Explain FSB to me?

[Deusfaux]

Ok so first of all... why is bus speed quoted in values 4x what the FSB as can be altered is?

like C2D are 1066 mhz.. which is "quad pumped" 266mhz. why both of those values?

clock speed is the latter x the multi anyways so..?

and the differences in bus speed between different C2Ds, like the E4XXX's being 800 and the refreshes being 1333.

Are these anything other than overclocked or underclocked FSBs? like the E4XXX's have a FSB of 200mhz and the refreshes 333mhz?


and if THAT is true, then assuming all other things equal, cant all differences be closed or erased with overclocking?

Isnt a C2D refresh with the same multiplier and cache as an original C2D the exact same chip, just with a higher factory clocked FSB, which you can up to yourself anyways?


Of course this leads to questions like would anyone find it worthwhile to upgrade fomr a 1066 C2D to a 1333.... but I also just want to understand the terminology and what's going on.


Do AMD procs have quad pumped FSB? are they measured and calculated in the same way?

sigh

 

[buffbiff21]

The front side bus is the electrical interface that allows the CPU to communicate with the northbridge chip (memory controller).

Ok so first of all... why is bus speed quoted in values 4x what the FSB as can be altered is?

like C2D are 1066 mhz.. which is "quad pumped" 266mhz. why both of those values?

Right. It's called quad data rate; DDR2 memory is double data rate. Wikipedia says:

can transfer up to four words of data in each clock cycle

...but I would put that in "English" by saying it allows for more memory thoughtput.

and the differences in bus speed between different C2Ds, like the E4XXX's being 800 and the refreshes being 1333.

The cpu communicates with the FSB via those tiny gold pins on the bottom of the PCB board. They are called BSEL. They tell the mainboard what FSB to run. All it is, is the stock setting. You can overclock this obviously.

Isnt a C2D refresh with the same multiplier and cache as an original C2D the exact same chip, just with a higher factory clocked FSB, which you can up to yourself anyways?

It depends on the core. E6600 uses Conroe, E4300 uses Allendale. (I believe the lower end E6300/E6400 now use Allendale as well.. if someone else wants to chime in on this)

 

[RamonGTP]

266MHz is the actual speed it runs at... Quad pumping means that data is sent 4 times for each clock cycle.

[CPU]-----fsb-----[northbridge]--------[DDR/2 memory]

The link between the NB and memory is double pumped, meaning data is trasferred twice per clock cycle, once on the rising edge and once on the falling edge. The FSB takes that a step further and transmits data 4 times per clock cycle.

 

[Glow]

Are certain chips 1333 / 1666 or is it dependant on the border if it will run 1333 or 1066?

 

[Deusfaux]

ok so i dont think the last one was quite addressed


but assuming everything else is the same, is a 1333 FSB conroe the exact same chip as a 1066 conroe, and the difference is overcome with overclocking?


Ie, the former has a 333 FSB, the latter a 266, and you can overclock the latter for hte former?


and assuming allendales were the same as conroes apart from the 800 FSB... you could overclock that 200mhz FSB up to "regular" conroe levels.... right?


if it werent for varying cache levels, multipliers and a few new little features in the freshes like TXT executions or whatever.... would there be any benefit into goin from a 1066 mhz to a 1333 mhz conroe?



why is it listed as a feature so prominently if its something that can already be acheived with relative ease? if the end clock speed is the same anyways too (by having a lower multi)... is it any better?

why do new motherboards have to come out with 1333 FSB support... or why is it listed as a feature... cant current boards get that fast... and much faster? (ive heard of 500 FSB clocks and such)

 

[arentol]

and if THAT is true, then assuming all other things equal, cant all differences be closed or erased with overclocking?

Assuming EVERYTHING else is equals, then yes, all differences can be erased with overclocking, though this can void your warranty if something goes wrong.

Of course this leads to questions like would anyone find it worthwhile to upgrade fomr a 1066 C2D to a 1333.... but I also just want to understand the terminology and what's going on.

Actually it might be an overclocking disadvantage to purchase a 1333 C2D...

Because the native bus is 1333 the multiplier to get to the same GHz will be lower. For instance, a 1333 C2D running at 2.66GHz will be sold for more than a 1066 C2D running at 2.4GHz. However, the multiplier on the 1333 will only be 8x while the multiplier on the 1066 will be 9x. This creates a situation in which it is quite possible your motherboard will limit the 1333 CPU far more than it will limit that 1066...

For example, lets assume your motherboard has an FSB limit of 1600 Mhz (400 actual FSB) and going higher makes the system unstable. In such a case the 1333 will cap out at 3200 Mhz (8x400) while the 1066 will potentially be able to reach 3600 Mhz (9x400).

So basically it may turn out that if you are willing to OC, but don't want to spend a ton on a very high end motherboard, then you might be better off with an 800 Mhz FSB CPU with a very high multiplier than with a high FSB CPU with a low multiplier.... Of course the lower end CPU's are usually binned that way for a reason, but considering how the E4300 has turned out the E4500 (11x multi I believe) might just be the low-end motherboard overclocking CPU of choice. (unless you consider the cache difference to be imporant)

 

[RamonGTP]

Everything else being equal (cache, multiplier, etc etc) and assuming the lower chip can overclock that high, then yes you can overcome the difference with OCing. If there are core improvements in one vs the other though, overclocking won't get you those.

 

[Deusfaux]

why was/is the E6300 so popular if the multi is the lowest of the range?

doesnt that mean its the hardest to oc to high levels... or at least harder than one of the ones with higher multis (they wouldnt have raise the FSB as much to acheive same end clock speed)

is it purely because its the cheapest? isnt that at least partially negated by having to buy more expensive mobos capable of handling really high and stable FSBs?

were the allendales (E4XXX) the same chips otherwise but with lower FSBs @ 800/200mhz and their higher multis... they'd be a much better choice than the lower end conroes cuz they would be easier to OC on top of being cheap.. right?

but they're a worse core overall so they're not always the better choice...

 

[RamonGTP]

Most people who are looking for a high OC will get a good board regardless of the multiplier settings though

 

[Glow]

For example, lets assume your motherboard has an FSB limit of 1600 Mhz (400 actual FSB) and going higher makes the system unstable. In such a case the 1333 will cap out at 3200 Mhz (8x400) while the 1066 will potentially be able to reach 3600 Mhz (9x400).

But lets say a 4300 has a multiplier of 9 isn't it stuck there? So how does the 1066 vs. 1333 help. (I'm still confused)

Good post OP btw.

 

[luihed]

I can do a simple mod to my cpu right now and turn it to an E6850 ( 1333fsb x 9multi ) thats not even available till fall......

Most people who are looking for a high OC will get a good board regardless of the multiplier settings though

If the available motherboard can clock high enough so it wont be a bottleneck then its ok..... If the E6400 have 1333fsb right now, your not going to see many people with them ovreclocking over 3gs....... with a multi of 6x, youll need a 500fsb motherboard just to hit 3gs, If you get really lucky ( and I mean really lucky ) and get a 600fsb motherboard you can hit 3.6gs.... just need to find rams thatll do 600fsb lol.......

@glow....You can multi down on Core 2....

 

[Lucio]

But lets say a 4300 has a multiplier of 9 isn't it stuck there? So how does the 1066 vs. 1333 help. (I'm still confused)

It depends on the chip, the selling point of the Xtreme edition chips is that their multiplier ISN'T locked to a specific value, you can tweak with it freely both up and down (within limits still). The other chips are locked to a specific multiplier (though some motherboards allow it to be lowered if you want to try say a 7 x 500FSB)

 

[Glow]

It depends on the chip, the selling point of the Xtreme edition chips is that their multiplier ISN'T locked to a specific value, you can tweak with it freely both up and down (within limits still). The other chips are locked to a specific multiplier (though some motherboards allow it to be lowered if you want to try say a 7 x 500FSB)

I'm really just curious since some boards can go 1066 and others can't so for me whos looking at a e4300 should I be looking at a board that could do 1066 or 1333?

Thanks OP now I'm uber confused

 

[Deusfaux]

why would you want to multi down ever?

isnt it harder to get a higher clock speed by raising the FSB?

so if you start out with a higher multi isnt that doing half the work for you?

 

[luihed]

To get maximum mhz you would always use the highest possible multi but there are situations where using a lower multi is better because of the higher fsb..... Nice to have the option to do it....

 

[Deusfaux]

are you saying having higher FSB in of itself can be a good thing?

its not purely a means to acheive a clock speed in the end?


the same core at 266 x 10 (2.66ghz) is NOT totally equal to one at 333 x 8 (2.66ghz)?

there would be applications or areas where the higher FSB would be better?

 

[RamonGTP]

A higher FSB equals more bandwidth between the CPU and the northbridge.

So a CPU running at 333x8 is going to perform better than one at 266x10 even though their clock speeds end up being identical.

Oh, and to answer your question about AMD's, they don't have an FSB at all. AMD processors basically have the northbridge chip built into the CPU, so on an AMD system, instead of the layout looking like the "diagram" I posted previously, it would look more like this...

CPU+NB------Memory

The link speed between the CPU and the memory is whatever speed you have the memory set to. AMD's do have a "reference clock" but all it is is a clock speed at which all other speeds are derived from. The default reference clock is 200MHz. From these reference clock, there are several dividers/multipliers to generate the CPU speed, memory speeds and hypertransport speeds. So if the CPU had a 10x Multi, it would be a 2GHz processor. There is also a 5x hypertransport multiplier which gives you a 1GHz hypertransport link. Hyper transport is double pumped so you often times see it advertised as 2GHz.

Hyper transport is a chip to chip interconnect which doesnt have much bearing on single socket CPU performance. The only thing you really NEED to know about hypertransport is to not overclock it... So if you raise the reference clock to 250MHz on an Athlon system, you'll want to lower your hypertransport multiplier to 4x instead of 5x. And don't worry about running the HT lower than it's rated speed, there will be no performance difference unless you take it down to like 200mhz.

 

[arentol]

why was/is the E6300 so popular if the multi is the lowest of the range?

The e6300 is popular for many reasons, but principally because there are plenty of motherboards these days that can exceed a 400mhz FSB quite easily. For just a little bit more than the price of an e6600 you can by a Gigabyte 965P-DS3, and e6300, and a decent HSF and easily hit 3000 mhz or higher in a decently cooled case.

The point about wanting a high multiplier is for when you are running a really cheap mobo, or one that just doesn't OC well for other reasons, like with many mini-ATX motherboards, or the Gigabyte 965G-DS3, which doesn't OC as well because of the on board video.

But lets say a 4300 has a multiplier of 9 isn't it stuck there? So how does the 1066 vs. 1333 help. (I'm still confused)

1066 vs 1333 doesn't matter at all inherently if your motherboard can already run 1333 since you can take any chip and up the FSB to 1333. The difference is that a 9x chip will then be running much faster (333 Mhz faster) than an 8x chip.

So why does Intel release a separate 1333 Mhz part? Because there is a big difference between OFFICIAL CPU speeds and what we OC'ers do. Intel will probably sell 10,000 times as many CPU's to businesses and normal home users than they will sell to enthusiasts. These non-enthusiasts will run 1066 CPU's at 1066 with stock intel HSF's, and when 1333 MHz CPU's come out they will run those at 1333 MHz FSB. Without the official release of these new higher FSB processors those people would never get the benefit of a higher FSB, nor would enthusiasts necessarily get the benefit of motherboards built to support those FSB's and far beyond.

I'm really just curious since some boards can go 1066 and others can't so for me whos looking at a e4300 should I be looking at a board that could do 1066 or 1333?

Thanks OP now I'm uber confused

Okay, first keep in mind that some boards are rated 1066 but can do 1333 or well beyond. This is especially the case with older enthusiast boards that don't have all the appropriate bios updates yet (though most do these days). Regardless, a 1333 mobo should always do 1333 quite easily, and therefore will likely go a bit further as well, certainly to at least 1400. An enthusiast 1333 mobo will likely hit 1500 or higher with ease.
If you are getting an e4300 then a 1333 board will inherently take you to 3000 mhz (333*9) as long as the processor can handle it. If you get a well known enthusiast board with a documented performance history (see the first thread on this board for some OC results) then you can go as far as that CPU will take you.

 

[arentol]

Oh, and to answer your question about AMD's, they don't have an FSB at all. AMD processors basically have the northbridge chip built into the CPU, so on an AMD system, instead of the layout looking like the "diagram" I posted previously, it would look more like this...

CPU+NB------Memory

That doesn't apply to 939's does it? I thought only the AM2's had the full IC and what not. ..

Also for OC purposes all that doesn't matter as far as I know. On an AM2 motherboard you just raise the "FSB" and everything else, CPU and otherwise, goes up as well, just like with an Intel.

 

[luihed]

are you saying having higher FSB in of itself can be a good thing?

its not purely a means to acheive a clock speed in the end?


the same core at 266 x 10 (2.66ghz) is NOT totally equal to one at 333 x 8 (2.66ghz)?

there would be applications or areas where the higher FSB would be better?

Higher fsb is not always better, it all depends what kind of cpu/mobo/ram combo you have, a lower fsb but using memory divider is pretty fast too.... When Im benching I use 8x and 9x multi depending on what benchmark and speed Im running....

 

[RamonGTP]

That doesn't apply to 939's does it? I thought only the AM2's had the full IC and what not. ..

Also for OC purposes all that doesn't matter as far as I know. On an AM2 motherboard you just raise the "FSB" and everything else, CPU and otherwise, goes up as well, just like with an Intel.

This applies to ALL Athlon64 CPU's as well as Sempron cpu's that run on sockets 754, 939, 940, AM2 and it will apply to the upcoming AM3 as well. The only difference between the AM2 memory controller and 939 is that AM2 is for DDR2 and 939 is a DDR controller.

And you're not raising the "FSB" since it doens't exist, you are simply raising the reference clock.

 

[Glow]

I miss OCing Bartons lol sooooooooo much easier.

 

[RamonGTP]

I miss OCing Bartons lol sooooooooo much easier.

It really isn't easier... I know A64 can be a bit confusing, but once you understand how it works, it actually gives you MUCH more flexability for overclocking, specifically when it comes to memory. With Barton's, or any Socket A CPU, you pretty much had to run the memory at a 1:1 ratio with the FSB, if you didn't, you would see a very significant performance hit due to added latencies being introduced in the northbridge if it wasn't 1:1. Since there is no FSB on the A64, the memory clock can be run at whatever speed you want (or whatever speed the memory will allow you to run).

 

[arentol]

This applies to ALL Athlon64 CPU's as well as Sempron cpu's that run on sockets 754, 939, 940, AM2 and it will apply to the upcoming AM3 as well. The only difference between the AM2 memory controller and 939 is that AM2 is for DDR2 and 939 is a DDR controller.

And you're not raising the "FSB" since it doens't exist, you are simply raising the reference clock.

I understand that you are just changing the reference clock, not the FSB since it doesn't technically exist on an AMD system (having been split into the mem controller and the HT bus). However, that has absolutely nothing to do with the point I am trying to make.

My point is that you can TREAT FSB and Reference Clock just about the same....

For instance, if you have been using Intel systems for the last 4 years and you get a new Intel system you can basically just treat the Reference Clock setting like you would have treater the FSB on an Intel, and vice versa if you have been running AMD for the last 4 years and you switch to Intel.

This is especially the case these days since the PCI and PCI-E buses are usually locked down by default, ensuring you aren't going to fry your videocard with a major FSB/Ref Clock overclock...

On an Intel if I change my FSB from 200mhz to 266mhz then the result is:
CPU runs 33% faster.
RAM runs 33% faster.
FSB runs 33% faster, which can cause lockups if my Northbridge and Southbridge can't handle it.
Basically everything else is locked down and doesn't change speed.

Meanwhile on an AMD if I change my Reference Clock from 200mhz to 266mhz then the result is:
CPU runs 33% faster.
RAM runs 33% faster.
HyperTransport Bus runs 33% faster, which, since it is already multiplied, causes lock ups. I cut it back to 3x (which is equal to the un-overclocked 4x setting) and everything works fine.
Basically everything else is locked down and doesn't change speed.

So from an OC standpoint the effective action and the effective result are the same, the only difference is the term used and some minor architectural differences in terms of what is actually happening.