Currently I'm benchmarking all my graphic cards I have (well, of each type) and I'm nearly finished benchmarking my 3dfx cards. With the Voodoo4 4500 I noticed something strange regarding perfomance:

I benchmarked the Voodoo5 5500 PCI and AGP in my Pentium III-S 1400MHz (Tualatin; 512KB L2) system which worked perfect. After that I benchmarked the Voodoo4 4500 PCI and it ran great, too. I noticed that the 'timerefresh' in Half-Life didn't use much CPU power so the scores with all the boards should nearly identical using a Pentium 2 450MHz CPU. With the 5500's and the 4500PCI this was the case.

I hadn't benchmarked the 4500AGP at that time but started doing something else. I don't remember which card (might be either Banshee, Velocity, Voodoo3 or ATi Rage/Radeon) but it ended up deleting the old VSA-100 driver. Now I wanted to benchmark the 4500AGP and I noticed it was a lot slower. In the P2 450MHz the Voodoo4 4500 AGP scored 177.52 with Half Life timerefresh at 1024x768. The P3 1400 only 92.95! I tried everything to 'fix' this. Checking settings like vSync, BIOS, reinstalling drivers, making sure no 'old' drivers stayed behind, etc. It just did not work out!

Then I remembered I benchmarked the 4500's in the P2 450 after I benchmarked the V5 5500. So, first I validated the V5 5500 PCI for performance (as it uses the same driver) and this card run great! The performance was comparable with the P2 450 which is normal. Then I removed the 5500 PCI and popped in the 4500AGP and guess what ...180.75 in Half-Life timerefresh! That's double the performance I had earlier!

I'm completely out of ideas now. Why is the Voodoo4 4500 getting faster if the machine fitted a V5 5500 once?! I haven't checked registry settings and neither I reinstalled the whole system but besides that I can't think of something. The driver is just exact the same. It doesn't even need 'reinstalling' or even a reboot, the switch from 5500 to 4500 goes troublefree.

For the record: I used the latest 3dfx Voodoo4/5 V1.04.01b driver on Win98 SE.

Hey, that is some weird stuff going on there, doesn't make sense whatsoever.

When I'm benchmarking, I have a clean image of the OS, then after everybench, put it back, then install the drivers of the specific videocard.

I removed all the 3dfx files in the \WINDOWS\SYSTEM directory. Dethkarz is running a lot faster now. Maybe Dethkarz is using 'old' drivers (Other then Voodoo4) which might stood behind in the SYSTEM directory. Although I still think it's weird that the 5500's performed like they should.

Anyhow, it seems to be working now. I will reinstall the system when I make a radical change of hardware (3dfx -> ATi or nVidia). I thought the 3dfx boards themselves would be similar enough to avoid driver issues. Probably not for the V4 4500.

I haven't tested more so I haven't found the way to recreate this problem. It'll most likely return after installing either a Banshee, Velocity or Voodoo3. For the Velocity I used the Voodoo3 driver.

[edit]
And it's still weird. I benchmarked the 4500 PCI which was performing perfect. So I was interested if the 4500 AGP would perform correct now so I switched the boards. Q3 performance was correct, Dethkarz was just slow. About 50% of the performance I should have. I rebooted, no effect.

Then I took a peek in the BIOS and changed the 'Init Display First' setting from PCI to AGP. I never ever had a graphics card (which is single in a system) which had problems if this option was set incorrectly. But, let's give it a go. Booted up the system and started Dethkarz and I got a whopping 138.09FPS which is the performance it should have. I rebooted again and got good performance. To verify it was the 'Init Display First' setting I switched it back to PCI and benchmarked again: still good performance! It makes no sense ...