@ekb - you are forgiven 😁

Here is what i got based on your pointers Andrew and EKB:

Red wire is pin B13, connected to +5V -> L1 WB enabled.
Yellow wire is pin R17, connected to GND -> 4x multiplier.

So i got the board running at 160MHz (4x40).
There was a bit of confusion initially. Apparently the BIOS can display maximum of 150MHz during POST, which confused me for a moment.
Even there is a bus divider option, 4x50 is a no-go, also there is no 60MHz option in the clock generator, so no dice for 3x60. Anyway.

Here is what SpeedSys reports for 4x40 and L1 WB.

CPU speed looks great, memory bandwidth is incorrect.
But i don't trust these synthetic tests 1 bit for a reason - they are more often than not misleading.

Running some real world game/apps tests show that 4x40MHz is quite slower than 3x50 on this motherboard. Didn't expect that.
Also, there is no measurable performance difference between L1 WT and WB. Didn't expect that either.

benchmark results

This FIC 86-VIP-IO2 motherboard is an oddball for sure.

Andrew, you said that you have similar board, wonder if you tried it at 3x50 ?
Did you observe the same performance advantage over 4x40 ?

on mobo 4DPS socket 3, there are undocumentary jumpers for 60 and 66 MHz. This is useful for Intel DX4 (60 * 2 = 120) or (66 * 2 = 133).
For AMD, 5x86 is practically irrelevant. Anyway, wait timings ruin the data bus speed (Cache or RAM).

I chose 50x3 = 150mhz AMD 5x86 for myself and made downvolting to 3.1 volts. To not get too hot. The drop in performance was 2-3%, but it's worth it to make the system quiet and not hot. No fans. 😉

Congratulations, everything turned out well 😀

Nice one pshipkov!
I'd expect that for games, since these
extra MHz on FSB can make a difference (for GPU).
On CPU hungry apps though I'd expect the opposite.
Haven't tried the 50MHz FSB though
not sure my GPU can handle it, but i'll give it a go.

My Jetway also report it wrong (150MHz) but
every utility I tried gives correct numbers.
I guess the problem is with the Award Bios.
Also, another revision of it only reported L2 cache
either 256k or none, even when I installed 1mb l2 on it!!
So, do not always trust your bios.

DataExpert EXP3406 rev 1.1 based on OPTi 82C495SLC.

Board is a looker with promising features:
- late/hybrid 386/486 chipset
- VLB slots
- write-back level 2 cache (10 cache banks)

Jumpers are scattered around and a bit illogical to follow. Had to consistently check reference manual, which was annoying.

Expected to be able to run 386 class CPU at higher than specification speeds, but unfortunately the board hard prohibits it.
Set of jumpers differentiates between CPU classes.
Each class is associated with its own set of base frequencies.
The 386 one is limited to 40MHz.
Bummer.

Board is picky about SRAM chips, but not too picky.

BIOS is full of options.

Used Ark1000 VLB in DOS, S3 Trio64 VLB in Windows, DTC2278E VLB EIDE controller.

--- AMD 386DX @40MHz

All BIOS settings on max, except FAST DECODE = DISABLED.
Small exception for rendering tests. Had to increase some wait states, otherwise the system hangs.
Not a visible issue for gaming and casual computing. Standard stability tests like CheckIt, etc. are happy too with tightest BIOS timings.

--- TI 486SXL2 @40MHz

Had to increase some of the BIOS timings to get through successful rendering tests:
DRAM read/write = 1 (from 0)
RAM timings = 3-1-1-1 (from 2-1-1-1)

DOOM test stutters during disk access. Not sure why. Using smartdrive minimizes the impact a lot.

--- IBM BL3

Ark1000VL
Promise EIDE 2300Plus. Turbo mode can occasionally hang the system during POST/BOOT. Fast mode works great.
16Mb RAM
256Kb L2 cache

Cyrix FasMath FPUs are unstable with BL3 processors.
Used ULSI DX/DLC 40MHz instead. Works great.

All BIOS settings on max.

Speedsys hangs.

---

benchmark results

All in all - not a bad motherboard.
Potential is there, but the 40MHz limit for 386 class CPUs is a limiting factor. Too bad.
Nice piece of hardware, definitely above average performer, but there are handful of better options out there.

A recent post by @mpe here reminded me of something i started before, but never finished properly - basically, try to find the best possible use of a slightly modified P24T (POD83) running at 100MHz.
The reason i left the job for later was the overall instability of the CPU with most Socket 3 motherboards.
Only few handled it properly. When it worked, performance significantly increased in several areas.

So, i replaced the 160MHz AMD processor with POD100 in one of the 486 PCs (showcased in this post) and gave it a quick spin with Quake 1. Shared the "achievement" in the same thread next to mpe's.

Here is the full story.

Applied @feipoa's mod to the CPU. With few small improvements.
Didn't cut the pin, but only unsoldered and bent it upwards, making sure it does not touch the heatsink.
Inserted a piece of insulator to form a bed for one of the diode legs.
Precisely bent the diode wires to fit exactly where i wanted them to be.
Used long and tin soldering hot end to ... well ... solder them in place, with minimal collateral damage.
Replaced the original fan with bigger one, was not sure at the time if the standard one will be enough when running heavy computation tasks.
Basically, made sure that the mod is easily reversible.
Here is the result:

Tested with the next motherboards:
Abit PB4 - will share info about it in upcoming post
Asus PVI-486SP3
Asus VLI-486SV2GX4 (picture shows 256Kb cache and 32Mb RAM, but is used with maxed-out to 1024Kb cache and 64Mb RAM)
Biostar MB-8433UUD-A (picture shows it with 512Kb, but is used with 256Kb)
LuckyStar-486EF
Hope i can obtain one day a PC-Chips M918i, looks like a good performer.

Why consider these 5 boards only ?
I tested many 486 ones, some of that is partially documented in this thread.
Many of them work with POD100, but are not the best performers and were discarded.
From the 5 above:
PB4 and LS were flaky before (with POD100). Gave them another good try - still flaky.
PVI barely worked before, spent some more time on it - improved on it - was able to run tests, worked decently well, but not stable enough to be a real computer.
VLI and 8433UUD did best.
So, down to 2 at the end.

Asus VLI in SpeedSys with all settings on max:

L1 cache in WB mode, Ark1000VL and Promise EISA 2300 Plus.
L1/L2 caches set in WB mode deliver peak performance. Most other 486 boards need L2 in WT mode with POD100 cpu.
If L1 cache is set to WB mode any configuration with more than one memory module (in the first slot) can be unstable in some situations. The motherboard cannot recognize modules bigger than 32Mb, so this is the limit.
Tested carefully with trusted 4, 8, 16 and 32 Mb ones that can handle up to 180MHz (3x60) and even 200MHz (3x66).
1x4, 1x8, 1x16, 1x32 - stable. Nx4, Nx8, Nx16, Nx32 - unstable.
At 83MHz (2.5x33) the issue is not present, only at 2.5x40MHz.

8433UUD in SpeedSys:

L1/L2 caches in WB mode for best performance.
EDIT:
The original post talked about the board operating with L2 cache disabled because of instabilities.
Comments by Feipoa below provided hints and confirmed that it can run stably with L2 enabled and tightest timings (*).
Turned out that the board is *extremely* picky about SRAM chips.
Had to find that specific permutation of 9 (UMC, 15ns) ones to make it happen. Will save the details, but I will not be going through the same exercise again for sure.
(*)All settings are on max, except IBC DEVSEL# DECODING, which had to be set to "medium" (down from "fast"), otherwise IDE controllers (on-board or external) do not recognize the CF card.

The standard set of benchmarks.
PVI and UUD were tested with Matrox Millennium PCI video adapter.
VLI was tested with Ark1000VL in DOS and S3 Trio64 (16-bit per pixel) in Windows.

POD100 > AMD 5x86 at 160MHz.
Matrox cards are way faster in Win GUI than anything else released in the mid 90ies, still S3 Trio64 VLB manages to catch-up but only when in 16 bits per pixel, otherwise perf is much lower.
Tested the Biostar UUD board withseveral other graphics cards. MGA was same or better, so it was used.

Quake 1 running on VLI + POD100 (L1/WB) + Ark1000VL = 27.3 fps. Pretty good. Probably the highest Quake 1 number on socket 3 hardware.

Quick Q1-PCPBench-3DS-SpeedSys roll with the VLI board:
https://www.petershipkov.com/temp/retro_pc_im … quake1_27.3.mp4

While i prefer non-hybrid systems, POD100 will probably stay in the VLI based PC, for extra flavor. Seems to be a good fit.

EDIT:
Decided to modify a second P24T processor. It went well. Satisfied with the result.

Finally - both P24T CPUs didn't survive POST at 50MHz base frequency in any of the motherboards.

Edom International Corporation 486VL3 (MP032F ?) based on OPTi 82C895/82C822/82C602 chipset.

Checked this board a while ago.
Cannot remember details anymore.
Going by the few notes i kept.

Notice how the VRM is (not) integrated in the PCB.
Very flimsy. Broke it twice by just tweaking the jumpers next to it.
Had to spend time to fix that.
Legs were already short and couldn't be soldered well enough, so sanded the bottom of the package carefully - effectively increasing the length of the legs which allowed me to solder them properly.
With that problem out of the way it was business as usual.

Board didn't want to run the AMD DX5 CPU at 4x40Mhz, so used the simple interposer from few posts above to force that behavior.
160MHz worked fine, but L1 cache refused to operate in WB mode.
Tested with 72pin SIMMs only.
Mr.BIOS on board - does not seem to offer anything special.
System was stable with all settings on max. Two exceptions below.

Used DTC2278E EIDE VLB controller.

Speedsys:

benchmark results

Favors VLB in DOS graphics - performance is pretty good. The best Wolf3D numbers with 160MHz CPU so far !
Failed the Doom test. Increased wait states didn't help. Looks like specific incompatibility with the Ark1000VL video card. Tested with two more VLB VGAs - S3 Tro64 and WD90C33-ZZ - worked fine. Didn't investigate in depth if there is a bigger VLB VGA compatibility issues, didn't feel that way.
Favors PCI in accelerated Windows graphics - performance is top notch. On par with some of the best 486 boards out there.
Computation tests are contradicting. Disappointing results in 3DS (DOS), but pretty good in LW3D (Windows).

Most OPTi based motherboards (well, at least the ones i had the chance to try so far) show mediocre performance, but this implementation is pretty good.
Board may look like your average citizen, but there is strength within. 😀

Not far ago in this post i talked about assembling a top-notch dual slot1/2 Pentium 3 based graphics workstation from year 2000-2001, using standard ATX/AGP motherboards.
That post is big and perhaps hard to follow. It was even bigger initially and completely out of hand, so trimmed it down to focus on 3 candidates only.
I already completed the PC and it is a done deal (will share the end result later), but figured it will be good to post the collected information for 3 more of the examined motherboards - one OR840 and two OR820.
Here is the first one:

---

Intel OR840 - IBM Intellistation M Pro (Type 6868), in mint condition

Majority of the computer hardware i have seen from companies like IBM, HP, Dell, Compaq, etc. tend to be too customized, restricted and quirky for my taste, so i avoid it in general, but this particular motherboard is pretty much vanilla Intel implementation with good IBM BIOS (updated to the latest version).
It lacks the magic touch of the Taiwanese manufacturers, which means limited/no options for tweaking and overclocking.
Requires custom PSU, which is inconvenient.
But other than that it checked all boxes, for me at least, so it was considered as a contender.

Clock generator is ICS9250BF-12, limited to 133MHz, so no dice for overclocking.
Boot times are a bit on the long side, but still within the reasonable limit.

IBM fully populated it with chips and other gear, compared to the Intel version. Integrated LAN, Sound and so on.
Swapped the coin cell battery socket. It was of non-standard type. The battery was less of a coin, but more of a satellite dish size and completely depleted. Soldered a standard one.
Also, added couple of jumpers that were missing in favor of proprietary IBM front panel.

Hit an interesting issue initially - boot process was unstable in a completely random way. 8 out of 10 tries led to no POST.
At some point i noticed few wrong text symbols in the BIOS settings. This hinted at a memory issue.
Until then, trusted 4x128Mb RIMMs were used. Switched to 4x512Mb 32ns ones - instant success.

Used these memory sticks at the end - 2Gb of RAM.
There is nothing special about them really.
The board handled well any 256/512 Mb RIMM modules i threw at it.

Initial tests were performed with P3 Coppermine 1000/256/133. All was good.
Replaced with Tualatins 1400/512/133 in Asus slotkets for the actual test cycle.
Silence. Not good.
Long story short - the widely praised Asus S370-DL rev. 1.02 slotkets just don't work with this board.
I am willing to pin it on the board itself, since they work in any other one i inserted them so far.
Bummer.
Ok.
I was already reaching for the soldering iron and considering which of the available slotkets should go "under the knife" in an attempt to make things work with the Tulies, but then i decided to first test things with the Coppermines and see how 840 fairs against Asus P2B-D which already emerged at the time as the front runner. If OR840 turned out to be a better deal than BX440, i would spend the time to find a way to upgrade it to Tualatins, otherwise - why bother.

So, here is what SpeedSys said.
Notice the impressive disk metrics.

Used the same NT4 installation outlined in the initial post that I linked above.
Tested with Quadro2 Pro and Quadro FX4000.
Quadro2 Pro seems to be the best overall period correct pro-grade graphics card from the entire pack (again - as outlined in the initial post).
Quadro FX4000 (GeForce 6800 Ultra) is the latest and greatest GPU that can run on these motherboards - to check the upper limit.
Compared with Asus P2B-D and the same dual Coppermine Pentium 3 1000/256/133 CPUs.
Both boards running at 133MHz FSB 33MHz PCI.
Also, included numbers of P2B-D running at 140/150MHz FSB with Tualatins, as a reference for peak performance.
And one last note - one of the rendering tests hangs. Kind of strange, because the board is rock stable otherwise.

It is obvious that Asus's implementation of BX440 is clock-to-clock faster than OR840 and can overclock on top of that, which further increases the performance gap between the two.
At that point i kind of lost interest in trying Tualatins on the IBM board.
It just didn't stand a chance.

Otherwise, this is really great piece of hardware. Solid, stable and just works.
If i was not on a journey for best compute/graphics performance, i would probably go for this one from everything i tested.

---

Quick test run captured on video.

pshipkov wrote on 2021-02-18, 21:30:

I applied @feipoa's mod to the CPU. With few small improvements. Didn't cut the pin, but only unsoldered and bent it upwards, ma […]
Show full quote

I applied @feipoa's mod to the CPU. With few small improvements.
Didn't cut the pin, but only unsoldered and bent it upwards, making sure it does not touch the heatsink.
Inserted a piece of insulator to form a bed for one of the diode legs.
Precisely bent the diode wires to fit exactly where i wanted them to be.
Used long and tin soldering hot end to ..Modifying the POD83's voltage regulator for overclocking. well ... solder them in place, with minimal collateral damage.

That is one very clean looking mod! Are you able to post your photo in the POD mod thread so others who stumble upon the thread can see how slick this hack can look? Modifying the POD83's voltage regulator for overclocking

pshipkov wrote on 2021-02-18, 21:30:

L1 cache in WB mode ! L2 cache is disabled. When enabled, the system becomes unstable, unless L2 Cache Wait States is set to 3-1 […]
Show full quote

L1 cache in WB mode !
L2 cache is disabled.
When enabled, the system becomes unstable, unless L2 Cache Wait States is set to 3-1-1-1, which turns-out to be actually slower than no cache. Curious.
Verified this very carefully, to make sure i am not missing something here.

Also, once L2 cache is off - the system is just rock stable - ticks fast and well.
All other settings are on max, except IBC DEVSEL# DECODING which had to be set to "medium" (down from "fast"), otherwise IDE controller does not recognize the CF card.[/quote]
You should be able to run the POD83 at 100 MHz just fine with L1-WB and L2 enabled. I used to have a setup with the MB-8433 and the POD-100 and GF2 and ran 3D games for hours without issue. Did you poke through the manual I wrote for this board? You need to ensure that you are using 256K of L2 cache double-banked. Do not use single-banked cache on this board with running a 40 MHz FSB and L2 set to 2-1-1-1. Also, it is best not to use the onboard IDE port. Ensure L2, if in WB mode, is set to TAG/ALTER BIT to 7+1. Or if WT mode, 8+0.

Shared the pictures there.

Read quite a few things about this board.
Where is the manual you got ?

The two boards i have behave similarly, so it is not a bad PCB at play, unless both are bad, but i doubt it.
512Kb is a no-go unless some hardware mods are applied. I think you shown that in a post around here (for 1024Kb).
I use 256Kb SRAM (8 chips).
Tried many different sets - IS61C256A, IS61C256AH, UM61256K, UM61256AK, UM612562CK, UM61256FK, UM61512AK, WB, GMC6512, TC55328P, WinBond, Mosel Vitelic, (exact chip designation escapes me right now), and so on.

With both 486@160MHz and POD100 CPUs the boards actually work really well - games from the era, DOS and Windows common tasks - all good.
I see instabilities only when the system is pushed hard - these are use cases that few people encounter, or care about.
Willing to attribute it to my "complete stability" disorder. 😀

Which cache chips you recommend based on your experience ?
BIOS settings here are 7+1 for WB, as you wrote.

Do you have pictures of BIOS+Jumper settings somewhere ?
To make sure i am not missing something obvious ?
This board holds a lot of punch, but the cache issues are nagging.

pshipkov wrote on 2021-02-18, 21:30:

Quake 1 running on VLI + POD100 + Ark1000VL = 25.7 fps. Pretty good.

Great score! What screen size was Quake set to? Also how much L2 cache was on the board?

The ver:2 manual I wrote is here:
http://www.vogonsdrivers.com/getfile.php?fileid=946
BIOS settings should be noted in the back of that manual. I do need to update the manual to include some more detail on the cache mod, but I'm concerned that I never will have time. My home life is a mess. I'd like to document how to make the board jumperable for 512K and 1024K double-banked with user removable jumpers before I die. I've already done this on one of my boards.

Could you document the reproducible steps which cause the instability? I've never had my POD100 crash on the MB-8433UUD with 2-1-1-1. Are you using EDO memory? Do not use EDO. Please try a single stick of 32 MB FPM (if L2-WB) or a single stick of 64 MB (if L2-WT). And please use a PCI controller card, like SATA a Promise TX2Plus. I remember having issues with hard drives on the built-in IDE controller when I was using 40 MHz at 2-1-1-1.

@Chadti99
This seems to be the highest Quake1 score on a socket 3 system that I have seen around.
Keep in mind that the computer runs with full gear and everything.
Cache size is 1024kb. See speedsys screenshot above. On page 6 i believe is a post with details about the PC setup.
Quake's screen resolution is the default one - 320x200, as it comes with Phill's benchmark package.

@feipoa
That IS detailed.
Thank you.

Ok, I was going to run few 486 boards with pod100 in the coming days, anyway.
Will spend the necessary time on the UUD one. Hope I make it work.

Will send you some stuff and instructions how to run the tests.
It will be really interesting to me if they pass on your setup.

Can you clarify if you were using EDO or FPM?

Fpm.
Tried multiple sets of trusted modules. Didnt observe any difference.
I remember testing with 8, 16, 24, 32, 64 Mb

What version of the MB-8433UUD are you using? I've had better luck with the later v2 (9620+ datecodes on chipsets), v3, and v3.1 boads. Ensure your cache isn't mixed mode. And use a PCI HDD controller card and no CF. I've had issues with CF on this board, but I don't recall all the particulars anymore. Also document the exact failure modes and I'll see if I can find some time [and space!] to reproduce them, likely after the 5 Linux systems are setup. 2 down, 3 to go.

picture

great tip about the ide controller. forgot to comment on that in previous post.
i have a note here to look for drivers for the integrated one but will use external as you suggested.
so far it was solid with any Hdds and CFs, but if you experienced problems with it before i may be hitting hidden issues there as well.

Here is the second of the examined OR800 motherboards.

OR820 as Dell Optiplex G300

Initially it supported 128Mb RIMMs only.
BIOS update allowed up to 2x512Mb, as well as added Tualatin support.
Rudimentary BIOS - 1 page with the basics.
Clock generator is limited to 133MHz. No chance for overclocking.

The board came without VRM, so only one CPU could be used.
Finding the right VRM required a bit of investigative work. Lonely post in a long gone Dell support forum provided the hint.
With the VRM module - two Coppermine CPUs can be used, but here is the bummer -> 2x Tualatins = no go.
Boot process stops with "No timer tick interrupt." error message.
Disappointing.

On-board (3Com) LAN and sound.
Drivers are still available and worked fine right away.

Non-standard CPU fan plugs.

Requires custom PSU.

Was not able to get SpeedSys screenshot.
Hangs at "Get DMI info".

Coretest reported 68756Mb/s buffered read speed with Promise ATA-100 adapter and fast CF card.
Pretty good.

Comparing this OR820 to OR840 (from the above post) and BX440 (Asus P2B-D from the previous page) using 1000/133 Coppermine CPUs.
Added P2B-D 1575/150 with Tualatins indicating max performance.
Some of the tests are performed with Quadro2 Pro, others with Quadro FX4000. These two cards deliver best perf from all examined PRO-grade AGP adapters. The one that did best in a given test was added to the chart below.
All tests performed on the same Windows NT4 SP6a as specified in previous posts.

Picture is pretty clear.
OR820 is equal in performance to OR840.
Both are clock-to-clock slower than BX440.
Don't stand a chance against overclocked BX440.

This motherboard is very Dell inside-out.
Too many quirks.
Trouble if you want more than the basics.

@Feipoa
Your comments about Biostar MB-8433UUD-A helped me to sort out that nagging problem related to L2 cache instabilities.
It took quite an effort, but once resolved, the board became even better than before. Updated all related posts in the thread with perf numbers and other details accordingly.
There is still one last remaining question - is there a driver for the on-board EIDE controller ?
By default it is kind of slow. But we know that standalone UMC based EIDE controllers from that time period are actually pretty good, given there is a driver loaded.
Did a brief search online, but nothing conclusive came up.

pshipkov wrote on 2019-11-04, 07:19:

A special guest on this weekend's matinee was MS-4144.

Without further ado - i expected a lot from this motherboard based on its impressive specs (256Mb RAM, 1024Kb SRAM), but got disappointed. The system is unstable when timings are too tight. Increasing them to keep things stable leads to worsened performance. Tried different sets of RAM and SRAM chips to eliminate the chance of incompatibility, but the problem seems to be the mobo itself. It is kind of strange, because the same SIS chipset works much better in other boards.

As a brief conclusion - a nice piece of hardware, extra RAM is a plus. Performance is not the best, but in the ballpark. There are better boards out there.

Did you also try those same tight timings with only 256K double-banked?

pshipkov wrote on 2021-03-14, 20:04:

@Feipoa Your comments about Biostar MB-8433UUD-A helped me to sort out that nagging problem related to L2 cache instabilities. I […]
Show full quote

@Feipoa
Your comments about Biostar MB-8433UUD-A helped me to sort out that nagging problem related to L2 cache instabilities.
It took quite an effort, but once resolved, the board became even better than before. Updated all related posts in the thread with perf numbers and other details accordingly.
There is still one last remaining question - is there a driver for the on-board EIDE controller ?
By default it is kind of slow. But we know that standalone UMC based EIDE controllers from that time period are actually pretty good, given there is a driver loaded.
Did a brief search online, but nothing conclusive came up.

Glad you go the Biostar working well. I get quite a few PMs concerning setting up the Biostar to run optimally and so far they all turn out successful.

I should have a collection of these IDE drivers for Win3.1 and/or Win95. I'll look for them for you when I continue my Linux installation battle. Next up is trying to get my no-name USB-to-Wifi drivers working on my garage/mechanic station. I didn't run ethernet to the garage and have been regretting it ever since.

Which OS drivers were you looking for?

EDIT: This is what I have (attached). Please note, I do not maintain archive file names (the *.rar or *.zip name). I rename them so I can actually find them, so for anyone else finding this, you may already have these archives but with a different naming convention.