Help: CPU & NorthBridge Communication Core 2 Duo

[GreenJelly]

A) Is the communication rate between the CPU to the Northbridge 4xFSB?

If (A) is true then
B) Increasing the FSB puts more strain (or maybe overclocks) the Northbridge?

If (B) is true then
C) How hardy is the 965 and 975 Northbridge? How far have people gone with the FSB on these chipsets?

 

[Guest]

The CPu runs at 4X the Internal FSB on conroe that gives 266X4.

On the motherboard it runs in dual channel at 2X 266.

Increasing the FSB does put some strain on the northbridge but there is some overhead and the 975 does suppport 1333FSB so its no problem.

The P5b/P5B-Dlx can attain 490-510 with proper cooling.
On a 975 the limit is around 400 and on the upcoming Ati RD600(only DFI will have a board apparently), is stated to hit 480+. All this X4 and then X multiplier give you the clock speed of the cpu.

 

[GreenJelly]

Too Answer my own question... The higher the FSB, the more likely it will cause failures... Most MoBo's and Most Intel Chips are able to handle around 400mhz max...

The P5B basic is limitted in that it has a non-copper heatsink and probably will fry at higher FSB's... But there is little data available to the temps these Northbridge chips run at underload on certain MoBos...

Swiftech MCX159 should work, but I have yet to test it out, and will have to wait till abit to

A) get my machine back
B) to make sure its stable for a few weeks.

I will be going down to frozen cpu in the near future to look at this Cooler first hand... and to make sure it fits BEFORE I buy it...

 

[1Tanker]

Too Answer my own question... The higher the FSB, the more likely it will cause failures... Most MoBo's and Most Intel Chips are able to handle around 400mhz max...

The P5B basic is limitted in that it has a non-copper heatsink and probably will fry at higher FSB's... But there is little data available to the temps these Northbridge chips run at underload on certain MoBos...

Swiftech MCX159 should work, but I have yet to test it out, and will have to wait till abit to

A) get my machine back
B) to make sure its stable for a few weeks.

I will be going down to frozen cpu in the near future to look at this Cooler first hand... and to make sure it fits BEFORE I buy it...

Why ask the question then, if you already know the answer? :roll:

 

[GreenJelly]

I asked the question September 11th

I answered the question September 16th

 

[1Tanker]

I asked the question September 11th

I answered the question September 16th

Sorry, didn't notice that.

 

[godman]

This will change when 965P rev.2 hit the market at October which will have a 600Mhz FSB wall (from reports so far).

8O wow!

right now i need to ask a question, Wusy or anyone else, if you got the fsb to 600mhz (2400mhz QDR) then isnt this faster than AMD's HTT, why cant intel therefore raise the fsb at stock to for example 500mhz (2GHZ QDR, therefore there would be no need for CSI)? or is this all to do with the quality of the components delivering clean signals etc..? :?

 

[GreenJelly]

AMD has their "NorthBridge"... well actually its the memory controller... Built Onto the CPU...

Many people feel this is a superior design, others say its opposite..

As for your question, because of the above... Its like compairing Apples to Oranges.

 

[godman]

yeah i know that the K8 has a on die memory controller hub but is the only reason for CSI because servers, for example woodcrest on a 4 dual core processor system (8 cores) (i read somewhere) doesn't work (or not well anyway) becuse of bandwidth issues etc...
therefore it woulsnt be worth having a 4processor woodcrest due to memory bandwidth problems, however HTT means that amd can have 8 core servers due to HTTT links which intel's (old) fsb doesnt have?

in conclusion i mean is this the only reason for intel developing their own "HTT" (CSI), more links? or is it because of bus speeds? :?

 

[GreenJelly]

I know that the intel core 2 duo processor can handle 4 instructions at one time... and that they share L2 Cache in a very advanced manor.

As far as the 400+ FSB on the DH and Deluxe... I would be amazed at this result... but hey... good luck

 

[sithscout80]

The speed of the FSB isn't the limiting factor in the move to CSI. As can be seen with benchmarks, the FSB isn't close to being saturated, (it can use the whole FSB bandwidth in short bursts). The major difference between HTT and the FSB is the latency time. The FSB has a much higher latency because it has to make multiple hops instead of the directness of HTT.

 

[ltcommander_data]

This will change when 965P rev.2 hit the market at October which will have a 600Mhz FSB wall (from reports so far).

8O wow!

right now i need to ask a question, Wusy or anyone else, if you got the fsb to 600mhz (2400mhz QDR) then isnt this faster than AMD's HTT, why cant intel therefore raise the fsb at stock to for example 500mhz (2GHZ QDR, therefore there would be no need for CSI)? or is this all to do with the quality of the components delivering clean signals etc..? :?
Finding a few CPUs and a few motherboards that can get to a 2.4GHz FSB may not be that hard. However, getting an entire product line of CPUs and motherboards from low-end to high-end to operate at those speeds is basically impossible with current implementations.

Besides, by the time you get the FSB above 1333MHz, you need to increase CPU voltage, motherboard voltage, memory voltage, all of which puts you out of spec. People who get high overclocks are using doing voltage modding, use special cooling, and have choice components all of which are not going to filter into the mainstream.

If Intel really tried (better binning, tweaked process, etc.), they should be able to get dual die chips like Kentsfield, Cloverton, and Tigerton operating on a stock 1333MHz FSB with standard cooling and voltages. A 1600MHz FSB stock is about the upper limit on a single die chip, but will probably need 45nm to get decent yields.

 

[godman]

thanks so the main reason is latency and comparing core 2 uarch and netburst uarch if you gave a netburst processor a 2000mhz fsb (motherboard considering) would that mean it would use up the available bandwidth due to (at stock) being strived for memory bandwidth. seeing as core doesnt make full effective use of the bandwidth and knowing as P4s strive for bandwidth the two uarchs wouldnt react the same to different fsb speeds.?

 

[sithscout80]

The reason Core 2 isn't affected by the FSB speed is because it hids the latency with the prefetchers and the large cache.
P4s excelled at deeply pipelined operations, like rendering media (as long as it predicted correctly). When the pipeline is full, the latency can be hid.

Also the reason that Core 2 doesn't use the full bandwidth of the FSB is not because it doesn't need, but because it can't. Due to the way that RAM is made and how many channels the RAM has, there is a max average memory bandwidth (I saw this demonstrated with the Stream benchmark at IDF).

 

[godman]

oh so thats the reason why ddr2 533 is better than ddr2 667 as it in sync with the fsb and because it can us the bandwidth, and by using memory multipliers/ dividers that allow the use of ddr2 667 causes latency

 

[Whizzard9992]

A) Is the communication rate between the CPU to the Northbridge 4xFSB?

If (A) is true then
B) Increasing the FSB puts more strain (or maybe overclocks) the Northbridge?

If (B) is true then
C) How hardy is the 965 and 975 Northbridge? How far have people gone with the FSB on these chipsets?

the data that goes on the FSB is a stream of data
the northbridge only takes the data it needs out of the stream
most of the data is for or from the ram and does not get processed to much by the Nbridge
the north bridge has little strain in this regard mostly you want to make sure it is running with a good divisor with the proc and ram


Correct me if I'm wrong, but the CPU talks to the Northbridge via FSB, and the Northbridge then communicates with the RAM, and sends the data back vis FSB, no?

 

[GreenJelly]

In another thread you mentioned that the memory speed is 1:1 when its at 533... and that only 533 is good for overclocking... What if you have 1066 memory? Why cant you increase the multiplier... Why do you say all of this? You just blew my mind....

 

[godman]

he means that 1066mhz ram works at a 1:2 (FSB;DRAM) ratio, meaning that it effectivly transfers 2 peices of data back to the northbridge for everty one peice of data transfered to the ram.

although core 2 cannot saturate the bandwidth (as i learned from annother post HERE) so 1066 mhz ram isnt much point in buying unless you're overclocking...in which itll be much easeier to overclock and can help you achive lower latencies...however lower latency ram also doesnt give that biga performance increase with core Uarch.

you might not be able to increaes the multiplier because at stock the motherboards are just not compatible with the fast ram, you have to overclock the fsb to be able to get the ram to work at 1066 spoeed, even then it would require a 533mhz (2132mhz QDR) fsb to get the cpu to make good enough use of the fsb and ram (the ram is therefore in sync with the fsb at that speed, although you havent overclocked the ram)

 

[GreenJelly]

I cant overclock the ram... its corsair and its said to run best at 2.2v

Of course P5B Basic doesnt support 2.2v, only 2.1... Even though all their other boards support the 2.2v (and more)

Asus sucks...

Anyways, I am lucky... The ram runs reliable at 1066, and the from all benchmarks, the 1066 5-5-5-15 runs faster then ANY other Corsair memory. Which means its the fastest memory out... Until the Dominator class comes out Then I wont be so uber... (like that maters)

I would drop the memory multiplier if I wanted to increase FSB... but right now I am not about to overclock this system...

After a few weeks, when I am confident in the machines reliability, then I will have a benchmark to compare when I OC... No real benchmark will absolutely prove reliability on a CPU... This is unfortunate... but its the way the world works...

Watercooling adds even more uncertainty... Sure you can benchmark the machine... But then you add more heat as the parts warm up... Use your 3D processor and things get warm.

Mike

 

[GreenJelly]

Life is good with good answers! I had forgotten about the access of memory from other then CPU!