper technical term for "bottleneck" (in this context) is "I/O bound". CPU's for a long time have been I/O bound. The way to alleviate that would be to operate the FSB and memory at CPU core speed. And that is technically and economically infeasible.