aitotat wrote on 2021-08-31, 05:14:

I found a fix for the non-working WSS support. The Sierra WSS drivers are hardcoded to IRQ7 and DMA1

Yeah, those are the parameters that seem to work best for WSS mode. Many games can use other values too, but some are more picky than others.

Of course now there is no need for WSS anymore because of the AWE64.

There are a few edge cases where the WSS driver might be better than the Sound Blaster driver. Aladdin is one such game, though Cloudschatze did provide a fix there. Another example would be Turrican2 which can run at 48 kHz in WSS mode. Those are the ones that I know of, but there could be more.

By the way. The AWE64 does not have the clicking that I heared with SB16.

IIRC, the DMA clicking bug is only present on original (non-Vibra) SB16 and AWE32 cards. Creative's newer cards don't have that issue.

I know that there is no DMA clicking on AWE64 but I just wanted to make sure since I've just recently tested it with one of the best SB16 models.

So there you have even more reasons why SB16 support is not necessary and sometimes even worse than WSS support. I'll have to try Turrican 2. I'd almost like to remove the AWE64 again but I'll keep it for now. Maybe I should connect PC-speaker to it. There is no PC-speaker connector on the Labway. Then again PC-speaker does not sound "correct" when it is coming from sound card but at least volume can be adjusted. Well, maybe it would be nice if one of my retro computers had PC speaker connected to a sound card. I've tried it few times but I've always gone back to PC-speaker, sometimes just to hear the POST beep that is not heard from uninitialized sound card.

aitotat wrote on 2021-08-31, 07:50:

So there you have even more reasons why SB16 support is not necessary and sometimes even worse than WSS support.

Well, there are around 200 DOS games which support WSS natively. Additionally, some games that have no native WSS support can be modified to use it as well (though results may vary). And most earlier titles sound very good in SBPro mode, so not having a SB16 is mostly fine.

On the flip side, there are also a few late-era DOS games which do benefit from 16-bit sound, but can't use WSS. Notable examples include Crusader: No Remorse, Quake and Duke Nukem 3D. For that reason, it's handy to have a SB16 as a backup option. But then again, one could always play such games on a different PC. Of course, the presence of a GUS in your system makes a SB16 even less necessary, since many games support it as well.

I've never played Crusader: No Remorse so I don't know how well it plays on a 486 or does it support GUS. Duke Nukem 3D runs on a 486 but only with low resolution so I would play that on a Pentium system rather than on this system. And it supports GUS. Quake is definitely Pentium game but I don't remember if it supported GUS or not. Maybe it did. Anyway, GUS + WSS sure is a great combination if there is no SB16 support. And seems that 16-bit samples are more of a Pentium era stuff. So not much use for SB16/AWE on this system.

But I just happened to find a game where AWE64 is better than Labway. Star Trek: Judgment rites. It was recommended to test the SB16 DMA bug but now I hear clicks with Labway but not with AWE64. Since I was adjusting volumes it also became clear that the game messes with the mixer values. I think the Labway suffers from clipping and AWE64 does not. I configured the game to Sound Blaster Pro and I was using floppy version. By the way, I just recently found out that Judgment rites have a CD-ROM talkie version too. I have the White label releases from both 25th anniversary and Judgment rites. The 25th is a talkie but the Judgment rites is a floppy version on a CD-ROM. So I just assumed there was no talkie released. Both are great games and I definitely have to play the talkie now that I know it exists.

But back to AWE64 and Labway. PC-speaker now comes from AWE64 and because of the Judgment Rites issue I'm more happy to keep the AWE64 now (although I lose stereo sounds from Judgment Rites).

Here are the two bat files to switch between cards.
SBPRO.BAT:

1@echo off
2rem ===== SOUND BLASTER AWE64 "inactive" =====
3SET BLASTER=A240 I11 D3 H5 T6 P0 E620 F0
4C:\DRIVERS\UNISOUND\UNISOUND.COM /C2 /V90 /VW0 /VF0 /VC99 /VL99 /VP50
5
6rem ===== LABWAY AUDICIAN YMF71x "active" =====
7SET BLASTER=A220 I7 D1 H0 P300 T4 J0
8C:\DRIVERS\UNISOUND\UNISOUND.COM /C1 /V85 /VW80 /VF85 /VC85
9@echo !!! SOUND BLASTER PRO mode set !!!

SB16.BAT:

1@echo off
2rem ===== LABWAY AUDICIAN YMF71x "inactive" =====
3SET BLASTER=A240 I11 D3 H0 P300 T4 J0
4C:\DRIVERS\UNISOUND\UNISOUND.COM /C1 /V85 /VW0 /VF85 /VC85
5
6rem ===== SOUND BLASTER AWE64 "active" =====
7SET BLASTER=A220 I7 D1 H5 T6 P0 E620 F0
8C:\DRIVERS\UNISOUND\UNISOUND.COM /C2 /V90 /VW85 /VF90 /VC99 /VL99 /VP50
9@echo !!! SOUND BLASTER 16 mode set !!!

Thanks to Unisound it is really that simple to switch between cards. Since Labway produces OPL3 sounds to both modes it is necessary to get volumes about the same for both cards. I think this is good enough. I set the FM/Midi volume bit louder to get the X2GS little bit louder that is connected to Labway. I like when music is the main thing and not sound effects.

New CPUs to test! Write-back version of Intel DX4 and AMD 5x86-133. But why?

WB-DX4 is easy to explain. I thought I already had one but I did not. AMD 5x86 is harder to explain. DX4-100 is an excellent clock rate. It is more than fast enough for every 486 game and for some Pentium games as well (like the Pentium detail on Tyrian) and with turbo disabled it is slow enough for most of the games needing slowdown. With turbo disabled, this motherboard runs the CPU at 1/3 clock rate so it is close to 486-33. CPU runs on full 100 MHz speed but chipset holds the CPU 8 microsecs for every 12 microsecs. Some clock rate detection programs detects the clock rate as 48 MHz so I had MHz display on the case set at 50 / 100 for a long time. But the system always felt much slower than 50 MHz. Now the MHz display has 33 and 100. And the benchmark results show that 33 MHz is quite correct.

But when I tested the AMD 5x86 133 on the Asus PVI-486SP3 I was very disappointed. The de-turbo speed was way too high (and icd and ice did not work on AMD 5x86)! I don't know how much it was because of the 133 MHz clock rate. It definitely matters something but I think the de-turbo speed was 2/3 of the clock rate (holds 4 microsecs for every 12 microsecs). That basically makes de-turbo useless on a DX4 CPUs. I can't remember the details but I do remember that de-turbo was too fast and the motherboard on this system with DX4-100 was a lot better choice for me. I couldn't test the 5x86 on this motherboard then because I had no way to flash the BIOS for 5x86 support.

Maybe you have guessed by now but I took a look at the SIS chipset datasheet to look how the de-turbo (as SIS calls it on the datasheet) is implemented. There are two modes: 1/3 and 2/3 and it could (and should) have been selected from BIOS setup but no, did any manufacturer put it there? And even better, de-turbo can be activated by software and not only with hardware switch. Great if you do not have actual turbo-button on the case. But even better, L2 cache can be controlled, and memory timings as well. Then there is of course the L1 cache that can be enabled and disabled from CPU. So lots of combinations for finer slowdown. I'd like to experiment them. Maybe it turns out that AMD 5x86 is the best for this system or maybe not. Lets find out. I also have couple of cool ideas for a slowdown program but let's do some testing first.

McCake now has heat sink and rubber pad to support weight of the McCake. I had some problems that rarely sound could not be heard but McCake leds were flashing. After noticing that Roland MPU401AT has shorter pins than the Labway it seemed most likely that there were connection issues. There is no way to secure the McCake to Roland so the wavetable connector gets all the weight and it easily bends a bit. McCake is heavier than other wavetable cards and now there is heat sink to add more weight. So I placed a rubber pad between Roland and McCake. Now it should be better.

Just before the most recent problem I think I actually was messing with SIMMs so it it very likely that I pushed the McCake somewhat. It did look fine but with the short pins...

Why did I mess with SIMMs? Well I removed two of them just to test 32MB RAM. There was some issues with limiting RAM to 32MB using RAMDRIVE or XMSDSK (I also tried LIMITMEM but it does not work at all on this system). I've been testing Sierra talkies lately. I installed all Sierra collections to my primary system using new installers from sierrahelp.com. Then I copied the games to DX4 system with a microdrive. I wanted to use sierrahelp installers to get all unofficial fixes the most easy way. But it was much more work than I thought it would be.

Anyway, when limiting RAM (it is not necessary for any of those games but I do it in autoexec.bat by default) and exiting for example LSL6 talkie it just frozes just before entering DOS prompt (I'm using MS-DOS 7.1 without Windows 98SE). But that happens only when I have RAMDRIVE or XMSDSK loaded. I don't know why. But I started to look for alternatives, like the LIMITMEM that did not work at all. I suspect it requires PCI system with more BIOS functions. And I also tested if it would be better to just have 32MB RAM.

But I finally found a solution. Himem.sys has /INT15=xxxx parameter to reserve RAM for BIOS extended memory interface for compatibility with some software that requires it. I don't know any such program but that parameter is perfect to limit RAM. So I set it to /INT15=32768 for two of my boot menu items.

Now Aladdin works (that is one game that does not work if more than 32MB RAM is installed) and Sierra talkies exit properly and no need to use conventional memory to load ramdrives.

aitotat wrote on 2021-09-02, 17:10:

Maybe you have guessed by now but I took a look at the SIS chipset datasheet to look how the de-turbo (as SIS calls it on the datasheet) is implemented. There are two modes: 1/3 and 2/3 and it could (and should) have been selected from BIOS setup but no, did any manufacturer put it there? And even better, de-turbo can be activated by software and not only with hardware switch. Great if you do not have actual turbo-button on the case. But even better, L2 cache can be controlled, and memory timings as well. Then there is of course the L1 cache that can be enabled and disabled from CPU. So lots of combinations for finer slowdown. I'd like to experiment them. Maybe it turns out that AMD 5x86 is the best for this system or maybe not. Lets find out. I also have couple of cool ideas for a slowdown program but let's do some testing first.

You might want to look at the SiS 496-497 data sheet also.
Compared to the 471 it has much finer-grained throttling capability.
You can control the STOPCLK very precisely.
There are two 8-bit counters, one for STOPCLK on, one for off.

For this reason the 486F55 mobo is the only 486 mobo I will keep.
The SiS 496/497 are quite common, so I guess it would be worthy to experiment with that.
Maybe a quite simple TSR might be sufficient to get a precisely hardware-throttable mobo.
Having a computer that can be throttled below XT speed would be nice for cheating purposes on some old DOS games, too 😀

I was hoping to avoid power management functions (to avoid need for TSR) but it should be considered as well. However I'm quite sure I'd need to have the 496/497 chipset for supporting it. The simpler stuff (turbo mode, L2 cache, memory timings) I should be able to support without having 496/497 chipset since they seem to be very similar to 471. Mostly just different register locations. I hope I have time next week to actually start experimenting.

By the way, there was another issue with Sierra talkies, not just the quit to dos prompt with ramdrive problem. Ever since my first Sound Blaster I've set IRQ to 5 that is default for all but earliest Sound Blasters. Only the earliest have IRQ 7 set to default. I wonder why since IRQ7 has always been default for printer port. I read from somewhere here at vogons that IRQ7 is the more compatible one for Sound Blasters (there area some games that are hardcoded to 7 and some to 5) so I decided to start using IRQ 7 when I replaced CT3900 with Labway (and AWE64). I did not reconfigure LPT IRQ since I have nothing connected to it anyway and everything seemed to work without issues. Until Sierra talkies.

Speech starts fine but quite soon there are some unwanted noise, or repeated parts and sometimes game crashes. That happened with both AWE64 and Labway. So I did a quick test with IRQ 5 and no problems at all, not with either of cards. So definitely IRQ conflict then. Since I don't need printer port I disabled it completely and just in case set printer IRQ to 5. I had to remove VLB I/O-card to access the jumpers. Removing and inserting VLB cards is always problematic and I don't like it at all. So definitely no IRQ conflicts anymore but the problem did not go away. IRQ5 works just fine and IRQ7 does not. I don't know why. Maybe problem with the Sierra drivers but I'll stay at IRQ5 as I've always had. Thanks to Unisound it is very easy to try IRQ7 for a game that needs it.

KQ7 crashed a lot with IRQ7 but now with IRQ5 I finished first chapter without any issues at all.

Before starting the SIS slowdown, I decided to do the MPU port trapping. And success. 330h and 331h are now successfully directed to 320h and 321h and that is where the X2GS is found. I have absolutely minimal implementation at the moment meaning no command line options or even any way to uninstall the TSR. But it works and I'll definitely implement the missing stuff during next week. Actually port trapping is the easy part (although EMM386 shows some weird behavior but I'll test that one more). The command line handling and uninstall part will take most of the code.

Maybe you remember but I used to have X2GS at 300h. Why is it at 320h now? Maybe I wrote previously that I have Lo-tech 8-bit IDE Adapter. If I did I was wrong, I have Lo-tech ISA CompactFlash Adapter (they are very similar). The latter always uses ports 300h - 31Fh so there were address conflict. It didn't actually matter since no drive was connected but I changed the MIDI port to 320h just in case.

I also made some other minor changes to SBPRO.BAT and SB16.BAT. I fine tuned volume a bit.
SBPRO.BAT:

1@echo off
2rem ===== SOUND BLASTER AWE64 "inactive" =====
3SET BLASTER=A240 I11 D3 H5 T6 P0 E620 F0
4C:\DRIVERS\UNISOUND\UNISOUND.COM /C2 /V90 /VW0 /VF0 /VC99 /VL80 /VP50
5
6rem ===== LABWAY AUDICIAN YMF71x "active" =====
7SET BLASTER=A220 I5 D1 H0 P320 T4 J0
8C:\DRIVERS\UNISOUND\UNISOUND.COM /C1 /V88 /VW80 /VF85 /VC85
9@echo !!! SOUND BLASTER PRO mode set !!!

SB16.BAT:

1@echo off
2rem ===== LABWAY AUDICIAN YMF71x "inactive" =====
3SET BLASTER=A240 I11 D3 H0 P320 T4 J0
4C:\DRIVERS\UNISOUND\UNISOUND.COM /C1 /V88 /VW0 /VF85 /VC85
5
6rem ===== SOUND BLASTER AWE64 "active" =====
7SET BLASTER=A220 I5 D1 H5 T6 P0 E620 F0
8C:\DRIVERS\UNISOUND\UNISOUND.COM /C2 /V90 /VW85 /VF90 /VC99 /VL80 /VP50
9@echo !!! SOUND BLASTER 16 mode set !!!

McCake gain values defined in mt32-pi.cfg are halfed from the default values.

I also had to change the way cards are chained. Now they go like this:

• PC-Speaker -> AWE64
• Labway -> AWE64 CD in
• Roland MPU401AT & McCake -> AWE64 Line in
• AWE64 -> GUS Line in (reversing stereo)
• GUS -> speakers (reversing stereo)

I wanted to fine tune McCake volume and with previous connection it also adjusted GUS volume. Now it is better although I need to use more RCA -> miniplug cables.

aitotat wrote on 2021-09-09, 09:34:

I was hoping to avoid power management functions (to avoid need for TSR) but it should be considered as well. However I'm quite […]
Show full quote

I was hoping to avoid power management functions (to avoid need for TSR) but it should be considered as well. However I'm quite sure I'd need to have the 496/497 chipset for supporting it. The simpler stuff (turbo mode, L2 cache, memory timings) I should be able to support without having 496/497 chipset since they seem to be very similar to 471. Mostly just different register locations. I hope I have time next week to actually start experimenting.

By the way, there was another issue with Sierra talkies, not just the quit to dos prompt with ramdrive problem. Ever since my first Sound Blaster I've set IRQ to 5 that is default for all but earliest Sound Blasters. Only the earliest have IRQ 7 set to default. I wonder why since IRQ7 has always been default for printer port. I read from somewhere here at vogons that IRQ7 is the more compatible one for Sound Blasters (there area some games that are hardcoded to 7 and some to 5) so I decided to start using IRQ 7 when I replaced CT3900 with Labway (and AWE64). I did not reconfigure LPT IRQ since I have nothing connected to it anyway and everything seemed to work without issues. Until Sierra talkies.

Speech starts fine but quite soon there are some unwanted noise, or repeated parts and sometimes game crashes. That happened with both AWE64 and Labway. So I did a quick test with IRQ 5 and no problems at all, not with either of cards. So definitely IRQ conflict then. Since I don't need printer port I disabled it completely and just in case set printer IRQ to 5. I had to remove VLB I/O-card to access the jumpers. Removing and inserting VLB cards is always problematic and I don't like it at all. So definitely no IRQ conflicts anymore but the problem did not go away. IRQ5 works just fine and IRQ7 does not. I don't know why. Maybe problem with the Sierra drivers but I'll stay at IRQ5 as I've always had. Thanks to Unisound it is very easy to try IRQ7 for a game that needs it.

KQ7 crashed a lot with IRQ7 but now with IRQ5 I finished first chapter without any issues at all.

Maybe my memory is off, and I was using an aztech card with sbp2 emulation, but back in the day, what I thought I did was have it set to IRQ 5 normally, and have a batch file to reset it to IRQ 7 with a new SETBLASTER string for stuff that didn't like 5... without disabling printer port. Since I thought that as long as I wasn't actively using the printer port, it wouldn't matter.... and I don't remember a problem doing that.

I haven't had time to do anything with my DX4, not even start experimenting the SIS slowdown stuff a little bit. I've spend all time I could on the MIDI port trapping (it will be called MIDIto) and I'm quite pleased with it. It works nicely on DX4 and on Copam as well. Having Dreamblaster X2 on Copam was quite useless since most General Midi games for that era PC are hardcoded to port 330h. But now that is fixed. I have to do some final polishing before releasing MIDIto. It prints a lot not so useful information at the moment and some sort of instructions need to be added. But it will definitely be ready to be released during weekend, maybe even before it.

So I haven't had time to test McCake for possible power issues. But I've been thinking about replacing the AT PSU with ATX and I think it would be a good time to go back to the original case again. I've been thinking about making a new HDD-CF-drive that would take two 3,5" slots. I'd like to add the McCake display to it and MHz display one way or the other. The first way would be to use McCake to display it. The other would be to use POST diagnostic card with 7-segment displays at the end of cable. I got two of those kind of POST cards (but I've not tried either of them yet). The SIS slowdown utility would then display appropriate (user defined) value on the display(s). But I'm thinking too far ahead. It will take a lot of time to get that far.

And here is MIDIto. I hope it helps someone other than me as well.

I got something new:

This is the CPU I think might be best for this system but we'll see. Not sure when. Midito has taken time way more than I initially thought. And I'm having many more ideas for it. Some may not be worth of the time. I even had this crazy idea of combining the AWE64 and Labway to act as a one card. Basically to redirect OPL3 and SB Pro commands to Labway and rest to AWE64. That is likely a fix for a problem that does not exist. I can easily switch between cards so why would I want a TSR with possible issues to make two cards act as one? Maybe not all ideas are worth implementing. I have better ideas though. But that means I do not have time to start the SIS slowdown program or test the CPUs. And now I have three waiting.

I also found ATX PSUs. Both are Nexus PSU. The other is NX-3500 and the other is NX-4090. They are from 2003 and 2004 based on this and this review.

So they were once quite good but why replace old AT PSU with old ATX PSU? Well the NX-3500 is just old enough to have -5V that the NX-4090 or any modern PSU does not have. But there are ways to get -5V so that would not be a big issue. The bigger problem is that I need much power on 5V rail and barely nothing on 12V rail. These PSUs give 30A and 28A on 5V rail and that is a big improvement over the 20A that the old AT PSU can provide. But if I would buy new PSU I would have to do some searching to get one that would provide more power to 5V than the AT PSU. And it would likely be expensive. And I don't want cheap PSU. I'd rather use old high quality one than new cheap one. So either of these should be excellent. NX-3500 has the -5V but the NX-4090 is supposed to be quieter and it likely is unused (seller wasn't sure but there is absolutely no dust on the fan or inside the PSU). The NX-3500 looks quite clean too but it has been used. And as an extra bonus these PSUs are from that time when I bought this computer. So one more thing I like about these. I still haven't tried either of them though.

No updates for some time but now there are. Actually I changed sound cards again but more on that on some later post. I still haven't replaced the power supply. I did install the 350W Nexus to another case to try how easy it is to replace AT power supply with ATX.

And there are many problems. First I needed to drill new holes so I could install the power supply correct way (fan down so it works like it should). Second the ATX power supply has too many molex (and SATA) cables and they are way too long. So it is a pain to hide all the wires. And possibly impossible on some cases.

Then you need the ATX to AT converter and that makes the motherboard power cable even longer. And those converter cables are badly designed. The soft power wires on them are too short since they need to be connected to the power switch on front of the case. So all those cables are just a mess that will affect ventilation for sure. And makes everything hard to reach.

But it gets even worse. Those Nexus PSUs are actually manufactured by FSP group. That wouldn't matter otherwise but they have that glue that turns conductive when aging. And bad caps. And the unused 400W unit was manufactured in 2006 and it still had bad caps! I'm in the process of restoring the 400W unit since it was unused. I've managed to remove the glue but I need to replace the capacitors. But I won't trust those Nexus PSUs enough to use them in this machine. So I recommend to service the old AT PSU if possible or use a new ATX power supply. Forget old ATX power supplies. Even the "quality" ones can be time bombs. And as mentioned above, there will be all sorts of cabling issues fitting the ATX power supply in AT case.

Edit: I forgot to mention one more problem. When using ATX power supply as AT replacement and the power is off, the ATX power supply is in stand by mode even though the motherboard is not. One great thing about AT PSUs is that there is no useless and possibly dangerous stand by mode. Power is on only when you need it and want it.

But I still need 25A on the 5V rail so I need new power supply (actually I think I could use linear voltage regulator to power McCake from 12V rail. Then I would not need new PSU). I just recently found used Corsair VENGEANCE 650M. It is a modern 650W power supply that can supply 25A on 5V rail. It has Japanese capacitors so that should be good as well. It is modular so I don't have to use all those completely unnecessary SATA power cables. The model I have is RPS0049. There seems to be different models under the same name. I don't have that single/multi rail switch as seen on some pictures. This is what I have and take a look what I did to it. This is a good time to warn that PSUs are dangerous and should not be messed with. I was careful and still managed to get a shock from the 400W Nexus.

I converted it to AT power supply! (It can be quite easily converted back to ATX if necessary).

1. I added second power switch so it works just like AT power supply would. Those soft power leads on the adapter can be just connected together and covered with tape. The actual switch on the pictures was for testing purposes only. Naturally it will be connected to the power switch on the case. I didn't remove the power switch on the back of PSU. Power switch that does nothing can be dangerous and removing it would just have left a hole that would have to be covered with something. So there are now two power switches but the one on the back can simply be left on. I didn't even have to drill a new hole for the AT power cable. It fit to the same hole as the motherboard power cable.

2. I added linear voltage regulator to provide -5V. There was a perfect spot where to glue it. See that very small piece of board with regulator and two capacitors. -12V and ground wires are soldered under the PSU PCB since it was the easiest. The -5V output wire goes to the cable sleeve with all the other ATX power wires. So the -5V mod is almost invisible!

The modified PSU is tested and now it is time to install it to the DX4 system. I finally decided to switch back to the original case so I'll do that also. But since I now need to remove the motherboard from the old case, this is finally a good time to test all those CPUs I have waiting.

I removed the mainboard from the old case. I'll do the testing without case since there will be a lot of messing with jumpers. Maybe I should replace tantalum caps as well. They look just fine but I don't want to wait tantalum explosion destroying the board. Looks like there are 13 tantalum caps that are 10µF 16V and one 1µF. It will be fast to replace those.

But anyway, here are the CPUs:

On the top left there is the Intel 486DX4-100 that came with this motherboard. It has 16k of write-through cache. Below is another Intel 486DX4-100. It has 16k of write-back cache and is the last and best 486 by Intel. Date code is a bit weird because this CPU looks to be manufactured in 2000 week 18. That is a very late for a 486. Also weird is the black bottom. It is normal for Pentium processors so likely a very late 486 would have it also. Unfortunately I don't have another DX4 board to test with so I'll have to take the risk.

There are two AMD CPUs on the middle. Top one under the heatsink is Am5x86 133 MHz and below it there is Am486 DX4-120SV8B. 8 means 8k of cache and B means write-back cache (T would have been write-through). I'm planning to underclock that one to better compare it with Intel DX4-100. If it supports 4x multiplier (I don't think the 8k version does), then I'll also overclock it to 133MHz. Remember, I'm not trying to find fastest CPU. The thing is to test them all how they perform after slowing them down with different ways (turbo On/Off with both modes supported by chipset (1/3 and 2/3 speed), L1 cache On/Off, L2 cache On/Off).

Finally there is Cyrix 5x86-100. I'm hoping most from this since multiplier can be changed from command line. But I somehow hope the Intel would be the best here because it kind of belongs to this board after all this time.

One last picture about Corsair and Nexus PSUs: Both are modded and it should be quite obvious why modular PSU is better for AT cases.

I went with a Cyrix 5x86 on my Asus PVI-486sp3. I am running it at 40x3 for 120MHZ.
I only had one VLB card (a Trident) it was super duper flaky so I ended up using a PCI Tseng ET6000.
I figure it is one of the faster 486 machines out there.

I am using a new Amazon special AT power supply with mine. I can't remember the name but it was brand new and tested stable at the time of install. I haven't had any issues.

Test bench is ready! Time to do some benchmarking tomorrow. I was supposed to make a SiS slowdown program but MIDIto took all the time. I did, however, four very simple utils:

1sisl2off.com
2-----------
3Disables L2 cache
4
5sisl2on.com
6-----------
7Re-enables L2 cache that has been disabled with sisl2off.com.
8Note that you cannot use this program to enable L2 cache that has been
9disabled from BIOS setup!
10
11sistrbo1.com
12------------
13Programs de-turbo function to 1/3 CPU speed.
14
15sistrbo2.com
16------------
17Programs de-turbo function to 2/3 CPU speed.

So I can now enable/disable L2 cache from command line and also change de-turbo to 1/3 or 2/3 speed. I think especially the de-turbo programs can be useful for others as well so here are all the four utils.

There are some problems with the L2 disable/enable programs. For example system just froze when trying to disable L2 when L1 is disabled. Some other issues as well. Likely the L2 is not flushed properly on some cases so data loss is possible. Do not use them. The turbo mode programs work just fine and can be useful.

Testing will take a lot more time than I thought. Running Quake and Doom benchmarks at lowest speeds take a lot of time. I've tested the WT cache version of Intel DX4 and the WB version tests are almost done. Yes, the Intel DX4-100 WB seems to be real and not fake! It identifies correctly and WB cache is enabled once jumpers are correct. Here are jumpering instructions but those will put the CPU in WT mode (even if configured to write-back from BIOS setup). WB mode jumpering instructions can be found from readme that is included with 0401 BIOS. I'll post the correct jumpers once I've done all the testing.

I was going to test AMD 5x86 next but the heatsink was glued to a position that did not fit to socket 3. Maybe the CPU was used in a laptop? The heatsink was quite thin. So I decided to test the AMD DX4-120 next while removing the heat sink from the 5x86.

Turns out that the DX4-120 is what is written on it. It could have been actual 5x86 since some of them are sold as SV8B CPUs. But no, my was what it was supposed to be. Only 8k WB cache and no 4x multiplier. I was hoping it would have really been another 5x86 but no. At least I can now test how much having only 8k L1 cache effects (and looks like 8k WB is slower than 16K WT). I started testing it under clocked to 100 MHz so I could compare it to Intel CPUs.

I managed to remove the 5x86 heatsink quite easily because it was taped and not glued. This was under the heat sink:

Looks what it was supposed to be. I removed thermal paste from the Intel DX4 WB with IPA and the printing did not come off. So looks like it was not relabeled or anything.

My money goes on either the Cyrix or the Am586, depends on what they clock to.
I doubt that the DX-4's will perform as well. Either from Intel or AMD.
My 100GP is very stable at 40x3 and walked all over my AM486 DX4-100