i believe isa is often not a separate bus but a pci extension; also this will wildly depend on settings for sound and gfx and also type of load (where eax with simple hi res framebuffer filled by soft routines is probably worst case but i have no clue what kind of load for 3d graphics there is on pci)

The impact is based on the prioritization of the PCI slots. Depending on the chipset, the lowest slot has the highest priority. You also have to make sure that the PCI slots are not sharing the same IRQ. This will ensure the best performance when working with PCI adapter cards.

You should probably be using an ISA card for DOS games anyway, but it's unlikely to affect performance. I mean PCI slots are intended to be filled, and often are with network cards, IDE controllers etc, and systems don't stop working. Maybe just avoid using PCI slots that share an IRQ and it will be fine.

I'd love to see benchmarks on this as well, although I would guess its a bigger factor for faster PCI videocards and less a factor for early PCI videocards.

Would it therefor make sense to use an ISA soundcard to maximize the performance of the system ?

Depends. Poor ISA bandwidth could drop performance even more, especially if 44kHZ 16-bit stereo is considered.

If you do experience slowdowns or stuttering sound, you will want to lower the PCI latency setting in n the BIOS. it can affect the speed of the game, but that is true of any non-hardware accelerated sound card.

If using an ISA sound card, depending on the ISA bus speed and wait states, it can slow down the game as well, especially due to the fact that the game has to keep the sound in sync with the video.

Just run tests yourself and see what the difference is.

cyclone3d wrote on 2020-10-28, 18:59:

If you do experience slowdowns or stuttering sound, you will want to lower the PCI latency setting in n the BIOS. it can affect the speed of the game, but that is true of any non-hardware accelerated sound card.

If using an ISA sound card, depending on the ISA bus speed and wait states, it can slow down the game as well, especially due to the fact that the game has to keep the sound in sync with the video.

Just run tests yourself and see what the difference is.

Reason I was asking is that I appeared to see a huge performance gain after I turned off the onboard sound. But I now have the impression it is totally unrelated.

I have a setup here and when running 3DMark99, I see big differences in different runs using the same hardware / software.
I always do a reboot of the machine, and then start 3DMark99 after win98 finished booting. I can imagine doing other things prior to running 3DMark99 could impact performance).

But with the same hardware / software / way of running 3DMark99 I still have results that range like this :

• 1772 3DMarks/ 4445 CPU 3DMarks
• 2174 3DMarks / 5032 CPU 3DMarks

This to me seems to be a pretty big difference.

Which board?

The Serpent Rider wrote on 2020-10-29, 13:23:

Which board?

It's a Packard Bell computer with a SiS 530 chipset. http://www.uktsupport.co.uk/pb/mb/980.htm

Seems legit, same findings here:

https://www.overclockers.com/forums/showthrea … efinitive-Guide

"Are you running any PCI cards? If you are take them out! (Unless you are running a PCI video card of course)
You see 3DMark does not use any PCI cards, but they steal CPU cycles slowing your Benchmark down, Sound Cards are especially guilty of this, you’d be surprised how much of a difference to your score this makes!"

Some soundcards steal a lot of CPU cycles (onboard is probably a big offender), any chance to put a Creative/Aureal/Crystal/Philips/Yamaha soundcard in that machine?

Hanamichi wrote on 2020-10-30, 17:19:

Seems legit, same findings here: […]
Show full quote

Seems legit, same findings here:

https://www.overclockers.com/forums/showthrea … efinitive-Guide

"Are you running any PCI cards? If you are take them out! (Unless you are running a PCI video card of course)
You see 3DMark does not use any PCI cards, but they steal CPU cycles slowing your Benchmark down, Sound Cards are especially guilty of this, you’d be surprised how much of a difference to your score this makes!"

Some soundcards steal a lot of CPU cycles (onboard is probably a big offender), any chance to put a Creative/Aureal/Crystal/Philips/Yamaha soundcard in that machine?

I did notice it runs a couple of fps faster in some tests when sound / network disabled (the sound is onboard, but on the PCI bus, and I've ran test enabled and disabled).

I used Final Reality to check the PCI bus transfer rate and that seems to be ok.

The thing is that with the exact same hardware config and software, the results vary from run to run. I'm not always getting a consistent score.
I can imagine that it will never be exactly the same, and it can vary from time to time, but the difference between a 3DMark score of 1900 Vs 2200 is substantial ( a couple of minutes in between runs, the PC not doing anything in between runs), and the extra fps the system gets in the 2 game demos (race and first person shooter) is big.

Thanks a lot for the link by the way

I distinctly remember myself trying to go "ISA-less" as quickly as possible around the turn of the millennium, due to the common wisdom of ISA cards slowing down modern systems back in the day.

I particularly noticed a severe slowdown when I used an ISA network card instead of a PCI network card.

Why wouldn't the same be true for sound cards? The ISA slot is limited to 8.33MHz, and the Intel 8237 DMA controller operates at even lower clock speeds than that. That's bound to cause some bottlenecks that you wouldn't get with PCI devices, right?

Well, I know the AWE64 and YMF718 ISA cards didn't do badly when Phil did his sound card performance comparison for Win98: https://www.youtube.com/watch?v=TC01uiyuJxI (albeit, with AGP graphics.)

Wouldn't go looking there for a performance boost, though. Better options sound quality wise in the PCI domain.

A similar test isa vs pci here too:
ISA (AWE64) vs. PCI (SB Live!) sound card benchmark

But again with an agp card.

Also I think results might vary massively from chipset to chipset.