Thanks. So you've changed the Slot 1/2/3 IRQ designation from Auto to NA and changed the PCI Latency Timer from 80 to 0? Do you recall which fixed the issue, the former or the later?

the first 3 make the difference, here at least.

OK, I'll test it out.

CoffeeOne, what does your Slot 1/2/3 IRQ designations look like?

There's a 307 BIOS for this board. You two are using 306. Was there any reason you wanted to stay with 306?

---Deleted duplicate post which occurred during website shutdown---

It was difficult for me to determine from the video, but I think you are using PCB rev. 1.22? Mine is 1.8, which is presumably newer?

Unfortunately, setting the PCI slot IRQ's to NA did not help, neither did playing with the PCI latency timer.

I do have a better understanding of what is going on though. When I have CACHE WRITE CYCLE set to 3, I can see the full hardware listing in Device Manager as shown:

But when I have CACHE WRITE CYCLE set to 2, this is what Device Manager looks like:

There isn't much left in there. The reason why my VLB graphics card continues to work is because it doesn't need an IRQ. The LPT-to-ethernet continues to work, even with the yellow exclamation; I'm not sure why. Voodoo2 continues to work because it doesn't need an IRQ. Even though the CD-ROM drive don't show up in the Device Manager, it still works fine. See the directory listing of the CD-ROM contents on the left.

It seems like the motherboard, nor the operating system are assigning IRQs when CACHE WRITE CYCLE = 2. Since pshipkov is also using Win95, I am guessing that there was some minor hardware change between motherboard PCB v1.8 and v1.22 which lets pshipkov's board continue to work with CACHE WRITE CYCLE set to 2. Perhaps related to freeing up IRQ's so the operating system can assign them.

Next, I booted into NT4 with CACHE WRITE CYCLE = 2, and it had no problem with a PCI network card, presumably because the OS is assigning IRQs. Perhaps Windows98SE can also setup IRQ's better than Windows 95 for the case where the motherboard's BIOS isn't assigning IRQs or the BIOS is reserving IRQs.

The only physical differences I can spot between our two boards is that mine is using IRFS640A for U8, while pshipkov's is using NDP406BL. I remember adding the IRFS MOSFET over 10 years ago, but I don't remember if U8 on my board was unpopulated, or if it had a burnt out MOSFET. The specs on the two MOSFETs match.

If anyone has any ideas on how to correct this peculiar problem, please let me know.

PVI versioning came up before in another thread, that i cannot find at this time.
Version ID is not a fraction.
It is usually composed by major.minor numbers which are independent from each other.
Both boards are major version 1. The one here is minor version 22, yours is version 8.

Your explanation is correct. I should clarify that more explicitly.
This was the reason i shown the device manager in the video.

I have one more check I can do. I only tried BIOS 306 with IRQ assignments at AUTO. I can put try 306 again with IRQ set to NA. Chances are slim, but worth the effort since the board is still connected.

OK, I've given up on this. The older BIOS didn't help with NA either.

It's possible that my board is just very particular on DRAM modules in relation to assigning IRQ's. I noticed I can have CACHE WRITE CYCLE = 2 and have everything work, but I must set DRAM timing to Faster instead of Fastest, and in that case, the DRAM throughput is the same as CACHE WRITE CYCLE = 3 and DRAM to fastest.

I hope you get your PVI board working again. I'll be interested to know what the issue was.

The video i shared shows the PC working again.
It looks like now you can run the video card properly, but have problems with the LAN card.
Can this be a prickly LAN card that does not click with the mobo (or the other way around) ?

I am running the system with VLB graphics now and only testing network cards. It takes less time for me to test PCI network cards rather than PCI graphics cards.

I tried a Intel Pro 100S Desktop PCI LAN card and a Realtek PCI LAN card. Both work with CACHE WRITE CYCLE =3, but don't work with CACHE WRITE CYCLE = 2.

I have had no improvement with this issue. I gave up.

Tekram P6B40D-A5 based on Intel 440BX chipset.

Early in the thread i went through multiple dual slot 1 and 2 motherboards while on a mission to build an early Y2K workstation.
Most boards got rejected early in the process because straightforward reasons like no Tualatin-S support and/or limited clock generators without options for modding.
But the one in question here was left unsolved.

Great looking motherboard.
Clock generator can provide up to 83/133MHz . Adjustable through BIOS (options for 100, 103, 112, 133 MHz), which is nice.
Latest available BIOS is already in place.

Initially the board didn't light with Tualatin-S CPUs.
At some point i realized that it was the on-board clock multiplier jumpers.
They were set for 8x.
Once i set them to some lower multiplier vlaue, everything started working.
This is unusual, since on all other boards when CPUs with locked multipliers are inserted the on-board jumpers get ignored.

Had to curate the RAM modules quite a bit to reach tightest BIOS timings.

When using 133 MHz FSB the board may not light-up after reset. Requires power off/on.
Things were fine in DOS and Windows 95, but NT4 was tripping when additional PCI extension cards were present.
In those cases the video cards may not get recognized. Stepping down to 112MHz FSB fixed the problems.
Finally, after quite a bit of effort i was able to make it work at 133MHz FSB.
While everything functioned properly i have this lingering feeling that the system is not fully stable.

Tested with dual Tualatin-S 1400/133 CPUs, 768Mb RAM, Quadro2 Pro for DOS and Quadro FX 4000 for Windows, CF card.
All BIOS settings on max, except:
VIDEO RAM CACHEABLE = DISABLED, otherwise POST never completes.

On par with P2B-D and P3C-D, but much more fussy at 1400/133 compared to them.

I was coming across boards that booted C stepping Coppermine and earlier but acted dead with D steppings unless having microcode updated.

edit: some details CC0 Pentium III vs CD0 P3 ... CPUID 6-8-10... are there real differences???

Good link. Thanks. Brings few extra bits of clarity to my understanding.
Yet, Coppers work fine on this board, just not in 133MHz FSB config.
Very strange.

After the impressive outcome with Chicony CH-471B, the very good overall characteristics of CH-881A, and the not too shabby TK8880F / 2066A2 at 160MHz, the brand name caught my attention.
Wanted to check more of their classic ISA/VLB motherboards, preferably ones based on UMC chipsets. So, here we go ...

Chicony CH-498B revision 3.0 based UMC's UM8498F

Still in perfect condition.
Supports up to 1Mb level 2 cache and at least 64Mb FPM RAM.
Tried with 2x32Mb 72-pin modules. Worked right away.
I don't have bigger than 32Mb FPM modules. Didn't fill the 32-pin slots in addition to that. Potential limit is probably 80Mb (64+16).

Clock generator supports 60/66MHz.
There is wiring for 3 and 5 volts to CPU only.

Windowed AMI BIOS. Meh.

Couldn't find any reference to this model online. This was concerning initially because the board refused to complete POST with Am5x86 CPUs.
But that first test was late on the day of arrival. On the next morning - no problem. Another case of stale ancient electronics at play.
L2 cache and clockgen tables are printed on the silkscreen. Quickly figured out the VLB wait-state jumpers, but the rest of the jumpers next to the CPU socket are a mystery. They don't seem to do anything. Blindly changing some of them results in no observable difference.

Tested with 32Mb FPM RAM (EDO is not supported).
1Mb level 2 cache. Board is not very picky, so i was able to quickly find a good set of chips.
Ark1000VL for DOS and S3 Trio64 for Windows.
Promise EIDE2300Plus with CF card.

--- Am5x86 at 160MHz (4x40)

All BIOS settings on max, except:
AT CLOCK SELECT = PCLK/3

System is completely stable.
Very satisfying.
Performance is mediocre at best but I was not disappointed - maybe this will be compensated with greater overclocking scalability.

benchmark results

--- Am5x86 at 180MHz (3x60)

All BIOS settings on max, except:
CACHE WRITE HIT WAIT STATE = 1WS
DRAM WAIT STATE SELECT = 1 WS
LOCAL READY DELAY SETTING = 1 T
AT CLOCK SELECT = PCLK/4

5V to CPU and 5V Peltier for active cooling.

System is completely stable.
Performance is great.
Very satisfying.

benchmark results

--- Am5x86 at 200MHz (4x50)

This didn't go well.
System is very unstable no matter what.

Was able to capture a SpeedSys screen at least.
Location of CPU socket is right next to the VLB slots which is a problem for Peltier + big fan for active cooling. Used standard ISA IDE controller.

--- Am5x86 at 200MHz (3x60)

No lights entirely.

--- Intel P24T (2.5x40) (POD100)

System is unstable no matter what, but works just fine at 2.5x33MHz (POD83) which is not interesting for me to test.

Was able to capture a SpeedSys screen at least.

---

All in all, not a bad motherboard.
Mediocre at 160MHz.
Great at 180MHz - getting close to PC-Chips M919 v3.4B/F, which the highest mark in this category.
Unfortunately the assembly does not scale well and caps at 3x60. This was disappointing.
Even more disappointing is its instability with POD100.

pshipkov wrote on 2023-05-29, 02:25:

Nice. […]
Show full quote

Nice.

Some of the results seems to be contradicting each other. For example:
18 fps in Quake 1 suggests well optimized system, but the 3DS test suggests otherwise.
Both of them lean heavily on the FPU.

Wonder if a different VGA BIOS will help the DOS pixel pushing (Wolf3D, Doom) ?
You probably saw this post by @WJG6260, also this one.
Plenty of S3 cards there with details about BIOSes and other stuff.

Do you have the rest of the VLI BIOS settings maxed out ?

Let me give a VERY late reply to trying a different VGA Bios.
I found today the mentioned Diamond Bios 1.01 and programmed it to a W27E512 chip (the bios is only 32kB so I appended the binary 2 times, the W27E512 is a 64kB EEPROM). It works!
It is really a bit faster and removes the strange behaviour of the Spea BIOS (need to set the refresh rate with a tool before starting Windows 98SE)
Then I increased again the memory clock to ~71.5 MHz

Dos values:

13D Bench 1.0c                      93.4
2Chris 640x480                      28.8 / 17.3fps
3PCPlayer 640x480                   11.4
4Doom Max                           1225 / 61.0fps
5Quake                              18.1fps
6Speedsys 4.78 Video                15961
7Wolf3D                             130.1

Windows 98SE Values 800x600 16bit:

1Wintune2                           11417 - 12100 kPixel/sec

So values on par, but Doom, Wolf and Speedsys are higher.
Windows 98SE Wintune is also higher, I ran it 5 times, values were 11417, 11777, 12061, 11635, 12100.
But don't ask about the Chris 3D 640x480 values, I have no clue why the old values were that high, I guess it was a typo last time. Or the Chris bench is fishy in general, I don't know.
And yes, I still can not use Transparent, have to stick to Synchronise.

Maybe I will also try another Bios for the Ark 1000. Ark 1000 is by far the king for me under DOS. And it runs with Transparent, not a lot of card are able to do it @40MHz together with my IO card.

Was the perf improvement based on the memclock increase, or there was contribution from the 1.01 microcode ?
Can you share the 1.01 version please ?

ECS Panda 386V rev 1.1 (also known as Daewoo AL486V-D) based on ALi M1429 A1, M1432 A2, relabeled to Panda PR3029 A1, PR3031 A2.

Clearly a new old stock.
I felt guilty about removing the CPU with that sticker on it but was very careful. After completing the examination and testing, the processor went back in and the sticker looked like it was never touched.

Jumper setup is very straightforward.
There are dedicated jumpers for several different CPU types, but as far as i can tell they don't make any difference.

Clock generator goes up to 40MHz only.

Default BIOS offers plenty of options, but locks the frequency multiplier for IBM BL3 processors to 3x.
Feipoa provided a BIOS that he and Jakethompson extended with bunch of additional options, including multipliers for SXL2 and BL3 CPUs.

Ark1000VL video cards don't agree on complete stability with this motherboard. Require increased wait states, which lowers overall performance.
Used S3 Trio64 instead. Works great with tightest BIOS settings.

Promise EIDE-2300Plus worked great as well. Used Transcend 2Gb CF card for local storage.

256Kb of 15ns level 2 cache.
16Mb of 60ns RAM. BL3 processors cannot cache more than that, or things get slow.
Some of the offline graphics tests can benefit from more memory, but didn't run them for 386DX and SXL2 processors. Notes below.

ULSI FPUs can cause hangs upon reset. Needs power off/on.
Used such CPU for the Quake 1 test. Quake really likes ULSI - performance is higher than with FasMath chips.
For everything else - 33MHz rated FasMath (gray top - slightly faster than the black top one).

--- 386DX at 40MHz, ISA bus at 20MHz

All BIOS settings on max, except:
CYCLE CHECK POINT = NORMAL (best is FAST)
System is completely stable.
That's pretty much it - things just work.

Used the standard BIOS.

One of the slowest boards overall, except for Windows accelerated graphics - number two so far.
Didn't bother with complex offline graphics tests. They would take forever to complete.

--- TI 486SXL2-50 at 40MHz, ISA bus at 20MHz

All BIOS settings on max, except:
DRAM READ WS SPEED = FAST (best is FASTEST). This was necessary only for the Quake 1 test. Everything else i tried works with this setting maxed out.
System is completely stable.

SpeedsSys hangs. Many motherboards exhibit this problem. Nothing unique here.

Intermediate performance when compared to the rest of the tested boards, which is actually not bad given the mere 40MHz FSB/CPU frequency.
Excellent score in Windows accelerated graphics - number 3 so far. Quite impressive given the above.
Didn't run some of the offline graphics test. SXL2 at 40MHz is too slow for them.

--- TI 486SXL2-66 at 66MHz (2x33)

This configuration is possible, but requires increased wait states.
My only TI SXL2-66 processor is flaky. It does better with Peltier on top, but too much trouble for this intermediate configuration.

Couple of numbers for reference:
Wolf3D: 64 fps
Superscape: 37 fps
Doom: 19 fps

Lost interest in testing it further.

--- TI 486SXL2-66 at 80MHz (2x40)

The TI SXL2-66 CPU i have is unstable in this configuration no matter what.
It really is time for upgrade ...

--- IBM BL3 at 80MHz (2x40), ISA bus at 20MHz

Slower than the 3x33. Didn't test.
Couple of numbers for reference:
Wolf3D: 83 fps
Superscape: 43.4 fps
Doom: 22.3 fps

--- IBM BL3 at 100MHz (3x33), ISA bus at 16.6MHz

All BIOS settings on max, except:
CYCLE CHECK POINT = NORMAL (best is FAST). Most of the time is ok with FAST, but eventually will run cause problems.
System is completely stable. Felt great.

Feipoa/Jakethompson BIOS:

The fastest thing around for Windows GUI, by far. Impressive.
Numero uno in Wolf3D as well.
Very good overall performance in Superscape, Doom, and PC Player Benchmark.
Below the average in Quake 1 and complex FPU-heavy tasks.

---

benchmark results

Excellent choice for IBM BL3 processors, less so for the classic 386DX.
Not clear yet where is the upper limit with SXL2. The current 40MHz tests show slow clock-to-clock system.

---

Additional tests and comments by @Feipoa here.

There's a typo, "IBM BL3 at 80MHz (3x40)"
80 != 3x40

I think it would help a lot of people to include photos of the BIOS screens for your final, most optimal settings.

For my Daewoo AL486V-D system with modified BIOS, I have arrived at these settings:

At the first few glance, it appeared as if the 16-bit ISA I/O Command WS, Memory Command WS, and LDEVJ WS were perfectly stable at 0 WS and 0 CLK2, but over the course of years having this system setup, I've arrived at the more conservative and stable values shown above. I generally run a lot more hardware in my system compared to pshipkov, though.

Nonetheless, using:
Diamond S3 968 + MVP mpeg decoder card, and
Promise EIDE2300plus

my DOOM score is 26.4 fps when using an IBM BL3 at 3x33 MHz. Was your score 26.6 fps?

I think CLK2 is 2xFSB, so 2x33.3 / 8 = 8.33 MHz ISA. I normally run the ISA at around 10-11 MHz, so I'd have to check my notes for why I'm running it so slow.

Are you able to enable the Flash ROM function on your setup? I thought this was similar to shadowing a memory range for I/O cards (the Promise EIDE2300plus), but when I enable it, the system doesn't work.

Fixed it - 2x40.

This board sat on the test bench for over 2 weeks. Pushed it around quite a bit.
SB16 and ASS1869 audio cards work just fine with the listed above BIOS timings.
Didn't test network cards. The presence of CF cards reduces the need for network connectivity for such old machines.

First BIOS page here is the same as what you posted.
The second is all on max except the listed items.
AT BUS CLOCK = CLK2 / 4. This can be derived from the listed ISA bus speed in each section.
I think you punish this 386 class rust with Win95 if i am not mistaken. I stay on DOS and Win3.1 only. Win95 is much more fussy. Maybe that's the reason you have the BIOS timings so inflated.

Yes, 26.6 fps in Doom. But i have you listed as 26.7 in the charts - remember that i picked it from conversation early in the thread. Not that 0.3 fps matters. Just pointing out.

FLASH ROM FUNCTION was enabled at some point but there was no observable difference in behavior and performance.
CYCLE CHECK POINT can be set to FAST, but at the expense of increasing other wait states. Keeping it at NORMAL allows for decreasing these other guys and achieving better overall performance.

May post BIOS screens, but have now Asus P3C-D board on the bench, so that will have to wait.

In my experience, 2 weeks is good as a starting baseline. After years, new discoveries unfold, as odd as that sounds. Small hardware changes to systems pushed to the max can require reduced timings or hardware swapping. It happened on most of my ~30 systems. Not necessarily related to this, but also prevalent is the element of time, which causes hardware degredation, like electromigration, growth if silicon and doping impurities, rust, corrosion, intermittent PCB vias, etc.

You are mistaken. I do not focus on w95 for 386 class hardware; I focus on DOS/w3.11. I only started adding alternate w95/nt4/nt35 CF cards years later, for the purpose of testing CPUID and HWINFO in the other threads. They had voids of the more exotic 386 CPUs.

It is only my socket 3 setups which insist on a dual boot environment with w95c and NT4.

I'm pretty sure that 26.7 fps in DOOM reduced to 26.4 fps when I swapped the SPEA S3 968 card in favour of the Diamond S3 968 card. The SPEA has slightly better performance. I did this swap so I could use the mpeg decoder card on my BL3 system, which was more fitting, seeing how the BL3 couldn't decode mpegs on its own.

In terms of the extra hardware, I almost always run 3.5" floppy, 5.25" floppy, CD-ROM, sound, and ethernet on my hyped up 386's. My BL3 also has a zip-100 drive and the MPEG decoder. This is what I mean when I write that my systems are more stuffed with hardware compared to your systems. I prefer them with this added amount of character, even if it comes at the expensive of a more troublesome setup, and it almost always does.

pshipkov wrote on 2023-10-22, 05:17:

CYCLE CHECK POINT can be set to FAST, but at the expense of increasing other wait states. Keeping it at NORMAL allows for decreasing these other guys and achieving better overall performance.

I'd have to check my notes to be sure, but either my DRAM couldn't do fastest/fastest with CCP on NORMAL, or I found fast/fastest with CCP on FAST to be more optimal (either speed or stability).

pshipkov wrote on 2023-10-22, 05:17:

May post BIOS screens, but have now Asus P3C-D board on the bench, so that will have to wait.

It was just a general comment on what I think would help others most. I noticed you had posted a BIOS screenshot for your Panda board, yet you were using the BIOS I provided for testing, as evident by posting the higher ISA speeds. The default BIOS has the ISA fixed at 7.1 MHz.