As your motherboard dosnt have an ISA slot I would go for a Vortex 2 or ESS Solo sound card, they should be $10 to $15 on Ebay, perhaps even cheaper. Other PCI sound cards have (even worse) DOS support and as you say Creatives PCI sound cards have a reputation for hogging the PCI bus more than necessary.

What sound chip is your motherboard equipped with?

The soundchip on the board is an ESS technologies 1989S. Its supposedly a decent chip, but ive tried 3 different driver sets and the problem is still there. If i play with the Windows 98 sound acceleration settings the does appear to be slight differences but maybe not.

SuperHanSolo wrote:

The soundchip on the board is an ESS technologies 1989S. Its supposedly a decent chip, but ive tried 3 different driver sets and the problem is still there. If i play with the Windows 98 sound acceleration settings the does appear to be slight differences but maybe not.

That is the ESS Solo if Im not mistaken so you already have arguably the best PCI sound chip for DOS.

Some other member probably has some good tips when it comes to solving the PCI latency issues.

Have you noticed if the issue affects you in normal use (game playing, using the OS in other ways) or is it limited to 3dmark? If you only see the issue in the benchmark, but it does not effect anything else, you might just ignore it as a quirk of the motherboard.

Another suggestion is to clock your cpu back to default speeds and see if the issue goes away.

My last suggestion is to try different VIA chipset driver versions. Phil's Computer Lab hosts some different revisions:
http://www.philscomputerlab.com/via.html

You may find that one version works perfectly. Let's hope so, at least.

Best of luck!

See my post here Socket 7 system stability issues
It should fix your problems too.

The on board trident blade 3d is a surprisingly decent chip for 3d games up to 1999. Great for DOS games too, including the ramdac speed sensitive ones. I would have used the on board + one or two Voodoo 2 cards.

@SuperHanSolo

I don't know if you have this already considered: The super-socket platform with for example the VIA or ALI chipsets are limited with cacheable RAM compared to the absolute max. size specified from the manufacturer.

The VIA MVP3 or MVP4 chipset has less cacheable RAM compared to the ALI chipsets.

Depending if you use 'write-back' or 'write-through' option in BIOS you have the following limits for your VIA (Apollo) chipset based mainboard with 512kB L2 cache:

With 'write-back':
512kB L2 cache = 64 MB SD-RAM

With 'write-through':
512kB L2 cache = 128 MB SD-RAM

Links:
http://www.amd-k6.com/cacheable-ram-on-socket-7-platforms/

http://www.anandtech.com/show/248/4

Thanks for all the info guys ! I thought i knew a fair bit about these old rigs but you guys are on another level.

Kanecvr, i was thinking about using the blade3D + voodoo 2 SLI but i was worried about how well the blade would work and getting the drivers to work right on the entire system since this board is from a very different era from the Voodoo 2.
Though Im glad you think its worth a shot since id really like to use 3dfx hardware. I already have a Voodoo 1 4MB but its probably not going to be enough for this setup.

The whole cacheable memory limit is a bit of a bummer, especially considering im using a K6-2 with zero L2 cache. Though im guessing having the extra 128MB is still better than having to use the page file.

SuperHanSolo wrote:

The whole cacheable memory limit is a bit of a bummer, especially considering im using a K6-2 with zero L2 cache. Though im guessing having the extra 128MB is still better than having to use the page file.

That's certainly true, but only if the extra RAM is actually needed. From what I've read, the uncached RAM is what tends to get used first under Windows. So this issue hurts performance even if only a smaller amount of RAM is being utilized, unfortunately. I don't know how serious the impact is.

However, sometimes it can make sense to exceed the cacheable limit if the RAM is truly needed. A long time ago, my sister had an old socket-7 machine with a 64MB cacheable limit. It was being used at a late date when it needed more than 64MB to avoid swapping, so in her case it was a lot better off with more RAM.

Back when I was on socket 7 I didn't know anything about the issue. My super7 machine ultimately had 320MB in it before I upgraded. In retrospect maybe I should have left it at 256MB (I think that was it's cacheable limit). I had a K6-3 though which might be why I never noticed a performance impact.

I didn't know the BIOS for VIA MVP3/4 boards had a setting for write-back/write-through cache, and that this changes the cacheable range. Interesting. I'll try to remember that if I build one of these again.

I dont seem to have that write-back or write-through option on my Mitac MVP4, advance chipset features in the BIOS has the following memory cache related setings:
DRAM Read pipeline - Enabled
Sustained 3T Write - Enabled
Cache R/CPU W Pipeline - Enabled
Cache Timing - Fastest

Based on the Anandtech article it might be that the MVP4 uses Write Through cache strategy by default. Though theres no way for me to really check it.

So far i tried:

Running 1x128MB RAM (So that its all cacheable...hopefully)
Tried VIA drivers 443, 435, 429 (currently on 429)
VIA latency patch
Detonator 3 drivers, Forceware (currently on 12.41)

The system seems to be running faster, i still get the sound glitches towards the end of 3DMark 2000, its somewhat reduced but still there. Im putting that down to some kind of bug related to 3Dmark 2000. Since the problem doesnt seem to be manifesting elsewhere now.

Thanks for the help guys, now to continue finishing up the system and seeing what hardware changes i can make.

SuperHanSolo wrote:

I dont seem to have that write-back or write-through option on my Mitac MVP4, advance chipset features in the BIOS has the follo […]
Show full quote

I dont seem to have that write-back or write-through option on my Mitac MVP4, advance chipset features in the BIOS has the following memory cache related setings:
DRAM Read pipeline - Enabled
Sustained 3T Write - Enabled
Cache R/CPU W Pipeline - Enabled
Cache Timing - Fastest
Based on the Anandtech article it might be that the MVP4 uses Write Through cache strategy by default. Though theres no way for me to really check it.

It's your second option you've found in your BIOS. For example i have the Shuttle HOT597 super socket 7 mainboard with VIA MVP3 chipset. And there is the following written in manual:

DRAM Read Pipeline
This item allows you to set DRAM Read Pipeline function Enabled or Disabled.
Sustained 3T Write
This item allow you to enable or disable direct map write back / write through secondary
cache.
The Choice: Enabled, Disabled.
Cache Rd+CPU Wt Pipeline
This item allows you to enable/disable the cache timing.
The Choice: Enabled, Disabled.
Cache Timing
This item allows the user to set the cache timing. The options are Fast and Fastest.

Im a bit confused, so does enabling Sustained 3T Write enable Write-Back or Enable Write-Through ?