misreading how fsb on the A64 and Core2Duo works?

[poohbear]

hey all, im overclocking a new C2D rig after spending 2 years on an overclocked AMD opteron rig w/ its hyper-transport "fsb" and im a bit confused as to how fsb works on the C2D.

i have an e2160 (800hmz) C2D running @ 250x9. now, quad pumped that should be 1000mhz, which is beyond the specs of this cpu, right? If its rated to run @ 800mhz, how can it handle 250x9 quad pumped? Even more confusing for me is that the ddr2 800 ram i have is running @ 1:1 (fsb:ram ratio) @ 250x9, which would mean its running @ 1000mhz, and that'd be a a huge overclock for that ram, am i right? am i misinterpreting the numbers here? on my opteron w/ ddr ram 250x9 for ram @ 1:1 would be unimaginable without some massive vdimm on expensive BH5 chips.

on an A64 200x10 would give u a HTT of 2000mhz, anything way beyond that would crash the comp w marginal gains. Is'nt the e2160 200x 4 to make 800mhz? and if i continue to overclock to 300x9 that would be 300 x4 (quad pumped) giving me 1200mhz for the fsb, wayyy beyond specs right?

i know im misunderstanding something cause my gigabyte P35C-DS3R has no options for me to reduce the ram divider beneath the fsb, only to increase it! lol there i was trying to ensure my ram is within the fsb specs, but the crazy mobo only offered to push them wayyy beyond specs!

can someone please explain what im misinterpreting? thanks in advance.

 

[dv8silencer]

FSB (unquadpumped) * 2 = DDR2 required
For you (from what you said), 250 * 2 = 500 mhz DDR2 required and you have 800 mhz, thus you have to run it at a 8/5 x FSB in order to get 800 mhz
Also, I don't know if you can run memory slower than FSB, which is why you can't reduce the ram divider to less than 1:1.

You should read the third thread for some basics?

Based on that thread, it seems that the stock e2160 should be 200 mhz * 9 = 1800 mhz
where 9 is the multiplier and 200 mhz is the unquadpumped FSB.
You seem to have overclocked the FSB to 250 mhz thus the CPU is overclocked to 250 * 9 = 2250 mhz.

For memory, like I said, the minimum (to run 1:1 ratio) frequency for DDR2 is 2 * FSB, which in your overclocked state is 250 * 2 = 500 mhz. Since you have memory that is rated for 800 mhz stock, you have to run it 8/5 x FSB in order to get it to 800 mhz.

 

[myocardia]

You're misinterpreting two things here. The first is that you're using an AMD Athlon 64. You aren't, you're using an Intel processor, and that processor does not have an on-die memory controller, like the Athlon 64's have. What speed you run your FSB has absolutely nothing to do with your processor, except for the fact that when you raise the FSB, you raise the processor's speed. This is why you have to raise your northbridge chipset's voltage to run at extremely high FSB's, and not your processor's Vcore (assuming you've lowered your processor's multiplier, of couse). The FSB you run is independent of the processor, except for the processor's overall speed.

The RAM isn't as easy to explain. Remember how, with your DDR (original, not DDR2) RAM, it ran at the same speed as your HTT bus, even though it's DDR rating was twice what it's actual speed was? Well, you aren't using DDR anymore, you're now using DDR2, which transfers data four times per clock cycle, instead of twice, like DDR did. Since you're running at an FSB of 250 Mhz, your RAM is actually running @ 125 Mhz, if you are running it 1:1, which gives you a DDR2 rating of 500 Mhz. BTW, DDR3 doesn't actually run any faster at all than DDR or DDR2. It just transfers data 8 times per clock cycle, which is twice as fast as DDR2, and four times as fast as DDR. If that wasn't clear enough for you, let me know, and I can explain further.

 

[poohbear]

ah i see, so ddr2 runs @ 4 times per clock cycle not 2 (like ddr)!!! that's what was messing me up!! lol thanks a bunch myocardia! ddr2 speeds at stock speeds would be 200/2 x 4? so, 250/2 x4=500mhz yea? i'll use THAT formula instead of just multiplying FSB x2 (as i mistakenly thought remains the same like ddr1) to get its rating. that explains the ram then!

now, im still a bit confused about the C2D's fsb rating of 800mhz (1333mhz for the newer ones). im not talking about the speed of the cpu, i already knew that was fsb x multiplier. but how is the 800mhz rating determined? is it the default of 200mhz fsb x 4? (quad pumped) furthermore, if i push up the fsb doesnt that take the fsb out of spec and does that mess up stability or it doesnt matter?

thanks in advance for the explanation folks! i've read plenty on the new chip but i guess lotsa folks assumed people knew how the fsb rating was determined but i started w/ a64s and they dont do fsbs & HTTs the same way.

cheers.

 

[myocardia]

Originally posted by: poohbear
ah i see, so ddr2 runs @ 4 times per clock cycle not 2 (like ddr)!!! that's what was messing me up!! lol thanks a bunch myocardia! ddr2 speeds at stock speeds would be 200/2 x 4? so, 250/2 x4=500mhz yea? i'll use THAT formula instead of just multiplying FSB x2 (as i mistakenly thought remains the same like ddr1) to get its rating. that explains the ram then!

Actually, the easieast way to figure your RAM speed is the way dv8silencer described. But use whatever way keeps you from getting confused about it.

now, im still a bit confused about the C2D's fsb rating of 800mhz (1333mhz for the newer ones). im not talking about the speed of the cpu, i already knew that was fsb x multiplier. but how is the 800mhz rating determined? is it the default of 200mhz fsb x 4? (quad pumped) furthermore, if i push up the fsb doesnt that take the fsb out of spec and does that mess up stability or it doesnt matter?

Yeah, the Intel FSB is also DDR2, so just like your DDR2 RAM, it acually transfers data four times per clock cycle. So technically, yours is only running @ 250 Mhz, but it's transfering the same amount of data as if it were running at 1000 Mhz.

 

[poohbear]

Originally posted by: myocardia


Yeah, the Intel FSB is also DDR2, so just like your DDR2 RAM, it acually transfers data four times per clock cycle. So technically, yours is only running @ 250 Mhz, but it's transfering the same amount of data as if it were running at 1000 Mhz.

and that wont make it crash if its running 200mhz over spec? what if i opt to drop my multiplier t0 8 and use a fsb of 350, would that make it 350x4= 1400mhz? it'd be way beyond spec (800mhz) and unstable no?

 

[myocardia]

Originally posted by: poohbear
and that wont make it crash if its running 200mhz over spec? what if i opt to drop my multiplier t0 8 and use a fsb of 350, would that make it 350x4= 1400mhz? it'd be way beyond spec (800mhz) and unstable no?

No, it's not "beyond spec". Your processor has no factory approved FSB. Well, technically it does, but not in the way you're thinking. Remember, the FSB has nothing to do with the processor, only the motherboard's northbridge chipset. And your motherboard is spec'd to run @ 333 FSB, so running it at 350 is about as much of an overclock to it as running your E2160 @ 1.9 Ghz would be.

edit: BTW, you're using the same motherboard I have in my Q6600 system. Not many of us have them, I've noticed.

 

[poohbear]

so what do they mean when they say e2160 (800hmz) and say a new 45nm C2D that is speced @ 1333mhz? what do those numbers mean? is that just the ram they can keep up w/? cheers and thanks again for any further insight into this.

i have my e2160 running @ 2.8ghz on stock cooler and stock voltage, these things are demons! and yea the gigabyte mobo i got is quite impressive and ridiculously easy to overclock w/. i guess overclocking has gone mainstream these days eh?

 

[Aluvus]

Originally posted by: poohbear
so what do they mean when they say e2160 (800hmz) and say a new 45nm C2D that is speced @ 1333mhz? what do those numbers mean? is that just the ram they can keep up w/? cheers and thanks again for any further insight into this.

Those numbers define the speed of the Front Side Bus. The FSB is the link between the processor and the Northbridge (Intel uses the term "Memory Controller Hub" rather than "Northbridge"). One of the major jobs of the Northbridge is to control the memory, except in the case of processors with an integrated memory controller, such as the Athlon 64.

Indirectly, this also affects the speed that memory can be operated at. Memory can be operated faster than would "match" the FSB, in which case it is simply run asynchronously. There is a performance penalty associated with doing so (which, depending on a variety of details, may mean running faster than the FSB is worse than running at the same speed). There is also a performance penalty for running the memory at a speed below what would "match" the FSB.

Originally posted by: myocardia
No, it's not "beyond spec". Your processor has no factory approved FSB. Well, technically it does, but not in the way you're thinking.

I'm not sure what you're getting at. From my reading, poohbear has it pretty much right. Processors are tested by the manufacturer to be able to handle a particular FSB speed, and that is the spec they are rated at. Processors have hardware that has to connect to the FSB, and that hardware is not infinitely fast. In practice they can often be run above that spec, but that is still "above spec".

Remember, the FSB has nothing to do with the processor, only the motherboard's northbridge chipset.

The FSB is a link between the Northbridge and the processor. Certainly the FSB speed has an impact on the processor's operation, as well as its performance. This impact is less extreme than it can be for the Northbridge, because it can be reduced by altering the multiplier. But it is still true that if the FSB is sped up, the processor has to keep up.

Originally posted by: myocardia
Yeah, the Intel FSB is also DDR2, so just like your DDR2 RAM, it acually transfers data four times per clock cycle. So technically, yours is only running @ 250 Mhz, but it's transfering the same amount of data as if it were running at 1000 Mhz.

Intel's FSB is QDR, or quad data rate. DDR2 is a memory standard that also (AFAIK) uses a QDR link. This is one of those "not all rectangles are squares" things.

 

[myocardia]

Originally posted by: Aluvus

Originally posted by: myocardia
No, it's not "beyond spec". Your processor has no factory approved FSB. Well, technically it does, but not in the way you're thinking.

I'm not sure what you're getting at. From my reading, poohbear has it pretty much right. Processors are tested by the manufacturer to be able to handle a particular FSB speed, and that is the spec they are rated at. Processors have hardware that has to connect to the FSB, and that hardware is not infinitely fast. In practice they can often be run above that spec, but that is still "above spec".

Remember, the FSB has nothing to do with the processor, only the motherboard's northbridge chipset.

The FSB is a link between the Northbridge and the processor. Certainly the FSB speed has an impact on the processor's operation, as well as its performance. This impact is less extreme than it can be for the Northbridge, because it can be reduced by altering the multiplier. But it is still true that if the FSB is sped up, the processor has to keep up.

Originally posted by: myocardia
Yeah, the Intel FSB is also DDR2, so just like your DDR2 RAM, it acually transfers data four times per clock cycle. So technically, yours is only running @ 250 Mhz, but it's transfering the same amount of data as if it were running at 1000 Mhz.

Intel's FSB is QDR, or quad data rate. DDR2 is a memory standard that also (AFAIK) uses a QDR link. This is one of those "not all rectangles are squares" things.

You must never have heard the phrase "know your target audience". Or do you think I should have explained to him how even while running his RAM at the same frequency and timings that I'm running mine, his data throughput will be lower? Hey, what if I explained skew to him? Feel free to explain Hi-K metal gates to him, using technical terms, if you'd like. I'll hide and watch.

 

[betasub]

Originally posted by: Aluvus
Processors are tested by the manufacturer to be able to handle a particular FSB speed, and that is the spec they are rated at. Processors have hardware that has to connect to the FSB, and that hardware is not infinitely fast. In practice they can often be run above that spec, but that is still "above spec".

Is that correct - Intel processors are tested for FSB capability? Or are they just tested/rated for clock speed, and then given default multiplier/FSB for a target maket)?

After all, it's the motherboard (BIOS) that sets multiplier and FSB values and determines the speed the CPU runs. As far as the CPU alone is concerned, is there any difference between 10x266 and 8x333 if both require it to run at 2.66GHz?

 

[Aluvus]

Originally posted by: myocardia

You must never have heard the phrase "know your target audience". Or do you think I should have explained to him how even while running his RAM at the same frequency and timings that I'm running mine, his data throughput will be lower? Hey, what if I explained skew to him? Feel free to explain Hi-K metal gates to him, using technical terms, if you'd like. I'll hide and watch.

what

There is a difference between a simplification that may mislead but makes things easier to understand, and a simplification that misleads for no beneficial reason.

Originally posted by: betasub

Is that correct - Intel processors are tested for FSB capability? Or are they just tested/rated for clock speed, and then given default multiplier/FSB for a target maket)?

After all, it's the motherboard (BIOS) that sets multiplier and FSB values and determines the speed the CPU runs. As far as the CPU alone is concerned, is there any difference between 10x266 and 8x333 if both require it to run at 2.66GHz?

Not all of the hardware in a CPU operates at the same speed. The execution core in your example would be operating at 2.66 GHz, but the interface components that talk to the FSB would not. Their operating speed is governed by the speed of the FSB, not the speed of the CPU's core clock. They don't care (and I suspect, don't know) what the multiplier is set at. They only need to care how fast data is moving across the FSB, which is not affected by the multiplier. After all, the performance of the bus is going to be limited by the device on the other end of the FSB (the Northbridge), which has no idea what the CPU's multiplier is set at.

 

[poohbear]

well thanks for all the replies and elaborations folks, its very much appreciated. I understand some terminology has to be dumbed down to keep things simple and pertinent as far as layman are concerned, so no worries on that end.

alot of it is making more sense now when you take over clocking out of the equation and just try to understand how the components are meant to run: an fsb of 800, 1066, 1333 means the cpu can handle that in accordance w/ the ram & northbridge, and overclocking beyond such specs doesnt guarantee they'll work beyond these fsbs. thanks again for the clarifications.

betasub, in my experience there is no diff between 10x266 and 8x333 as they're 2 different ways to reach the same speed (cpu speed = fsb x multiplier), assuming the NB can handle a fsb of 333 etc. just my 2 cents

 

[ShockwaveVT]

Lets do a comparison and perhaps that will clarify it for you

E2160 9x multiplier, default FSB is 800MHz (200*4) ... CPU clock is 9x200 = 1.8 GHz

E8400 9x multiplier, default FSB is 1333MHz (333*4) CPU clock is 9x333 = 3.0 GHz

Since regular Core2Duos and Core2Quads have a locked multiplier (i.e. you can't raise the multiplier on your E2160 above the stock 9x) you must increase the FSB to overclock the CPU. For example, my E6300 at default is 7x 266(1066) = 1.86 GHz. I have it overclocked to 7x 429(1716) = 3.0 GHz.

Overclocking your E2160 to 3.0 GHz is as simple as increasing your FSB from 200(800) to 333(1333). Typically you also need to increase your CPU's vCore a bit as well to compensate. Yes this puts you "over spec" for your particular CPU, but hey that is why it is called Overclocking! (Note there are limits on vCore, 1.5v max for 65nm C2D's and about 1.35v max for the 45nm C2D's).

Most motherboards can handle 333 MHz (1333) FSB without breaking a sweat, as that speed is the default for many of the newer Core2Duos. Alot of boards can handle 400 MHz (1600) as well, since some Intel CPUs are starting to run that as default. And of course most enthusiast boards can handle FSB's well over that, up into the 500's. You may find that plain jane motherboards don't support overclocking at all, and in those cases you can usually fool them by doing a BSEL pin-mod to your CPU so that it identifies its default FSB as 266(1066) or 333(1333) instead of 200(800)

Since RAM speed is directly linked to FSB, you must make sure that your RAM is fast enough to handle the higher than default FSB when overclocking. DDR2 800 RAM is rated for 400 MHz (1600) FSB at a 1:1 ratio. Since most C2D's have a lower default FSB than 400 MHz (1600) there is some built-in headroom for increasing your FSB and therefore overclocking your CPU. Beyond that you start overclocking your RAM, which many people are able to do successfully. (my DDR2 800 is running fine at 429 FSB, meaning the RAM is operating as if it were DDR2 858)

 

[ShockwaveVT]

Originally posted by: poohbear
betasub, in my experience there is no diff between 10x266 and 8x333 as they're 2 different ways to reach the same speed (cpu speed = fsb x multiplier), assuming the NB can handle a fsb of 333 etc. just my 2 cents

Actually there is some gains from running a higher FSB and lower multiplier.

8x 333(1333) = 3.0 GHz will slightly out perform 10x266(1066) since memory bandwidth scales up with FSB. (in theory).

Thats why Intel keeps increasing the default FSB on their new chips, first 266(1066) then 333(1333) and now 400(1600) and sets it lower, at 200(800) for their budget line.

 

[poohbear]

Originally posted by: ShockwaveVT

Originally posted by: poohbear
betasub, in my experience there is no diff between 10x266 and 8x333 as they're 2 different ways to reach the same speed (cpu speed = fsb x multiplier), assuming the NB can handle a fsb of 333 etc. just my 2 cents

Actually there is some gains from running a higher FSB and lower multiplier.

8x 333(1333) = 3.0 GHz will slightly out perform 10x266(1066) since memory bandwidth scales up with FSB. (in theory).

on a 1:1 divider, but you can always use a ram divider to increase the rated speed of the ram to correspond w/ the fsb. i.e. since 10x266 @ 1:1 is far beneath ddr800's 400mhz, you can use a divider to bump it up closer to 400, right? however, i always wondered how much of a performance penalty this really incurs? is running ddr800 ram @ ddr 650/700 that much of a penalty in real world apps? i doubt it.

infact, how much does the speed of the system's fsb (800mhz, 1066mhz, 1333mhz etc) effect overall system performance? is it tangible or significant? i'd be curious to see the difference between 2 cpus running @ the same speed but w/ one @800mhz fsb and the other @ 1333mhz. is the fsb saturated @ 800mhz, or does 800mhz already provide enough bandwidth between the NB & most C2D cpus and 1333 is overkill?

 

[jesterb84]

Originally posted by: poohbear

Originally posted by: ShockwaveVT

Originally posted by: poohbear
betasub, in my experience there is no diff between 10x266 and 8x333 as they're 2 different ways to reach the same speed (cpu speed = fsb x multiplier), assuming the NB can handle a fsb of 333 etc. just my 2 cents

Actually there is some gains from running a higher FSB and lower multiplier.

8x 333(1333) = 3.0 GHz will slightly out perform 10x266(1066) since memory bandwidth scales up with FSB. (in theory).

on a 1:1 divider, but you can always use a ram divider to increase the rated speed of the ram to correspond w/ the fsb. i.e. since 10x266 @ 1:1 is far beneath ddr800's 400mhz, you can use a divider to bump it up closer to 400, right? however, i always wondered how much of a performance penalty this really incurs? is running ddr800 ram @ ddr 650/700 that much of a penalty in real world apps? i doubt it.

infact, how much does the speed of the system's fsb (800mhz, 1066mhz, 1333mhz etc) effect overall system performance? is it tangible or significant? i'd be curious to see the difference between 2 cpus running @ the same speed but w/ one @800mhz fsb and the other @ 1333mhz. is the fsb saturated @ 800mhz, or does 800mhz already provide enough bandwidth between the NB & most C2D cpus and 1333 is overkill?

I've done some searching around on this myself, and I think the general consensus on the net is that increasing the FSB yields minor performance gains in synthetic benchmarks, but yields negligible impact on real-world performance. You can benchmark yourself by lowering the multiplier and increasing the FSB.

But remember, 9x333=3GHz and 8x375=3GHz stresses different components. A higher multiplier places more stress on the CPU while a higher FSB places more stress on the NB. Since I have a cheap E2160 CPU, I'd sooner prefer zapping my $70 CPU...

 

[poohbear]

yep fsb increases from 800mhz to 1066 prolly show tangible performance increase, i figured as much since this was also true of the A64s and even the old athlons. kinda like the ddr timings fiasco when the ram companies were always advertising tight timings for their premium ram like 2-3-2-5 and charged premium prices. then when ddr2 emerged they were like "forget ram timings, they were never important anyway, 5-5-5-15 is perfectly fine". sure enough the bencmarks proved that they were indeed very negligible in terms of a performance loss. going to looser timings but higher bandwidth provided the same performance.

but how would a default 9x multiplier put too much stress on an e2160? its meant to run on the 9x multiplier right?

 

[jesterb84]

Originally posted by: poohbear
but how would a default 9x multiplier put too much stress on an e2160? its meant to run on the 9x multiplier right?

No, I just used the above example as a comparison. Of course the 9x on my E2160 is the default one, I was just comparing between a lower and a higher multiplier and how they affected the CPU/NB.

 

[poohbear]

oh i see. well thanks for all the info folks, really clarified the way everything works on the newest cpus and mobos. cheers.