xjas wrote on 2020-07-09, 03:22:

Have I missed some magic setting that I'm supposed to turn on or off, or what?

Turns out the answer to this is 'yes' - it's A3D.

It still slows down, but I'll take 12 FPS in a heavy battle over 3 FPS in a light one any day. I can even bump it up to 640x480 now.

Still curious what you guys are getting if you have this game on a similar system.

As you found out by yourself, using A3D on an already very CPU limited scenario is recipe for disaster. Resolution doesn't make much of a difference in your case, because you are CPU limited, the Voodoo3 is just chilling in the back, I reckon that even if you bump it up to 800x600 it will still be performing the same.
Descent 3 is demanding game from 1999, so it's no wonder that the K6 is having a poor time. However, looking at benchmarks from this forum, it seems like the timedemo runs significantly better than your experience. Perhaps you have another bottleneck in there? Any USB 2.0 cards installed perhaps? There was a discovery fairly recently that these cards kill performance on VIA chipsets, I'm wondering if network cards affect performance as well 😁

I do have an NEC USB2 card installed, but I need it for my flight stick, foot pedals, mouse, flash drives (which I use all the time), etc. I don't have a compatible rear panel bracket for the motherboard USB 1.x headers. Disabling joystick + mouse in-game was good for 3-4 FPS, but that's no fun. I'm well aware of the performance hit though.

Where are these other benchmarks you found? D3 doesn't have a "standard" timedemo AFAIK, you have to record your own and run it or download one from somewhere. I'd love to know what demo they were using to do a proper comparison.

No NIC installed - I ran out of PCI slots.

Yeah, that USB2 card probably destroys performance. Take a look around the forum, it was all the rage a while ago.

Welp. I tried disabling the USB2 card in the device manager. I benchmarked secret2.dem (the famous one used in all the Tom's Hardware tests back in the day) and one I recorded in a map that was seriously chugging. Here are the results:

Secret2.dem (Average / Min / Max FPS)
With USB2: 38.2 / 14 / 87
Without USB2: 71.5 / 17 / 135

Gammachugs.dem
With USB2: 33.0 / 8 / 82
Without USB2: 62.3 / 17 / 134

...I was ALL SET to brush this off and say, "it's fine, a few FPS doesn't matter to me", but I really can't ignore that. Not sure what to do now. Running off the USB2 card, I have my flight stick, throttle, DIY foot pedals, and a mutli-card reader/USB hub. The motherboard header has two USB1.1 ports but that's not enough ports & losing fast file transfers via flash drives would really suck.

Weirdly, Quake & UT99 show almost no change with the USB2 card disabled. D3 seems exceptionally sensitive.

Here are the demos I used if anyone else wants to try (mine requires Reactor Gamma.) Benchmarking instructions here.

What causes the issue with USB 2 cards?

Told ya, it's pretty crazy! I don't know what causes this, but it seems to be the VIA chipset in conjunction with Win9x. Not sure if someone's benchmarked under Win2K and XP to see if it still produces the same issues. Perhaps a solution can be found if someone's brave enough to look into it.

Personally, I can live without "fast" file transfers. I usually only copy a lot of files during an initial installation which is not a big issue as I can leave the system copying and go do something else around the house. Once that's done, any further file transfer is not really that big a deal as it's usually some driver or patch and those are usually small. Then, there's networking which can be just as fast as USB2 and doesn't require drives being plugged into computers all the time. Consider it as an alternative!

^^ I'm not using a VIA chipset at all. The board is ALi Aladdin 5 & my USB2 card is NEC.

My Shuttle XPC (VIA KT266, same 98SE) has USB2 integrated and it's fine.

Yeah, you're right, had a little brainfart! Not sure if this affects MVP3.

The other thread was reporting the same issue on VIA MVP though... I think USB2 is just badly designed for older systems. Too much overhead.

xjas wrote on 2020-07-09, 21:09:

Weirdly, Quake & UT99 show almost no change with the USB2 card disabled. D3 seems exceptionally sensitive.

was about to say that this makes sense considering all of the peripherals you are running off that usb 2.0 card (i'm assuming you don't play quake with pedals etc.), but if it still shows such a huge difference in a canned benchmark run where input isn't polled on the fly, there's really something off here...

what i would try is get a bracket for the onboard usb, as they should have a standard pinout and be much easier to find than PS/2 ones, and use a hub if possible. also if you have the hardware, this subject lends itself to doing a comparison with the same components built around a p3 500 or such, which would show how the game could actually perform back then at best. if i recally correctly SSE was aggressively marketed for this game as well, though i've so far not looked into whether it actually has a practical benefit here.

^^ yeah, I was really surprised by that. The FPS running the demo I made was a bit better than when I played it in-game, maybe because it wasn't handling inputs even though polling the USB card was still hogging resources?

Descent 3 supports 3DNow! and was well optimized for the K6 series. I'm not bothered if it "could" perform faster on a P3, as long as it's running as well as it should on my K6-2, it's fine. (This is literally the same motherboard I used to compete in the Descent 3 tournament at Uberfest 99, and in all the online ladders back in the day.) I could always run it on my Shuttle XPC if I need more FPS.

I stick those NEC USB2 cards in everything, BTW... or at least, I *used* to. I wonder if a different make of card would make any difference.