Hi,

I was just wondering. On a motherboard like Abit A17, which supports the quad pumped 800MHz P4 chips, what takes priority when overclocking? The motherboard clock speed (which is 133 I think) the motherboard chipset FSB which is 266 or the memory/cpu FSB which I guess should be DuelDDR400 and 800MHz FSB CPU?

Thanks

Mark

No, the system bus (what you referred to as "motherboard clock speed") is 200mhz...the FSB is 800mhz (200mhz x 4).

When you overclock, you increase the system bus...the FSB automatically increases along with it. On some systems, you can lock the memory speed (200, 266, 333, etc), but still increase the system bus...but running the system bus & memory bus at different speeds is not recommended.

So what effect does overclocking the system bus (and therefore the system FSB) have on the CPU external bus? Does it change?

thanks

Mark

I believe external bus & system bus are different names for the same thing? Internal bus is the same as the CPU's overall clock speed. If you increase the external bus, the internal bus will increase as well...unless you lower the multiplier to compensate.

In other words,

10.0 (multiplier) x 100mhz (external bus) = 1000mhz (internal bus)...if you change the external bus to 133mhz, the internal bus would run at 1333mhz. But if you want to run a 1000mhz CPU with an external bus of 133mhz, you'd have to lower the multiplier to 7.5

should have directed you to the post just 2 below your's titled, "Help with FSB"

Jam; everything you sais is valid.

<<but running the system bus & memory bus at different speeds is not recommended.>>
This applies more to AMD Athlon. P4 800 mhz with dual memory band has much less impact. Using even down to 3:2 ratio or PC2700 still provides excellent bandwidth at better than 4800mb/s.

Hi,

I know I am getting confused about something, I can't pinpoint exactly what it is yet. The post below has helped though.

Could I run this by you guys:

FSB = Front Side Bus

Motherboard has:

A System Bus (for instance 133)
A System Front Side Bus (which is related to the system bus but is new terminology for DDR) (for a 133 system bus this would be 266)

A CPU has:
A CPU FSB (which is not related to any bus on the motherboard but is used to calculate the overall CPU Clock speed, like 3.2GHz P4)

A Clock Speed (this is calculated as the CPU FSB x CPU Multiplier)

Memory has:
A memory bus (for instance, if it's 133, the theoretical FSB is 266)

So, for overclocking purposes, when you increase *the FSB* instead of the CPU multiplier are you

a) increasing the system FSB( and also other buses like PCI/AGP), which in turn affects the CPU FSB, and therefore the CPU clock speed

or

b) increasing just the CPU FSB and increasing the CPU clock speed and not affecting the system FSB

or

c) The system FSB and the CPU FSB are one and the same thing?

Sorry if this makes no sense but I'm just typing the confusion in my head!!

Thanks for your help

Mark

Say you have a 3.2ghz p4, 800fsb.

The doorway of the cpu is 64 bits wide. It can let in 64x800,000,000 bits in a second. The path entering the cpu is called the front side bus, the fsb, the cpu fsb. They all mean the same thing.

At the other end of that path is the motherboard northbridge. A motherboard is based on a chipset, which has a northbridge and southbridge (leaving out new technology here). The path between the northbridge and the cpu is the fsb.

That fsb is based on a system clock, or system bus, or motherboard clock, or motherboard bus, or cpu frequency. They all can be used to describe this one clock that is running at 200mhz in this case. Cpu frequency is somewhat misleading here but it is used in the bios sometimes to refer to this clock.

For this cpu the fsb is always 4 times the system clock.

The path between the northbridge and the memory is called the memory bus. If you have DDR memory this bus runs at twice the memory clock, also known as the dram frequency or the memory frequency.

Not to be critical, but there are a number of statements in the post above that are incorrect:

"Motherboard has:

A System Bus (for instance 133)
A System Front Side Bus (which is related to the system bus but is new terminology for DDR)"

DDR is a memory term and does not fit here. There is only one fsb and we might call it the system fsb or mobo fsb or cpu fsb but its still the same thing. The fact that on an amd system the fsb is double the system clock is due to double-pumping, not DDR. In practice they are the same thing but DDR does not fit in a discussion of the relationship of a cpu and a mobo, especially an intel mobo where the bus is quad-pumped.

"A CPU has:
A CPU FSB (which is not related to any bus on the motherboard..."

The cpu fsb is exactly the same thing as the motherboard fsb.

"A Clock Speed (this is calculated as the CPU FSB x CPU Multiplier"

The cpu clock speed is the multiplier x the system bus. The fsb is not a part of this convention. The clock speed of the cpu in question is written as 16x200 = 3.2ghz. The multi is 16, the system clock is 200mhz, and the cpu clock speed is 3.2g. The fsb, which is 800mhz, is not mentioned.

"Memory has:
A memory bus (for instance, if it's 133, the theoretical FSB is 266)"

If the dram clock is 133mhz then the memory rate (if you have DDR memory) is 266mhz. This may or not be related to the fsb. Fsb does not fit in a discussion of the relationship between the northbridge and the memory.

to answer this question:

"So, for overclocking purposes, when you increase *the FSB* instead of the CPU multiplier are you

a) increasing the system FSB( and also other buses like PCI/AGP), which in turn affects the CPU FSB, and therefore the CPU clock speed

or

b) increasing just the CPU FSB and increasing the CPU clock speed and not affecting the system FSB

or

c) The system FSB and the CPU FSB are one and the same thing?"

a) is correct, b) is impossible, c) is also correct.

lazyman,

I'm admittedly not an Intel guy, but I've read numerous times that running the RAM & CPU "in sync" always gives the best performance. I've never seen a disclaimer that said, "except when running a P4".

Are you saying that since the bandwidths are so "ridiculously" high anyway, that it really doesn't matter?

I'll answer that one jam, and you are right. The 800mhz fsb p4 has enough bandwidth to make it less of a weak link than is the case for an amd fsb. 4 years ago Intel and Amd had the same fsb speeds, but then amd stuck themselves to someone else's bus technology (DEC ev6), so the bus speed improvements over time were simply the result of the manufacturing/technology improvements and did not incorporate the design improvements and optimizations that Intel achieved. The result is that the P4 ended up less constrained by its fsb than did the amd xp, making it more tolerant of sub-optimum memory bandwidth. Note that the highest speed amd air-cooled mobo maxes out at about 250mhz double pumped, or a rate of 500mhz, but the fastest intel mobo maxes out at about 300mhz quad-pumped, or a rate of 1200mhz. Its no coincidence that intel makes the chipset.

So take the common maximum air-cooled oc rig you can get from each, expressed using the "real" multiplier, which is the cpu ghz divided by the fsb (as opposed to the system clock). In these terms you would have a 3.6ghz p4 running at 3x1200, and a 2.5ghz amd xp running at 5x500. Its that much higher amd multiplier (5 vs 3) which indicates the bandwidth constraint, and the intolerance for less-than-in-syncness

This is why amd's multiplier locking was such a significant downer. The unlocked multi allowed you to dial into that last mhz of fsb your mobo could handle.

2001 article about the amd xp bus

http://www.extremetech.com/article2/0,3973,1148757,00.asp

It is also why the new 64-bit cpus from amd are such a leap forward. Amd has gone from having the most constricted bandwidth to the most wide-open, thanks to the on-die memory controller. They went from a lame bus to no bus at all.

Your point about in sync being the best way to run is indeed true, but far more for amd than intel, and probably should have the intel disclaimer you mention.

Johnoh and Jam - Thankyou!!

Johnoh, you have definetly cleared up the terminology problems for me and I think I am on the way to understanding the hardware architecture now. I'm still unsure about how increasing the system clock affects the memory bus (if it does) but apart from that I think I have all the other basic principles.

I have 'knocked together' a diagram based on what you have said in this post, please could you let me know if I have deciphered your explanation correctly.

The diagram is here on my PC:
http://80.6.33.110:8080/hardware.htm

Is this statement true?
FSB = system clock + any performance enhancers like 'double pumping or quad pumping' depending on the CPU chip.

Many Thanks, you've both been very helpful... it was the terminology that has been stopping me make any progress in fully understanding hardware architecture, and now when I read about these things in my magazine (customPC - fantastic by the way) I will have a clearer understanding.

Mark

Asus A7A266
Athlon XP1800
512MB PC2100 DDR266 RAM

Good explanation, Johnoh, can't do it better myself.

The A64 finally beats Intel in max bandwidth. The only thing holding me back from getting one is the uncertainty of which socket AMD will stay; 940 or 939.

Mark your diagram is right on, though "Internal Bus" is perhaps best left off the chart since its not a commonly used term and there are a lot of things happening in the cpu at 3.2g.

There is a back-side bus and it does happen to run at 3.2ghz as well. The bsb is the path (entirely within the cpu) between the L1 cache and the L2 cache. Back in the day when L2 cache was off-die (not on the cpu chip), the bsb would run at some speed lower than the cpu speed, like 1/2 or 1/3. But since the bsb has gone on-die it runs at the same speed as the cpu. Since the bsb will remain on the cpu for the foreseeable future, calling out its speed isn't usually done anymore. Here are two pages that give a lay-overview of what happens in a cpu:

http://www.sysopt.com/articles/p4/index3.html

http://www.apple.com/g5processor/executioncore.html

"I'm still unsure about how increasing the system clock affects the memory bus"

On most new mobos it does not, as they are set independantly.

In particular, on your memory stick is a tiny chip

http://www.pc100.com/dimmspd.jpg

called the SPD (serial presence detect), which is read by the mobo bios immediately on startup and if the mobo bios has "default" or "auto" set for the dram clock, the speed which is hard-coded into the spd chip is reported to the bios and the bios sets itself to run the memory at that speed.

If you want to manually adjust the dram clock, some mobos will only allow you to do it in sync with the fsb, or in proportion to the fsb. For instance, I have an amd mobo which runs at a system clock of 133mhz and if you put in pc2700 memory (2x166mhz) and adjust the system clock to 140, the dram clock autmatically goes to 175. The reason for things like this is that the mobo maker can save a few bucks by having only one clock generator on the board, and every other speed uses that as their basis, multiplying it by 5/4 or 3/4 or whatever the case may be.

"FSB = system clock + any performance enhancers like 'double pumping or quad pumping' depending on the CPU chip"

That statement is true. A shorter version of it is:

fsb = 2 x system clock for amd xp
fsb = 4 x system clock for intel
fsb = does not exist on amd 64-bit, as hypertransport replaces the fsb. The cpu-to-memory-controller hypertransport path is within the cpu die since the memory controller (normally on the northbridge) is on the cpu die. And unlike an fsb which goes in both directions resulting in bits waiting around at the ends like cars using a one-lane bridge, hypertransport uses two lanes, one dedicated to coming in and the other dedicated to going out. This greatly reduces latency (latency=waiting) even without changing the speed.

Things go fast enough nowadays that performance improvements often come from eliminating wait-states, not increasing maxed-out speed states.

This two-lane vs one-lane concept is also the next step in the pci/agp world, as pci-e has lanes in both directions but pci/agp shares the same physical path for coming and going. pci-e also has separate paths per device as opposed to several devices sharing one path (the pci bus). Open a file on your CD drive, your hard drive, play a song and download something through your ethernet adapter and it all goes through the one pci bus. With pci-e they'de each have their own path. Less waiting.

Thanks for above posts. So helpful in clarifying these issues for a non-techy like me. Even so, I'm still not sure about the following and wonder if someone can help: when the blurb for the Abit IC-7 mbo says "800 fsb (400 fsb *not supported*)" what is the implication of the 400 fsb not supported? What are they telling me I can not do/add to that mbo? Also what memory would folks recommend for this board with a P4 3.2 800fsb (Prescott)? And finally, can someone explain what difference the level 1 cache makes. Is it worth the extra bucks to go for 1mb instead of the standard 512kb? Would be grateful for any replies.