Hi,

I think these are caused by two different issues:

Socket3 wrote on 2026-06-02, 19:20:

- in Starcraft, whenever a sound clip that hasn't been played before loads in, the game freezes for about a second or less - otherwise the game is as smooth as on a 133MHz pentium

I think this most likely caused by HDD read delay, if it would be a problem with sound mixing then it would happen every-time the sound plays.

Socket3 wrote on 2026-06-02, 19:20:

- in Red Alert 95, when there are many sounds playing the framerate tanks and audio begins to stutter - otherwise the game runs very smooth

This is most likely caused by mixing the sounds together, as far as I know none of the Yamaha OPL3 sax chips supports HW mixing of digitized samples, so most likely the CPU has to do the sound mixing and that tanks the performance.

Socket3 wrote on 2026-06-02, 19:20:

- in Age of Empires when a new audio track begins to play the game freezes for a second. When many sword hit noises play audio begins to stutter.

This seems to be the mixture of the two issues, the new audio track is most likely a loading speed problem, the many sword hit noises stutter is most likely having to do the mixing in SW on a Cyrix MediaGX CPU.

Socket3 wrote on 2026-06-02, 19:20:

This got me thinking - what if I switch to a PCI sound card? Would that get rid of freezing and stuttering? Google claims some PCI sound cards like the SB Audio PCI can help offload the CPU - is there any truth to that? I'm talking about playing digital audio in games, not EAX or 3d audio.

I have heard that communicating with ISA devices slows down the processor on some motherboards, so moving to PCI will help if it is one of those motherboards, some cards do offer HW mixing of multiple digital sound channels, but whether the games will utilize it is another question.

MagefromAntares wrote on 2026-06-02, 19:39:

I think this most likely caused by HDD read delay

Maybe enabling DMA mode will fix the issue. Then again the IDE controller on the Cyrix 5530 chipset is pretty flaky... it might actually cause more problems then fix. Thing is, this issue is very common the socket 3 machines, regardless of HDD or storage controller (VLB, PCI, IDE or SCSI).

MagefromAntares wrote on 2026-06-02, 19:39:

I have heard that communicating with ISA devices slows down the processor on some motherboards, so moving to PCI will help if it is one of those motherboards, some cards do offer HW mixing of multiple digital sound channels, but whether the games will utilize it is another question.

And witch cards offer hardware mixing? I've got a CT4810, a CT5880, a CT5808 and a YFM724 on hand. I think I might also have an ESS Solo-1 and maybe a vortex II stashed someware. Witch one would you try?

Socket3 wrote on 2026-06-02, 21:14:

And witch cards offer hardware mixing? I've got a CT4810, a CT5880, a CT5808 and a YFM724 on hand. I think I might also have an ESS Solo-1 and maybe a vortex II stashed someware. Witch one would you try?

For the CT chips it is hard to determine if they support HW mixing of digital sounds the way games are using it, most data-sheets for them only mention the simultaneous Wavetable synthesis and simultaneous MIDI voices, not really used by games for audio effects. However if I read the specs right all of these CT chips can do PCI Bus Mastering to get sound data, so at least the final mixed sound doesn't have to go through the slow 16-bit 4-8Mhz ISA bus interface, so they will at least reduce the time that the bus is used to transfer data from memory.

YFM724 - While the documentation I found doesn't exactly mention the number of HW sound channels the "Direct Sound Hardware Acceleration" usually means that at least some amount of HW mixing is being done by the card.(https://theretroweb.com/expansioncard/documen … 44294516815.pdf) This card also supports PCI Bus Mastering, so the comment that I gave regarding the CT chips about it applies to this card as well.

ESS Solo-1 - There is a quite detailed data-sheet available for this(https://theretroweb.com/chip/documentation/ds … 27339683597.pdf), the document mentions a 8 channel mixer but only in regards of input, not output, on page 12 it details that there are 2 digital audio channels, although the documentation doesn't explicitly says it, but it seems that the second digital channel is supposed to be used when the card is in duplex mode, so it is there to help outputting sound during recording, not for mixing. On the same page it mentions that the card supports both "Normal DMA Mode", which means that the CPU has to poke the card to transfer data from system RAM and "Auto-Initialize DMA Mode" which means that the card itself can request data without having the CPU poke it first. The data-sheet is quite detailed so I don't have the time to read it completely now, but it seems to me that HW mixing is not a feature of this card, but the ability to do the "Auto-Initialize DMA", if Direct Sound supports it then it will give a performance advantage even without having HW mixing available.

vortex II - This seems to be the best candidate, its documentation(https://theretroweb.com/chip/documentation/ds … 2d537237231.pdf explicitly mentions "discrete 32-input, 16-channel digital mixer", which means that HW mixing is available on this card when using Direct Sound. This also seems to be overall the best card in the list according to the specs.

Socket3 wrote on Yesterday, 07:19:

Thank you for this information. I started testing the cards in my MediaGX. […]
Show full quote

Thank you for this information. I started testing the cards in my MediaGX.

Since there isn't any exact information on CPU offloading, I decided to test all the PCI cards. Here are the results so far:

- Creative CT4810 - hang on POST screen right before detecting the disk drives. *
- Creative/Ensoniq CT5880 - hang on POST screen right before detecting the disk drives.
- Yamaha DS-XG YMF 724 - correctly detected and installs just fine under win98. Noticeable performance improvement in Red Alert - the slowdown and sound stuttering are gone. Starcraft no longer hangs for 1 second before playing rarely used sounds. Somethimes it will stutter instead of hanging, but it only happened twice in 10 minutes of playing. I have not tested AoE yet, but overall, games feel like I'm playing on a 120MHz pentium.
- Aureal Vortex II 8820 - not yet tested
- ESS SOLO-1 - not yet tested

*I think the Creative cards are causing a resource conflict with the IDE controller. Either that or they require PCI features the P5GX-M and Cyrix 5530 doesn't support.

NOTE - I've seen discussions about ISA 8 and 16 bit recovery time both here and on other forums. By default, the P5GX-M has both set to 5. This setting as I understand is a "delay" in clock cycles meant to offer better compatibility with some ISA cards. It's how many clock cycles are skipped before the CPU pesters the ISA card with data. Too low settings will cause loss of data because the ISA card can't respond in time due to the limited 8mhz/11mhz ISA bus speed. Should be 8mhz on the P5GX-M. Too high a settings invites performance penalties. It seems "1" is required for best performance. I've had it down to :"3", and it made little to no difference. I'll retest with 16 bit I/O recovery time set to "1" and report back.

Thank you for this information. I started testing the cards in my MediaGX.

Since there isn't any exact information on CPU offloading, I decided to test all the PCI cards. Here are the results so far:

- Creative CT4810 - hang on POST screen right before detecting the disk drives. *
- Creative/Ensoniq CT5880 - hang on POST screen right before detecting the disk drives.
- Yamaha DS-XG YMF 724 - correctly detected and installs just fine under win98. Noticeable performance improvement in Red Alert - the slowdown and sound stuttering are gone. Starcraft no longer hangs for 1 second before playing rarely used sounds. Somethimes it will stutter instead of hanging, but it only happened twice in 10 minutes of playing. I have not tested AoE yet.
- Aureal Vortex II 8820 - not yet tested
- ESS SOLO-1 - not yet tested

*I think the Creative cards are causing a resource conflict with the IDE controller. Either that or they require PCI features the P5GX-M and Cyrix 5530 doesn't support.

NOTE - I've seen discussions about ISA 8 and 16 bit recovery time both here and on other forums. By default, the P5GX-M has both set to 5. This setting as I understand is a "delay" in clock cycles meant to offer better compatibility with some ISA cards. It's how many clock cycles are skipped before the CPU pesters the ISA card with data. Too low settings will cause loss of data because the ISA card can't respond in time due to the limited 8mhz/11mhz ISA bus speed. Should be 8mhz on the P5GX-M. Too high a settings invites performance penalties. It seems "1" is required for best performance. I've had it down to :"3", and it made little to no difference. I'll retest with 16 bit I/O recovery time set to "1" + the ISA OPL3sax and report back.

Socket3 wrote on Yesterday, 07:19:

NOTE - I've seen discussions about ISA 8 and 16 bit recovery time both here and on other forums. By default, the P5GX-M has both set to 5. This setting as I understand is a "delay" in clock cycles meant to offer better compatibility with some ISA cards. It's how many clock cycles are skipped before the CPU pesters the ISA card with data. Too low settings will cause loss of data because the ISA card can't respond in time due to the limited 8mhz/11mhz ISA bus speed. Should be 8mhz on the P5GX-M. Too high a settings invites performance penalties. It seems "1" is required for best performance. I've had it down to :"3", and it made little to no difference. I'll retest with 16 bit I/O recovery time set to "1" + the ISA OPL3sax and report back.

This does seem to imply that the Cyrix MediaGX and its chipset combination slow down when accessing the ISA bus, which means that most likely any PCI sound card working in the system will have a performance advantage versus ISA cards simply by not slowing the CPU down to do the ISA interfacing.

Some of these cards may have drivers that require Pentium instructions.

I wonder what's up the the hangs at the BIOS.

There has been a ton of MediaGX discussion over the years on here so you should search too. Perhaps @feipoa will sense the topic too.

swaaye wrote on Yesterday, 21:33:

Some of these cards may have drivers that require Pentium Instructions.

Note that Socket3 in the opening posts mentions that the CPU is a Cyrix MediaGXm(https://datasheets.chipdb.org/Cyrix/MediaGX/gxmdb_v20.pdf), not the base MediaGX without the "m" at the end, the m postfixed version actually not only supports Pentium instructions, but also supports MMX.