Ok... seems like there is no such software, so maybe I should ask a different question:

Which AGP graphics cards won't tolerate AGP frequency higher than 80 MHz?

I need to know the exact model of such a card.

I found there is a way to check it in SiSoft Sandra, but I am not sure if that software shows the correct values...
https://www.overclockers.com/forums/showthrea … -of-PCI-and-AGP

Seems like there is not software that would tell the truth for sure...

OK, I figured it out 😉
ATI Radeon R3** seem to be very sensitive to AGP BUS speed.
I testes Radeon 9500 by Sapphire, and Radeon 9800 PRO by HIS, and they won;t boot when the FSB is set to 163-164. At 160-162 they work but are very unstable. The maximum speed they tolare without errors is 158-159 MHz FSB. For sure they won't even post at 166MHz FSB!!!

I checked it on the following motherboards just to be 100 % sure:
- ASUS CUV266 (DDR)
- ABIT VH6T (SDR)
- ASUS TUSL2-C (SDR)
- ABIT ST6 (SDR)
On each board the results were almost the same.
Conclusion: Radeon R3** cards do not tolerate AGP frequencies higher than 80 MHz 😀

Can someonce else confirm that, please?

So It seems there is one motherboard that has the right AGP divider (2/5) for 166 FSB! 😁

I need to find out the same!

I'm not sure, if my MSI motherboard sets the divider for AGP at 133 mhz FSB to 66 mhz (1:2), because it starts at 100 mhz FSB (autodetected) and overclocking is only through BIOS in 1mhz steps.
I saw motherboards, that ignore setting dividers, when there is manual setting for FSB, and you'll reach next step (133/166), some of motherboards have set it through jumpers, and then, they considered it as overclock, even if you set jumper to 133 and in BIOS set 166 (it will overclock AGP).
I need to be sure, my motherboard set AGP frequency to 66 mhz (default), because it starts at 100 mhz and there are no jumpers. I am using 166 mhz sempron CPU, but it cannot detect correctly, and set it at 100mhz fsb * 10.5 = 1050 mhz.
After I raise FSB to 133, to have 1400, I need to be sure, it has autoset divider to 1:2, so AGP is not overclock.
Unfortunately, I've checked several programs (hw info, power strip, cpu-Z), no program is showing what is AGP frequency, or tries to detect it.
In BIOS, unfortunately, this motherboard doesn't show it also (some better motherboards shows the actual frequency, so you can see, in case you manualy selecting FSB, if it is overclocked, downclocked (sometimes, it will underclock to 60mhz, when you are approaching next step, eg when you are on 120 mhz FSB, it will not clock it higher, but rather use lower multiplier, so it runs stable)

I try find out, if motherboard let AGP at 66mhz, or it is overclocked, after I raised it to 133 (processor has 166, but it didnt detect)

Any help would be appritiated. Best would be program, that can detect it.

It seems there is other way, than the one described above,(besides using the oscilloscope) to check the PCI dividers 😉

I have found a differnt way to check if there are proper PCI dividers by using an external PCI SATA RAID CONTROLLER 😁

I've used ASUS TUSL 2-C for testing, which is known for having maximum PCI divider of 1/4

Two SSD were used for testing: one "slow" and old Kingston SSD 60 GB:

100 FSB (PCI 33MHz) - 75,5 MB/sec MAX

133 FSB (PCI 33MHz) - 76,4 MB/sec MAX

166 FSB (PCI 41MHz) - 103,8 MB/sec MAX

200 FSB (PCI 50MHz) - 104,7 MB/sec MAX

The other one FAST and stable SSD SAMSUNG 850 PRO 128GB:

100 FSB (PCI 33MHz) - 88,0 MB/sec MAX

133 FSB (PCI 33MHz) - 89,2 MB/sec MAX

166 FSB (PCI 41MHz) - 120,4 MB/sec MAX

200 FSB (PCI 50MHz) - 123,1 MB/sec MAX

Results:
The increased FSB from 100 MHz to 133 MHz gives minimal increase in maxium transfer rate (1.2% - 1.4%) and it is due to the higher clock speed od the CPU. On the other hand, the increase from 133 to 166 (the PCI clock jumps from 33 MHz to 41 MHz) is HUGE! 39% [/ b] in the case of "slow" SSD kingston disk and 35% [/ b] in the case of SAMSUNG 850 PRO 😀 Another step from 166 MHz to 200 MHz is smaller because we are limited by the controller itself. What is important is the fact that this test shows PERFECTLY that if you do not have the correct dividers for PCI then the transfer rate will increase SIGNIFICANTLY!

So what does it look like in the case of ECS P6S5AT? Of course, there will be no gain in tranfer rate because ECS HAS the "1/5" divider for PCI 166! 😁


KINGSTON:

133 FSB (PCI 33MHz) - 78,2 MB/sec MAX


155 FSB (PCI 30MHz) - 71,3 MB/sec MAX


166 FSB (PCI 33MHz) - 78,7 MB/sec MAX



SAMSUNG:

133 FSB (PCI 33MHz) - 92,4 MB/sec MAX


155 FSB (PCI 30MHz) - 83,3 MB/sec MAX

166 FSB (PCI 33MHz) - 92,6 MB/sec MAX



RESULTS:
Just as this table suggests:
https://imgbb.com/84xmqJ6
The ECS has a divisor of 1/4 for FSB 133 MHz and 1/5 for FSB 166 and FSB 150 MHz. In the case of 166, the divider of 5 is not a problem, but in the case of 150 MHz things are already underclocked ... because AGP is only 30MHz and AGP frequency is only 60 MHz, which affects to some degree performance. However, for the sake of the hardware, I think that it is better to have a slightly underclocked bus than a highy overclocked one 😜 It all depends on what you want to achieve of course.... But to build a stable and reliable machine I would stick to the proper dividers 😉 That's why this ECS rules:D hehe (I couldn't find so far any other socket 370 motherboard with a correct 1/5 divider at 166 MHz)

Those tests with the SATA PCI controller seem like an ideal solution for checking the motherboard's dividers if you don't have an oscilloscope and want to make sure 😉

Thats a good trick. Thank you.
Will it work also with non-SSD disk? With HDD?

I see a problem, it will show PCI frequency this way, but AGP can be set different way. In other words, maybe it sets only PCI frequency on next divider, but for AGP, it use overclock. Good example are Intel BX chipset boards. That have limited AGP dividers, but PCI have greater range od dividers.

Maybe I've found the answer. But I have it only theoreticaly.
If you find the board, that can limit AGP to 1x, and find demanding game, where that bandwidth of AGP 1x starts to be bottleneck. You actually don't get any performance by overclocking the card, or CPU (throug multiplier, or using higher multiplier CPU). (same FPS). Because of that AGP bandwidth limitation. Then, only way, how to get increased FPS, would be if AGP bandwidth will be increased, and that happens through overclocking. Of course, it is reasonable only on boards without AGP/PCI lock, so for example on socket A, up to KT600 chipset.

How to test it.
Lets start at 100Mhz FSB, the AGP will be 66 mhz for sure. Set AGP 1x. Run benchmark, lets say you'll get 50 fps.
Now increase speed of CPU, with same settings, if you don't get any FPS, then graphic card is bottleneck, eighter by it's max performance cap, or by bandwidth.
Increase performance (overclock) graphic card. If you don't get any FPS, it's AGP 1x, that is limiting factor. And here you go, that's exactly, what you wanted.
Now increase FSB, so it overclock AGP slot, now you should see increased performance. Hit for example 124 mhz FSB, some boards use here lower multiplier. (like underclocked from 133 divider). In this case, you should get less performance, than on 123 mhz FSB (big drop in performance).
But scenario can be different, as you go above 124mhz FSB, you still get performance increase. Then you know, underclocked next multiplier wasn't used, so AGP frequency is still raising. You'll hit 132 mhz, with yet max performance. And now go for 133 mhz. If there is huge drop in FPS, you know, motherboard used correct multiplier here, to get 66 mhz AGP. But if FPS will raise over 132 mhz FSB, you know, it continue with overclocking, so AGP is still overclocked, and motherboard didn't used next divider.

W.x. wrote on 2022-10-10, 04:33:

Thats a good trick. Thank you.
Will it work also with non-SSD disk? With HDD?

Yes, but the results will probably have smaller differences in performance since normal HDD drivers are slower than SSD ones.

W.x. wrote on 2022-10-10, 04:41:

Maybe I've found the answer. But I have it only theoreticaly.
....

In the case you are describing it's better to use an oscilloscope 😀

VooDooMan wrote on 2022-10-10, 16:11:

In the case you are describing it's better to use an oscilloscope 😀

Not everyone have one, or know how to use it. I do not have any. It's usually needed only on few motherboards, and it's only one-time only procedure. To me, more weird seems looking for oscilloscope, just to find out this one attribute. That's much more work, and money.

Just out of curiosity, are there maybe PCI diagnostic cards that tell the bus frequency as well?

IDK about that, but maybe there's cheap build it yourself options.. https://hackaday.com/2022/10/14/to-turn-an-at … the-spec-sheet/ though that would only be useful on full lot of PCI speeds, and lower AGP speeds, tell you if it's gone over 133 I'd think though.

Original Hardware: PCI / ISA / AT Clock measuring device.

VooDooMan wrote on 2018-12-14, 09:56:

Ok... seems like there is no such software, so maybe I should ask a different question:

Which AGP graphics cards won't tolerate AGP frequency higher than 80 MHz?

I need to know the exact model of such a card.

This is from my personal experience with an Intel 440BX-based board and FSB set to 133 MHz:

nVidia RivaTNT/TNT2: no
nVidia GeForce 2 MX: yes
nVidia GeForce 3/Ti: yes
nVidia GeForce 4 Ti4200: yes

ATi Radeon 9200: no
ATi Radeon 9600/9600 Pro: no
ATi Radeon 9700 Pro: no

It's by no means scientific, just anecdotal / personal experience.

Edit: This has come up before - check out https://www.anandtech.com/show/574