atom1kk wrote on 2022-11-23, 18:45:

6000 with a Ti 4400? I mean i get 6150 with a gf2 ti. The card should be alot faster

GF2 came out in 2000 when ~1GHz was the fastest CPU clock around. GF4 2002 with 2.4-2.8GHz clocks. Even GF2 is being cpu choked here.

> but im still curious why i wont reach your scores of 8091

try CL2 anyway, you can test stability with https://www.memtest.org

MrSteelrose wrote on 2022-11-23, 21:10:

i did three things. […]
Show full quote

i did three things.

1 overclocked system speed from 100 to 133 and thereby going from 700 to 933 Mhz
2 increased voltage by 10%
3 i changed timing on my SDRAM from CL2 to CL3 because they are labeled as 3.

now i got 6750 points. this is performance i can live with, but im still curious why i wont reach your scores of 8091. is the difference between 4200 and 4400 that big? or is it because you can run in 4x AGP speed?

The 440BX chipset does NOT support AGP 4X (so my tests were also done at AGP 2X). Also, again (people don't seem to understand how CPU / GPU bottlenecks work), all of these GPUs are bottlenecked by the CPU, so increasing the AGP speed even to 1000X wouldn't make a difference. 😀
AGP 2X is more than enough for this platform and even generally speaking, a GeForce 4 Ti 4600 would reach 99% of its full potential at AGP 2X.

CL2 is better (faster) than CL3. If your RAM modules work fine at CL2 (even if they're labled as "3"), then leave them at 2 (my tests were also done @ CL2).

Now, you did not mention anything else about your build.
For example, besides the motherboard, CPU, RAM, GPU & HDD... do you also have some other device/card? (maybe a PCI USB 2.0 adapter or something similar?). If you do, that card is the most likely culprit for the performance drop, so remove it immediately!
I would do the following test:
- remove everything, leave just the motherboard, CPU, RAM, GPU and HDD
- use a PS/2 mouse + keyboard
- load the BIOS defaults and maybe set CAS latency to "2" if your RAM supports it (i.e. if it does not crash). The performance improvement with CL2 won't be THAT high, though, so don't expect miracles.
- do a clean Win98SE reinstall
- after that, install only the following:

• DirectX 7.0
• 440BX chipset drivers (although they're not actually needed - Win98SE already comes with very good Intel chipset drivers which will give you the maximum performance out of the box - yes, I tested this many times!)
• nVIDIA Detonator 30.82

If, after doing all of the above, you still get 1000 points less in 3DMark2000... that means that something else is going on - something specific to your motherboard, maybe. In that case, you should see if there are any BIOS updates available for it (or, if you are already using the latest BIOS, a downgrade to an older version might help).
Good luck!

Ok now i did the following.
* I removed the USB 2.0 PCI card
* I set my monitor to 60 hz and windows to 16bit color
* I forced AA off, i think it was off already but not sure.

V-sync was already off since before and Bios settings were set to default and CL timing changed to 2 instead of 3

Now i get 7592 Score in 3dMarks 2000.

The USB PCI card was the main culprit i guess.

i have almost doubled my original score that was on 4137 points, i know i did some overclocking too but its stable so that counts 😀

You still beat me with 6.5% but that might be because you have a slightly better GPU

bloodem wrote on 2022-11-24, 06:17:

do you also have some other device/card? (maybe a PCI USB 2.0 adapter or something similar?)

fantastic guess!

Just curious... can anyone explain why the USB 2.0 card would be slowing things down so much?

Is it a resource sharing issue?

MrSteelrose wrote on 2022-11-24, 21:30:

i have almost doubled my original score that was on 4137 points, i know i did some overclocking too but its stable so that counts 😀

You still beat me with 6.5% but that might be because you have a slightly better GPU

Man... this is like the fourth time that you've mentioned the GPU, and I keep telling you that the GPU has nothing to do with it. 😁
Even with a GeForce 2 GTS I still get 8000+ points, because, as I mentioned countless times before, we are NOT GPU limited in this test (at the default 1024 x 768 x 16 resolution and color depth).
Hell, even with a GeForce 2 MX/MX400, if you drop the 3DMark 2000 resolution to 640 x 480 x 16, you should still get ~ 7500 - 8000 points with that CPU running at 933 MHz.

So, regarding the 5 - 6% difference... well, we are now close to margin of error. First of all, the board I've tested on is pretty fast (about 2% faster than other 440BX boards I have).
Secondly, based on my experience, 440BX motherboards tend to gain another 2 - 3% in performance if you populate all memory slots with identical memory modules (still unclear why this is the case, since the 440BX chipset doesn't support memory bank interleaving, however I've consistently replicated this behavior on many boards). This might also depend on the actual memory modules that are being used (though I've seen this behavior with various modules that have both high and low density memory ICs).

rasz_pl wrote on 2022-11-25, 04:10:

fantastic guess!

Haha... yeah, it was one of the only things left that would explain such a steep decline in performance.

Ozzuneoj wrote on 2022-11-25, 06:07:

Just curious... can anyone explain why the USB 2.0 card would be slowing things down so much?

Is it a resource sharing issue?

Hard to say... did not do a lot of digging into this issue. Originally, I thought it was somehow related to USB polling rate, but the polling rate shouldn't matter if there's no USB device inserted + I've had boards where I completely disabled the adapter in Device Manager, and the drop in performance still persisted after that. So, this points more towards some incompatibility with these older platforms.

bloodem wrote on 2022-11-25, 07:03:

Hard to say... did not do a lot of digging into this issue. Originally, I thought it was somehow related to USB polling rate, but the polling rate shouldn't matter if there's no USB device inserted + I've had boards where I completely disabled the adapter in Device Manager, and the drop in performance still persisted after that. So, this points more towards some incompatibility with these older platforms.

hmmm. I wonder what exactly disabling in device manager does. Judging by this it just unloads the driver but hardware might still be in initial state, and in this case this state might keep triggering interrupts? Afaik USB2.0 is so braindead it does trigger interrupts on every pooling interval. Still is shouldnt hurt if nothing is plugged in, as that state is detected (pulldowns on data pins). Now plugged in device is a different matter, SOF are send every 125 µs = might be even 8000 interrupts if its as braindead as I suspect.

New record 7621 😀. i closed some process that was running. i consider this issue solved

rasz_pl wrote on 2022-11-25, 08:40:

hmmm. I wonder what exactly disabling in device manager does. Judging by this it just unloads the driver but hardware might still be in initial state, and in this case this state might keep triggering interrupts? Afaik USB2.0 is so braindead it does trigger interrupts on every pooling interval. Still is shouldnt hurt if nothing is plugged in, as that state is detected (pulldowns on data pins). Now plugged in device is a different matter, SOF are send every 125 µs = might be even 8000 interrupts if its as braindead as I suspect.

Yes, it's very possible that something like this is causing the slowdown, but it's still unclear why it doesn't happen on every sub-1GHz platform. Some seem to be unaffected (even though they're still not particularly fast).

MrSteelrose wrote on 2022-11-25, 14:43:

New record 7621 😀. i closed some process that was running. i consider this issue solved

Great, enjoy your retro PC, it's a very nice one! 😀

256mb RAM should give you the optimal performance score for your build.

I tried to remove my usb 2.0 card snd got only 130 points more. So in my system it doesnt affect tio much.

atom1kk wrote on 2022-11-26, 13:56:

I tried to remove my usb 2.0 card snd got only 130 points more. So in my system it doesnt affect tio much.

The chipset of the card makes a difference. The old P4 system I saved from a scrap pile had an Nvidia USB 2.0 card installed but I wanted a card that had a nine pin USB header that I could attach the front USB cable to, and the one I found had an NEC chip. I measured the system's performance with both cards, neither, and individually - the Nvidia USB card caused a drop of about 300 points in 3D 2001 while the NEC one didn't measurably affect performance.

i have two voodoo 2 12 mb cards in SLI connected to the geforce 4 card 😀 for GLide support, and im getting a Roland canvas sc-55 soon for some sweet midi magic. DOS 6.22 here i come 😀