FSB vs performance

[Alice in Chains]

LS,

In my earlier time (pentium 1 /K6-2) the FSB was always 1 to 1 related to the memory bus. It should be as high as possible. 66x2 (133 MHz) outperforms 50x3 (150 MHz). Since the membus speed is not fixed to FSB anymore, I wonder what the impact is.

Which settings has the best performance?

(Board = KT7, proc = tbred B)

Setting a:

FSB = 100 MHz
PCI = 33 MHz
MEM Bus= 133 MHz (HOSTCLK + PCICLK)
multplier = 22x

CPU = 2200 MHZ.

Setting B:

FSB= 116 MHz
PCI= 33
MEM Bus = 116 MHz (HOSTCLK)
multiplier = 19x

CPU= 2204 MHz.

My guess is that the membus speed is also the bottle neck, and should be as high as possible.

What do you guys think?

Thanks,

Mike.

 

[Wes Newell]

On Tue, 28 Oct 2003 16:56:04 +0100, Alice in Chains wrote:

LS,

In my earlier time (pentium 1 /K6-2) the FSB was always 1 to 1 related to the memory bus. It should be as high as possible. 66x2 (133 MHz) outperforms 50x3 (150 MHz). Since the membus speed is not fixed to FSB anymore, I wonder what the impact is.

The pentium FSB was 1-1 for data and clock. The memory bus has always been seperate and some socket 7 boards allowed different memory and FSB speeds. Many had a compatibilty jumper that locked the memory bus at 66MHz so you could change the FSB speed for the newer cpu's like Cyrix that ran at 75, 83, etc. fsb.

Which settings has the best performance?

(Board = KT7, proc = tbred B)

Setting a:

FSB = 100 MHz
PCI = 33 MHz
MEM Bus= 133 MHz (HOSTCLK + PCICLK)
multplier = 22x

CPU = 2200 MHZ.
...

How about C?

C:
FSB= 116 MHz
PCI= 33
MEM Bus = 149 MHz (HSTCLK+PCI)
multiplier = 19x

I've got PC150 ram in mine. But honestly, I don't think you'll notice the difference with either setting. A 33MHz jump in bus speeds provides only about 3-5% improvement in performance. Approx. 90% of data requests from the cpu are serviced by the internal cache.

 

[- HAL9000]

Ok, I'll jump off a cliff....

In my Kx7 motherboard the FSB clock is "fixed" to the memory bus clock. It is fixed by ratio. The common term for this is not "fixed", but "locked". Meaning, the FSB isn't fixed to one single value, rather it can take on many values, values in ratio to the FSB. The memory clock to FSB clock ratio is "locked". Perhaps you already knew these things, not sure, just thought I would explain. Part of the confusion in terminology is the translation of Asian words to English words.

Most users have the ratio set at 1:1 (eg FSB = memory) but 4:5 setting appears to be possible (eg FSB = 133, memory = 166) but useless. The excess (frequency) bandwidth of the memory is wasted at the CPU - North Bridge interface, aka FSB.

One point worth mentioning is that the Kx7 designers did not allow the memory to be lower in value then the FSB. That is, the bios options won't allow you to set a 5:4 ratio (eg FSB = 166, memory = 133). Not sure why. The effect would have been the same though - wasted bandwidth. In this case the FSB bandwidth instead of the memory bandwidth.

So in scenario A, the bottleneck has become the FSB clock since it is less than the memory bus clock.

Hehehe, nor sure I want to jump off this cliff again. Hope this made sense...

Forrest

Motherboard Help By HAL web site:
http://home.comcast.net/~hal-9000/

On Tue, 28 Oct 2003 16:56:04 +0100, "Alice in Chains"
wrote:

LS,

In my earlier time (pentium 1 /K6-2) the FSB was always 1 to 1 related to the memory bus. It should be as high as possible. 66x2 (133 MHz) outperforms 50x3 (150 MHz). Since the membus speed is not fixed to FSB anymore, I wonder what the impact is.

Which settings has the best performance?

(Board = KT7, proc = tbred B)

Setting a:
...

 

[Wes Newell]

On Tue, 28 Oct 2003 19:36:38 -0600, - HAL9000 wrote:

So in scenario A, the bottleneck has become the FSB clock since it is less than the memory bus clock.

I think you forget that the memory bus isn't DDR on the KT7 boards, so there's no way the memory bus can bottleneck the FSB on these boards.

On Tue, 28 Oct 2003 16:56:04 +0100, "Alice in Chains"
wrote:

Which settings has the best performance?

(Board = KT7, proc = tbred B)

Setting a:

FSB = 100 MHz
PCI = 33 MHz
MEM Bus= 133 MHz (HOSTCLK + PCICLK)
...

 

[Alice in Chains]

"Wes Newell" wrote:

How about C?

C:
FSB= 116 MHz
PCI= 33
MEM Bus = 149 MHz (HSTCLK+PCI)
multiplier = 19x

I've got PC150 ram in mine. But honestly, I don't think you'll notice the difference with either setting. A 33MHz jump in bus speeds provides only about 3-5% improvement in performance. Approx. 90% of data requests from the cpu are serviced by the internal cache.

Yeah, I tried that, but my PC133 Mem don't like it :-(

Thanks for the info!

Mike.

 

[- HAL9000]

Ahhh yes, thanks Wes. Hehehe, I thought I saw a cliff...

On Wed, 29 Oct 2003 07:23:41 GMT, "Wes Newell"
wrote:

On Tue, 28 Oct 2003 19:36:38 -0600, - HAL9000 wrote:

So in scenario A, the bottleneck has become the FSB clock since it is less than the memory bus clock.

I think you forget that the memory bus isn't DDR on the KT7 boards, so there's no way the memory bus can bottleneck the FSB on these boards.

On Tue, 28 Oct 2003 16:56:04 +0100, "Alice in Chains" wrote:

Which settings has the best performance?
...