In an effort to test and troubleshoot the Windows 3.11 audio issues, I decided to set up a standalone test bench using another DFI G7S620-N motherboard.

This way I can more easily experiment with things as I try different configurations.

I have an Orpheus II LT card I'm using for this setup. Since it's basically the same as the full Orpheus II but minus the GUS part, it should hopefully suffice.

So far I have installed DOS 6.22 + WFW 3.11, along with audio drivers. It's replicating the same audio lagging issue I have on my main setup. This at least rules out some of the non-audio related drivers (e.g. SVGA video, CD-ROM, etc.), since I don't have those installed yet.

But I still don't know what the main issue is. I may start playing with different audio settings (ports, DMA, etc.). Alternatively, I might try the regular Crystal drivers instead of the Orpheus II version.

Man this is a great thread. Its very similar concept to my system, although mine is Socket 478. I really need to get mine back up and running.

Do you think this will ultimately cover the wide gamut of games DOS up to 98 (Even early XP?)

My system was a P4 3.06ghz with a FX 5900XT

Bancho wrote on 2024-04-03, 11:22:

Man this is a great thread. Its very similar concept to my system, although mine is Socket 478. I really need to get mine back up and running.

Do you think this will ultimately cover the wide gamut of games DOS up to 98 (Even early XP?)

My goal with this build is to cover 1993 to 2003. For DOS games that aren't speed sensitive, it does quite well. I haven't run into any major incompatibilities.

While I've done some initial testing of speed sensitive games and results are promising, I need to fully play through some of those games to know how well they'll work.

For Windows 98 and XP it's been rock solid based on initial testing/usage. I'll know more in the coming weeks and months as I put it through its paces.

My system was a P4 3.06ghz with a FX 5900XT

That looks like good specs for a similar build. Especially with the all important ISA slot(s) for DOS compatible sound cards. 😁

I'm using a slightly older GeForce 4, so not sure how they compare in terms of compatibility with the FX5900XT. I might try the latter at some point, since it would yield more performance headroom for XP.

Shponglefan wrote on 2024-04-03, 21:01:

I'm using a slightly older GeForce 4, so not sure how they compare in terms of compatibility with the FX5900XT. I might try the latter at some point, since it would yield more performance headroom for XP.

Both cards are similar in terms of Win9x game compatibility. But early GeForce 4 Ti cards can use 30.82 drivers which are more suitable for Need for Speed games (especially High Stakes). Also, the text corruption is more pronounced on FX cards.

That said, FX cards are better for gaming at high resolutions like 1600x1200. They also have less of a performance penalty for cranking up AA and AF. Lastly, Radeon X800 cards have even more power, but it comes at the cost of some minor compatibility issues.

Joseph_Joestar wrote on 2024-04-03, 21:15:

Lastly, Radeon X800 cards have even more power, but it comes at the cost of some minor compatibility issues.

How is the X800 for DOS compatibility?

I have an X800 and haven't really decided what I'm going to use it for.

Shponglefan wrote on 2024-04-03, 21:34:

How is the X800 for DOS compatibility?

I have an X800 and haven't really decided what I'm going to use it for.

I haven't used it in pure DOS a whole lot. The system where I have it isn't really geared toward that environment.

I would assume compatibility is similar to the other Radeon cards, which Gona has tested here.

Joseph_Joestar wrote on 2024-04-03, 21:43:

I would assume compatibility is similar to the other Radeon cards, which Gona has tested here.

So not as ideal as a GeForce4 in all likelihood.

I suppose I can always set up a testbench and run it through some tests...

I've been continuing to test audio under Windows 3.1 using MS Arcade.

I set up another test bench with a Socket 7 + Pentium 100. This was to hopefully rule out any speed sensitivity or other issues with the Pentium 4 setup. Unfortunately I still get the same audio lagging issues in MS Arcade.

I even tried an original Orpheus I for good measure, but the issue persists. Tried both the regular Crystal drivers and the modified Orpheus drivers. Even tried clean installs of both Windows 3.1 and Windows for Workgroups 3.11. Nothing has made a difference.

edited:

After some searching, it looks like this issue was reported back in 2022: Re: Orpheus Soundcard: a new DOS soundcard with SPDIF/OPL3/MPU support

Unfortunately doesn't look like there was any follow-up with a resolution.

edited part 2:

Just tested everything on a 486 DX2-66 setup and same results: system lag whenever audio plays. 🙁

Installed the GUS PnP drivers for Windows 3.11. These drivers worked much better than the Crystal ones. No more system lagging / audio stuttering with games.

They did overwrite the MIDI map as noted by keropi in the Orpheus II thread. But otherwise, everything seems to work.

I did have an initial issue with an audible tone/buzzing noise after installing them. This seems to be related to the Crystal Audio input. Changing the input to "Loop" fixes the problem and the buzzing goes away.

Not sure exactly what the issue is, since I don't have the same problem with other operating systems.

In other Windows 3.11 news, I did run into a problem with SimAnt. The game crashes after a few minutes with a General Protection fault in the modified SVGA driver. I haven't experienced crashes with other games or programs yet, so I'm hoping this is an isolated issue.

I intend to install a number of Maxis games so we'll see how those perform.

Installed video and audio drivers for Windows NT 4.0.

For audio drivers I used the Crystal Audio 4237 NT 4.0 drivers. They seemed to install fine and basic audio playback is working. Haven't tried any games yet, but that will be the next thing to test.

Also still need to install an MPU-401 driver to try to get proper wavetable card support.

For video drivers I installed nVidia 43.45 drivers from the Vogons library. Also installed fine, but will need some games testing to ensure things are fully working.

Continued testing various Maxis games under Windows 3.11.

Speed sensitive is definitely a challenge with these games with the faster speed options enabled. Fortunately, they all contain options to run at slower speeds or even paused.

I'm going to continue to experiment with throttling options and see if I can strike a balance between performance and playability (at faster settings). This would be especially nice for SimTower which only has two speed settings: fast mode enabled or disabled.

I do have a 486 DX-33 setup alongside the Pentium 4, so I might try duplicating my Windows 3.11 install on the 486. This would give me a period correct benchmark for how quickly these games run in comparison.

Played some more SimAnt and curiously did not see a video driver crash. I'm still not sure what triggered those previous crashes.

On an unrelated note, I didn't realize the Windows 3.x version of SimEarth lacked audio support expect for PC speaker. Though I suppose that makes sense, since it's a Windows 3.0 game and that version of Windows lacks native sound support.

I've been doing some throttling and DOS games testing this past week.

I had previously have done extensive testing using the previous Prescott 2M processor (Pentium 4 650). But hadn't done much since replacing it with the Cedar Mill Pentium 4 651 (D0 stepping).

An issue the previous Prescott 2M had was depending on the throttling settings it would produce a coil whine sound. This naturally limited the options I could use in practice. However, the Cedar Mill is dead silent regardless of the settings used.

The second thing I discovered is the range of multiplier options is greater with the Cedar Mill 651 processor. With the Prescott 2M I had two options: 14x or 17x.

This Cedar Mill processor supports a range from 12x through 17x. In combination with other throttling options, this gives more control over fine tuning relative performance.

Finally there was the revelation that CPUSPD could disable both L1 and L2 cache at the same time. Previously I'd been using SETMUL to disable L1 cache only. Since the motherboard doesn't support disabling L2 cache via SETMUL, I just assumed I could only disable L1 cache.

With CPUSPD while I can't independently disable L2 cache, I can use it to disable both L1 & L2 cache at the same time. Incidently, the motherboard's BIOS also only supports enabling or disabling both caches at the same time. There are no independent options for L1 or L2.

Disabling both L1 & L2 cache reduces performance to a lower end 486 approximating a 486DX-33 or a bit faster (benchmarks in next post).

I tested a number of games that have various issues when to run too fast: graphics issues, freezing/crashing, music issues, etc.

• Blackthorne
• Day of the Tentacle
• Dynablaster
• Rise of the Dragon
• Warcraft

All of these games appeared to work perfectly with L1 & L2 cache disabled.

The most notable was Blackthorne. It has an issue where music won't play back if the processor is too fast. With previous throttling options I could get in-game music to work, but not the music during the intro cutscene. By disabling both L1 & L2 cache, all music played back perfectly just how it runs on my real 486.

My original goal for this build was to cover speed-insensitive DOS gaming. But if this system can reliably emulate an early 486, it extends its usefulness for retro gaming.

I've done some basic benchmarking of various throttling options and compared these to a real 486 DX-33 and 386 DX-40.

I haven't tried all the permutations yet, but wanted to get a rough idea of how scalable South Bridge and multiplier settings are.

Between the eight South Bridge settings (1 to 8 ) and six multiplier settings (12x to 17x), there are a total of 48 combinations available. Adding in three cache states (both enabled, L1 disabled, both disabled), increases the total combinations to 144.

The benchmarks are also not a perfect approximation of the real hardware. I imagine differences in graphics pipeline also has an impact on the final numbers.

Did some more benchmarking at the other end of the spectrum, running 3DMark2001 SE and 3DMark03 under Windows XP.

The 3DMark03 results were predictably poor, especially since it couldn't run all the tests due to lack of Shader Model 2.0.

At some point I need to test some later AGP cards to see how they would compare.

Shponglefan wrote on 2024-03-18, 13:34:

Attempted to install Windows NT 4.0. Unfortunately it spontaneously rebooted just after loading CD-ROM drivers. The final image […]
Show full quote

Attempted to install Windows NT 4.0. Unfortunately it spontaneously rebooted just after loading CD-ROM drivers. The final image before rebooting was this lovely series of stripes.

NT4 Install Attempt.jpg

I recorded a video of what happened available here: Windows NT 4.0 install fail

I tried disabling hyperthreading and L1/L2 cache, but that didn't change anything. From what I've researched, it seems NT 4.0 has difficulties installing on later Pentium 4 systems.

I may have to try installing on a different platform and then migrating to this system.

I'd try the single CPU kernel to start with. Iirc this can be chosen in the f8 menu or is part of safe mode.
Also, I'd probably try it with no other cards installed.
Assuming that works, I'd add sp6 before trying the multicore kernel.
Worth also playing with those bios settings like "mps version" 1.1/1.4

myne wrote on 2024-04-22, 02:53:

I'd try the single CPU kernel to start with. Iirc this can be chosen in the f8 menu or is part of safe mode. Also, I'd probably […]
Show full quote

I'd try the single CPU kernel to start with. Iirc this can be chosen in the f8 menu or is part of safe mode.
Also, I'd probably try it with no other cards installed.
Assuming that works, I'd add sp6 before trying the multicore kernel.
Worth also playing with those bios settings like "mps version" 1.1/1.4

I managed to solve this previously with a Max CPUID BIOS setting: Re: Pentium 4 (LGA775 / i865) Multi-OS Build Log (DOS/3.11/95/NT/98/Me/2k/XP)

After which NT 4.0 installed without any issue. I also installed the SP6 service pack before switching the CPUID setting back in the BIOS.

I haven't done much with NT 4.0 yet, but I did install graphics and sound drivers and those seemed to work fine.

Did some more DOS benchmarking to get a better idea of how CPUSPD settings affect system performance.

L1 and L2 cache were enabled for these benchmarks. EMM386 was enabled and FASTVID was disabled.

First up is a series of benchmarks using 3DBench 1.0c. I used CPUSPD to set both throttle (ACPI) and multiplier settings. The throttle settings on this board range from 1 to 8. There are 6 multiplier settings (12x to 17x).

Changing the throttle settings created a larger step decrease in performance. Changing the multiplier settings created a more granular decrease for each throttle setting. For the most part this resulted in a relatively smooth downward trend. There is a bit more of a gap between throttle settings as performance gets lower.

However, this can also be further adjusted by disabling L1 cache. For example with L1 cache disabled and throttle of 2, I can get a range of FPS from 88.5 to 95.9 (12x to 17x multiplier). This fills in some of the gaps.

Next up are some Doom benchmarks. I haven't done a full 48 runs of Doom since it takes much longer to run that benchmark. However I included two charts, one showing decrease in throttle performance (17x multiplier) and the other showing effect of multiplier on performance (throttle set to 8 ).

Changing the multiplier has far less impact on performance for each throttle step than it does on 3D Bench. While changing throttle settings do result in a relatively linear downward trend.

Another round of benchmarks, this time testing the on demand clock modulation (ODCM) settings. EMM386 was again enabled, while FASTVID was disabled.

With ODCM the frame rates don't decrease as much as the ACPI throttle settings, and also don't exhibit the same linear drop. There was also a more significant gap between ODCM setting 2 versus 1.

Multiplier settings also behaved differently. With the ACPI throttling, reducing multiplier had progressively less of an effect as the throttle setting decreased. Whereas with ODCM the opposite was the case and multiplier settings had a progressively greater effect on performance.

3D Bench 1.0c:

Doom (high detail) (17x multiplier):

3D Bench 1.0c - ODCM vs ACPI (17x multiplier):

Doom (high detail) - ODCM vs ACPI (17x multiplier):

Did some testing of some speed sensitive games including Ultima VII, Titus the Fox, and Wing Commander.

Ultima VII is notorious for re-enabling a disabled cache. However, I found that it doesn't seem to do that on this system when I disable both L1 & L2 cache with CPUSPD.

With both caches disabled and the multiplier set to 12x, it runs a bit faster than my 486 DX-33. I'd estimate it's performing similar to a 486 DX-40.

It is possible to run it closer to 486 DX-33 speeds by setting either ODCM or ACPI throttling to 7. However, I noticed this introduces some music slowdown during the earthquake (screen shake effect).

Another option might be to tweak the game with the Ultima VII frame limited patch, but I haven't tried that yet.

Speaking of the earthquake, I did run into a GPU compatibility issue regarding that effect. With a GeForce2 or GeForce 3, the effect works fine. However, with a GeForce4 or GeForceFX, the screen shake causes the game to freeze.

After some poking around, I found a TSR called QFIX: Re: Ultima VII Earthquake effect (screen shaking)

It replaces the vertical screen shaking with a horizontal screen shake. It was apparently designed so Ultima VII could run on Riva 128 cards. But it also works with the GeForce4 and FX cards I tested.

In testing Titus the Fox, music playback is wonky with no throttling. Disabling L1 & L2 cache fixed that problem and the game seemed to play and sound fine.

I compared it to my 486 DX-33 and couldn't hear any difference.

Finally, I tested Wing Commander.

I do have a confession: I've never actually played Wing Commander before. I don't have a good sense of how fast it should normally run, though the 386 DX-33 seems most recommended.

To compare in-game speeds, I tried it on my 486 DX-33 with turbo off (approximates a 386 DX-33) and also compared the speed to Phil's Wing Commander 386 DX-33 video.

With the Pentium 4, I could approximate the same speed by disabling L1 & L2 cache, setting either ODCM or ACPI throttling to 4 and then tweaking the multiplier. 14x or 15x seemed to give a good approximation of 386 DX-33 speeds.

Using ODCM or ACPI throttling did make the mouse move slower on a couple screens, but didn't seem to otherwise affect gameplay. I'll probably have to play it a lot more to confirm and especially see how it plays compared to my de-turbo'd 486.

I also did some 3D Bench scores with both ODCM and ACPI set to 4 and going through various multiplier settings.