memory bandwidth is cut in half which can definitely play a role.

What are the rest of your clocks doing? Is PCI still running at 33MHz?

The simplest explanation is that these games run fine on a 667MHz CPU but not on a 333MHz CPU...

That's normal when setting the FSB to 66 MHz, which in addition to lowering the CPU frequency it also reduces the RAM bandwidth.

ok, but i mean a game like sega rally, which is made for P1 processors shall manage this, those cpus ran also on 66 FSB. it depends on game, most are fine but there are some games which do not like the settings. i mean NFS3 for example runs fine.

I think the PCI should not be affected by the FSB

atom1kk wrote on 2023-11-08, 18:26:

ok, but i mean a game like sega rally, which is made for P1 processors shall manage this, those cpus ran also on 66 FSB. it depends on game, most are fine but there are some games which do not like the settings. i mean NFS3 for example runs fine.

I think the PCI should not be affected by the FSB

Are some of these games software rendered? And at what resolution?
See this table, and note the MTRR write combining difference as well.
50 to 133MHz FSB on a BX Mainboard

all the games which struggle are software rendered. i change the FSB on the MoBo with the switches not in the bios. the pci is untouched.

at the moment there are two games which i found, the one is sega rally the other is Grand Prix 2. the funny thing is, i used a 800 mhz cpu before which i used with 400 mhz. and there GP2 run worse than on the 667 at 333 mhz.

this is what i do not understand.

Sounds like write combining isn't enabled. In games that uses software rendering, performance in higher resolutions will suffer without it.

It can be enabled in the BIOS settings by setting "Video Memory Cache Mode" to UWSC. If that isn't available or doesn't work, then use MTRRLFBE or FASTVID (they can used in AUTOEXEC.BAT) to enable it only for the LFB region.

atom1kk wrote on 2023-11-10, 11:34:

all the games which struggle are software rendered. i change the FSB on the MoBo with the switches not in the bios. the pci is untouched.

at the moment there are two games which i found, the one is sega rally the other is Grand Prix 2. the funny thing is, i used a 800 mhz cpu before which i used with 400 mhz. and there GP2 run worse than on the 667 at 333 mhz.

this is what i do not understand.

What motherboard are you using ?

Also, are you certain that you do not need to change jumpers to keep the PCI bus at 33MHz when changing FSB ? PCI clock speed is implemented as a clock divider on at least some motherboards. Some boards have a manual setting for this divider.

If you set FSB to 133MHz, the clock divider for PCI should be 4. If the FSB is lowered to 66MHz with the divider still set to 4, you get a 16MHz PCI bus.

I have the Gigabyte ga-6vxc7-4x-p board. yeah but why do som egames run than better on a lower cpu at 66 fsb than with a faster on 66fsb

atom1kk wrote on 2023-11-10, 14:44:

I have the Gigabyte ga-6vxc7-4x-p board. yeah but why do som egames run than better on a lower cpu at 66 fsb than with a faster on 66fsb

If you have the same or similar board revision to the one I looked at the manual for, your FSB and PCI clock divider are set on dip switch block SW1.

DIP switches 1 and 2 on that block should be the ones setting the PCI divider and both should be set to ON to get a PCI divider of 2, which is appropriate for a 66MHz FSB to get a 33MHz PCI bus (66/2=33).

My theory :

If DIP switches 1 and/or 2 on block SW1 are either not set to ON, OR are not working as expected, the PCI divider will be set to a higher value than 2, which would slow down the PCI bus abnormally.

Also does your 800MHz CPU have a nominal FSB of 100MHz ? If so, it might actually help narrow down the issue.

atom1kk wrote on 2023-11-10, 14:44:

I have the Gigabyte ga-6vxc7-4x-p board. yeah but why do som egames run than better on a lower cpu at 66 fsb than with a faster on 66fsb

Are you switching CPU now? Please rephrase.

darry wrote on 2023-11-10, 15:55:

DIP switches 1 and 2 on that block should be the ones setting the PCI divider and both should be set to ON to get a PCI divider of 2, which is appropriate for a 66MHz FSB to get a 33MHz PCI bus (66/2=33).

In this generation of motherboards the PCI frequency is hardly ever an issue. And at the the common FSB speeds (66,100,133) the board should have an automatic 33 MHz PCI.
It can be verified by looking op the type of the PLL chip on the board, next to the AGP slot. I can't see what type it is now.

gerwin wrote on 2023-11-10, 17:47:

Are you switching CPU now? Please rephrase. […]
Show full quote

atom1kk wrote on 2023-11-10, 14:44:

I have the Gigabyte ga-6vxc7-4x-p board. yeah but why do som egames run than better on a lower cpu at 66 fsb than with a faster on 66fsb

Are you switching CPU now? Please rephrase.

darry wrote on 2023-11-10, 15:55:

DIP switches 1 and 2 on that block should be the ones setting the PCI divider and both should be set to ON to get a PCI divider of 2, which is appropriate for a 66MHz FSB to get a 33MHz PCI bus (66/2=33).

In this generation of motherboards the PCI frequency is hardly ever an issue. And at the the common FSB speeds (66,100,133) the board should have an automatic 33 MHz PCI.
It can be verified by looking op the type of the PLL chip on the board, next to the AGP slot. I can't see what type it is now.

Based on my interpretation of the DIP switches, the divider setting is manual, if the FSB is also set manually (I could be wrong, of course).

The first 2 DIP switches on SW1 are always the same for a given divider setting ( 2, 3 or 4). The remaining DIP switches likely set the FSB clock.

If I am correct in the above deduction, wrongly set or dysfunctional DIP switches 1 or 2 could massively curtail PCI bus performance.

My understanding is that OP is downclocking 1 of 2 CPUs using FSB . He is setting a 667MHz one (5*133) to 333 (5*66) and an 800MHz one to 400Mhz (undisclosed nominal FSB, 100MHz I suspect). When downclocked to 400MHz, that CPU is slower than the other one at 333MHz. To me that smells of a possible FSB divider issue.

darry wrote on 2023-11-10, 18:00:

If I am correct in the above deduction, wrongly set or dysfunctional DIP switches 1 or 2 could massively curtail PCI bus performance.

Yes I understood that. But I would still rather see the PLL type code and then the Datasheet for it.

gerwin wrote on 2023-11-10, 18:03:

darry wrote on 2023-11-10, 18:00:

If I am correct in the above deduction, wrongly set or dysfunctional DIP switches 1 or 2 could massively curtail PCI bus performance.

Yes I understood that. But I would still rather see the PLL type code and then the Datasheet for it.

Ack, TY For clarifying.
I edited my post just before you answered.

This is the board manual for what I believe is rev 5 of the board (OP may have a different one).

Maybe you can find some board photos where the PLL is identifiable.

EDIT: @gerwin I just realized that I am talking to someone who know more about this than I likely ever will.

darry wrote on 2023-11-10, 18:12:

Maybe you can find some board photos where the PLL is identifiable.

I tried, but either the pictures are not sharp, or the chip is obscured by an AGP bracket or card. I do see brand logo "ICS".
EDIT1: Manual says "ICS 9248DF-39"
EDIT2: Datasheet is available on the net. Lowest PCI clock it can have is 31MHz, at the 124MHz FSB setting that is - No problem there.

gerwin wrote on 2023-11-10, 18:26:

I tried, but either the pictures are not sharp, or the chip is obscured by an AGP bracket or card. I do see brand logo "ICS". ED […]
Show full quote

darry wrote on 2023-11-10, 18:12:

Maybe you can find some board photos where the PLL is identifiable.

I tried, but either the pictures are not sharp, or the chip is obscured by an AGP bracket or card. I do see brand logo "ICS".
EDIT1: Manual says "ICS 9248DF-39"
EDIT2: Datasheet is available on the net. Lowest PCI clock it can have is 31MHz, at the 124MHz FSB setting that is - No problem there.

You are quite right. My hypothesis is not possible.

All bit combinations are accounted for and PCI divider is not configurable independently according to the datasheet.

EDIT: Maybe It would be worth running some CPU, memory and PCI/VGA bandwidth benchmarks with both nominal and slowed FSB FSB settings to compare ?
@gerwin Could a faulty PLL possibly cause something like this ?

darry wrote on 2023-11-10, 20:25:

@gerwin Could a faulty PLL possibly cause something like this ?

Never heard of that being an issue, and we have too little information anyways.
I suppose it is up to the topic starter to decide how to proceed.

gerwin wrote on 2023-11-11, 17:28:

darry wrote on 2023-11-10, 20:25:

@gerwin Could a faulty PLL possibly cause something like this ?

Never heard of that being an issue, and we have too little information anyways.
I suppose it is up to the topic starter to decide how to proceed.

That makes sense to me as well.

Hi, sorry was some days off.

my 800 MHZ cpu is also a 133 mhz fsb version which i ran an 66 MHZ. I did some time ago. the sheet from darry is also the one which i used for the setting. I have seperate switches for DIP und system bus.
when i changed to the 667 mhz cpu i aslo have to change the DIP switches to the according multi, otherwise the pc will not boot sometimes. in this case i ran both cpus on 66 fsb with corresponding DIP. For 667 mhz cpu 5 x 66 and for 800 6 x 66