3 (+3 more) retro battle stations

Topic actions

Reply 40 of 79, by pshipkov

User metadata
Rank Member

Glad you liked my gibberish.

4x1Mb Samsung 60ns parity FPM 30 pin SIMMs


Had to pick and choose from a bunch to find the right ones. As far as i remember 2 more memory sticks with LG chips on them worked fine too, but used the 4 Samsung ones for uniformity.

The BIOS does not offer option to toggle parity checking, but these memory sticks have been battle tested on multiple motherboards for all sort of things - they are solid.

Just remembered something - there is a similarly looking type of SIMMs with KM41C1000CJ-6 labeled Samsung chips on them, they gave me quite a bit of trouble. They are a bit taller and have wider resistors. There is a "SAMSUNG Korea" label in the upper left corner.
But maybe it is just a coincidence that mine are not great and i am only spreading rumors here.
So i leave it to you to decide what is what. 😀

Last edited by pshipkov on 2020-01-10, 07:03. Edited 3 times in total.

Reply 41 of 79, by Ekb

User metadata
Rank Newbie

nice 😀
i didn't know that samsung is good too 😀

if I'm not mistaken, then "Parity error check" is it J18, you need to install a jumper 1-2.
2-3 - OFF parity error check.

I see chips "Parity Error Check" are also high speed 60ns 😀

Reply 42 of 79, by pshipkov

User metadata
Rank Member

I tried that and got an error message.
Plus the system hangs in a weird way.

Tried the same memory sticks in two other motherboards - 286 and 386 - parity check went just fine.
Given the fact that the VLSI system is very stable, i am wondering if its parity checking is just being funky ?

Btw, do you have the datasheet of the motherboard ?
There are couple of other jumpers that seemingly do nothing.

Last edited by pshipkov on 2019-08-25, 00:13. Edited 1 time in total.

Reply 46 of 79, by pshipkov

User metadata
Rank Member

Thanks for the kind words.
It is a nice hobby to spend an hour or few with, when possible.

You are right - i didn't provide any numbers for the voodoo. Let me know if there is a specific benchmark, or a game, that you think will be a good fit and i can give it a try.


I want to righten something that was wrong, or better - something that was unjust.
Couple of posts above i labeled the Unichip 367C motherboard an underperformer.
The motherboard shown all symptoms of turbo turned-off - no L2 cache, very low performance, etc.
I was unable to wake it up, so i assumed that it is just a badly desgned piece of hardware.
Until yesterday.

Check this out:
- I noticed that one of the turbo-switch pins was little shorter. Upon close inspection i realized that it actually was broken and does not make contact with the corresponding line. Fixed it. Shortened the pins (turbo-on). Back in business ! Until ...
- Now there is an error message about bad cache.
- I know that the cache chips are good. They work fine in other motherboards. Whatever. Changed couple of sets, until 4xEtronTech 20ns ones did it. Weird. Finally ! Well, until i ...
- Turned-off auto-config and optimized timings = black screen.
- I was using CL-GD5426 and CL-GD5428 ISA video cards. Had to bring my preferred weapon of choice - ET4000AX (1998) when something is funky on that side.
- Back online. No, not yet ...
- The system is now very unstable. I know that the installed memory may not be the best out there. Swapped them with trusted ones. And finally that did it !
Quite a tall stack of problems that i think gives me a good excuse for my initial opinion about this motherboard.
Once that was cleared up, the motherboard turned out pretty nice.

Updated stats:

Performance comparison with M321 for reference:

Reply 47 of 79, by pshipkov

User metadata
Rank Member

Back in the day there were plenty of S3 Virge based video cards.
Spent a moment today to check the unaccelerated DOS and accelerated Windows GUI performance of 3 Virge generations (capped at 1996).
Added an S3 Trio for reference.
Used the ASUS PVI-486SP3 motherboard.

S3 Virge/GX 2Mb (Compaq)
https://www.petershipkov.com/temp/retro_pc_im … s3_virge_gx.jpg

S3 Virge/DX 2Mb (forgot to put the second Mb of memory for the photo session)
[url]http:s//www.petershipkov.com/temp/retro_pc_imag ... rge_dx.jpg[/url]

S3 Virge/VX 8Mb (STB Velocity 3D)
https://www.petershipkov.com/temp/retro_pc_im … s3_virge_vx.jpg

S3 Trio64V2/DX 2Mb
https://www.petershipkov.com/temp/retro_pc_im … dx5_s3_trio.jpg

https://www.petershipkov.com/temp/retro_pc_im … nchmarks_s3.png

As expected, DOS results are even, a bit of a difference in Windows GUI. Same old, same old.
Will try to run some accelerated 3D tests tomorrow to see if there is anything different on that side.
If i can trust my distant memories - the results will be close, but also dismal. 😀

Last edited by pshipkov on 2020-04-05, 04:36. Edited 3 times in total.

Reply 48 of 79, by pshipkov

User metadata
Rank Member

Ran a quick test with latest version of glQuake1 against the 3 Virge cards using the GL library provided by S3.
Used Voodoo 1 as a reference, but had to set "cache burrst write" to 3 cycles (lowest possible is 2), otherwise Voodoo and glQuake don't cooperate. This was not necessary for the Virges.
I expected VX to be the slowest, but surprisingly it does better than the rest. Still the Virge line is clearly a 3D decelerator type of hardware.
On top of that, picture was dithered and the GX card produced additional texture stretching artefacts.

Reply 49 of 79, by pshipkov

User metadata
Rank Member

In continuation to my post on Page 2 about 486 motherboards from the time period 1995-1996, i am adding one more to the list - Biostar MB-8433UUD-A rev.2

http://www.petershipkov.com/temp/retro_pc_ima … motherboard.jpg

The motherboard came with the usual dead Dallas battery, which required unsoldering and socketing.
The cache chips were faulty, so fresh high-quality ones had to take their place.
Memory was problematic too, so high-quality modules had to replace them as well.
Adjusted BIOS timings to the lowest possible wait states for best performance.

Below is what the standard set of DOS syntethic benchmark programs report.

Biostar MB-8433UUD-A with AMD-X5-133ADZ 160MHz

Biostar MB-8433UUD-A with Intel Pentium P24T 83MHz

Used the Asus PVI-486SP3 rev. 1.22 as a reference point.
http://www.petershipkov.com/temp/retro_pc_ima … motherboard.jpg

Asus PVI-486SP3 with AMD-X5-133ADZ 160MHz

ASUS PVI-486SP3 with Intel Pentium P24T 83MHz

Numbers are more or less the same. The only noticeable difference is in IDE performance:
PVI achieves 9-10Mb/s while Biostar lingers at 3-5Mb/s.

Synthetic tests are ok, but they don't show what really is going on, so i ran the usual mixture of unaccellerated DOS and accellerated Windows graphics, as well as 3D rendering tests ...

Biostar kicked PVI's butt in PCI video graphics. Tested with Matrox Millenium, Tseng Labs ET6000 and Cirrus Logic GD-5480, results were consistent, so only the Matrox numbers are shown below for compactness.
But the ASUS board had the final word on graphics performance with the monster Trident TGUI9440AGI VESA card.

The Biostar board ran into trouble with the Windows accellerated graphics and 3D rendering tests using the 486DX5 CPU.
I had to significantly relax the wait states to get it to stable condition, otherwise Windows was hanging, 3D Studio cannot get to UI and LightWave3D crashed during rendering.

For Windows and 3D Studio:
L2 cache wait states 3-2-2-2
host clock / pci clock 1:1

For LightWave3D:
read/write wait states 1 1
L2 cache wait states 3-2-2-2
host clock / pci clock 1:2/3

Synthetic tests taken with the LightWave3D wait states for the 486DX5 CPU for fully stable system:

I was conflicted about the situation.
In the one hand - i really like the great DOS performance of the Biostar board despite the fact that the wait states used to achieve it lead to unstable system for Windows and 3D rendering, but in the other hand - i really like fully stable systems.
At the end i decided not to discard Biostar's great DOS results and provided two bars for the 486DX5 - one for fully stable system and another for the lowest wait states.

Ok, enough talk - let's take a look at the results.
Well, one last note - orange bar denotes best result, orange outline denotes best PCI result.


The Biostar board has better PCI graphics performance even with increased wait states necessary for fully stable system for the execution of more sensitive tasks. Its IDE performance is about half of what the Asus mobo does. The Asus board saved face with its support for VESA graphics which came on top at the end of the day. But the metrics difference is more or less marginal.

Last edited by pshipkov on 2019-12-09, 07:28. Edited 1 time in total.

Reply 50 of 79, by pshipkov

User metadata
Rank Member

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

https://www.petershipkov.com/temp/retro_pc_im … motherboard.jpg

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. Used the same Matrox Millenium PCI video card for the tests below.


The results (used Asus PVI-486SP3 rev 1.22 for comparison):

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.

Last edited by pshipkov on 2020-03-15, 08:21. Edited 1 time in total.

Reply 51 of 79, by pshipkov

User metadata
Rank Member

Another weekend - another motherboard examined - this time is the 386 FX-3000/CL rev 1.0.
Actually the examination happened a while ago, only sharing the findings today.

Motherboard is fully stable at 45MHz with all system timings at their lowest (fastest) available values. Also, fully maxed out system - 32Mb RAM, 256Mb SRAM, etc.
The mobo is fully stable at 50MHz with lowest timings, but without FPU. If FPU is present (a 40MHz rated FasMath, for example) some compromises need to be made for things to stay stable - relaxing some timings and enable caching for the first 16Mb RAM only. Obviously this is a no-go for me, so i don't consider it as an option.
The board works at 55Mhz as well, but timings have to be relaxed which leads to lower performance than 50MHz and even 45MHz.

Surprisingly at 50MHz the Adaptec AHA-1542CF SCSI controller can operate at 10Mb/s which is a real boost for the disk performance. You really feel the 5Mb/s read speed ...
At 45MHz i can get it to work at 6.7Mb/s max.

FX-3000 didn't like 12ns ISSI SRAM chips. Not sure why, because they work just fine in other systems, so i went with 15ns UMC ones, which nicely match the chipset brand of the board.

All tests were performed with the same Genoa CirrusLogic GD-5426 video card with latest (for 1993) Diamond BIOS (great performance), that i use in my main 386 build.

https://www.petershipkov.com/temp/retro_pc_im … motherboard.jpg

Syntethic tests at 45MHz

Syntethic tests at 50MHz

The numbers that matter (compared to PC-Chips M321):

Reality is that FX-3000 is a bit slower (clock-to-clock) than PC-Chips M321 (which seems to be one of the fastest 386 motherboards) at the same clock speed, but PC-Chips has one deficiency operating at 45MHz. As i mentioned in my initial post here - between restarts, it needs some cool-down time - 30 sec to a minute. This compromises my requirement for a fully stable system, so i ended-up replacing it with FX-3000 in my main 386 build.
Also, if one does not care about esoteric stuff like 3D rendering, the motherboard can operate really well at 50MHz without FPU, which is a total pass for all games and other standard activities performed on a 386 machine. Obviously at 50MHz FX-3000 is faster than 45Mhz M321 (which cannot go past that).

Last edited by pshipkov on 2020-03-15, 08:28. Edited 2 times in total.

Reply 52 of 79, by pshipkov

User metadata
Rank Member

Finally got some time to get back to the retro hobby.

Played a bit with this QD-U386386DX ver.1 motherboard based on UM82C482AF/UM82C481BF chipset.

https://www.petershipkov.com/temp/retro_pc_im … motherboard.jpg

The mobo looked very promising at first - clean and well organized layout. DIP14 crystal oscillator for the ISA bus ...
But then came the disappointments.

It was very picky about the SRAM chips. Took me a while to get it to 256Kb.
Very slow HDD I/O and below average VGA performance.
Swapping the ISA oscillator with 16MHz one does not result in measurable difference.
The board does not overclock at all - 40MHz is the max.

For some reason the screen grabber was unable to capture any screens other than SpeedSys and LandMark6. Provided below.

Here is the standard set of benchmarks i use. Didn't bother with 3D rendering - "picture" is pretty clear.

Succeeded to get the mobo running at 45MHz but without FPU.
Unstable in Windows 3.11. Couldn't complete the WinTune2 tests.
Numbers are actually pretty ok, but in general - without FPU and unstable Windows - system is compromised.

Expected more, but still, it was fun messing around with it.

Last edited by pshipkov on 2020-03-13, 01:21. Edited 2 times in total.

Reply 53 of 79, by pshipkov

User metadata
Rank Member

Octeck Jaguar V rev.1.3 - a 386 motherboard with potential, hampered by wacky Mr.Bios.

https://www.petershipkov.com/temp/retro_pc_im … motherboard.jpg

An interesting board.
On par (clock-to-clock) with some of the best performers out there.
There is only 8Kb cache installed with no options for upgrade. This hurts memory intensive tasks.
Very stable, takes 100MHz crystal oscillator with ease.
The hardware is clearly pretty capable, but the installed Mr.BIOS lacks any sophistication, other than 2-3 settings in the "chipset" section. Really annoying. Will probably give it another spin with proper bios at some point later.
For now - here is what the board does at 40MHz, tested with ET4000AX 1Mb video card:



And below is how things look at 50MHz, again with ET4000AX 1Mb video card.
The WinTune 2 comparison is not really fair between the AX and W32i ET4000 cards, so take it for what it is.
Had to lower the "system wait state" setting in the BIOS to be able to boot. Notice the lowered IDE performance.



Not a bad piece of hardware, the 8Kb of cache is a limiting factor, very snappy - especially between restarts - it take a second only to get to DOS prompt.
Didn't bother with 3D rendering - it takes a lot of time and won't reveal anything surprising for this motherboard.

Last edited by pshipkov on 2020-02-18, 09:40. Edited 4 times in total.

Reply 54 of 79, by pshipkov

User metadata
Rank Member

I have been searching for the right 386 motherboard for quite some time.
Went through a lot of them (partially documented in this thread), but never found one that fully satisfies my requirements for speed and stability.
Well, until now.

Symphony 'Haydn' SL82C461/SL82C362 chipset
256Kb cache, 12ns
32Mb RAM, 60ns

This motherboard is a real beast.

(with naked CPU)
https://www.petershipkov.com/temp/retro_pc_im … motherboard.jpg

1. The fastest (clock-to-clock) 386 hardware i have seen so far.
2. Overclocks to 55MHz without FPU and more importantly - overclocks to 50MHz with FPU with the tightest BIOS timings (* small exception for 3D rendering - noted below).
- A 40MHz FasMath FPU core and 12ns SRAM are needed to operate at this frequency.
- The slightly faster (clock-to-clock) 33MHz marked FastMath FPU core hangs the system.
3. Impressive IDE performance (3.5Mb/s) and even more impressive SCSI performance in 10Mb/s mode (5.2Mb/s read and 3.2Mb/s write).
4. Fully maxed-out system (ram, cache, fpu, scsi, sound, lan, etc.).
5. Very stable !

And the best part yet:
Usually getting 386 hardware to perform at 45Mhz or higher involves significant effort.
Not this time - the thing just works.

In addition to that:
Clean and easy to work with layout.
Not a fan of the RTC Dallas modules, but once socketed they are ok.
There is something to be said about the SIMM RAM sockets - the metal clips are so convenient and with just the right grip, so i caught myself messing around with them more than i should, it's kind of addictive in a retro-nerdy-way. Whoever built that board, really put the extra effort back then.

Few comments about the stats and tests:
Until now i relied on CL-5426 based Genoa video card with Diamond BIOS for my 386 activities, but it refused to operate reliably at 50MHz FSB. So i switched to ET4000/W32i for good. It is on par with CL-5426 and ET4000AX in DOS, but does so much better in Windows. At 24bit color it approaches the performance of Artist Graphics's WinSprint accelerator, which operates at 8bit colors only (limited by it's Win3.1 driver - i am playing period correct at all times, so everything has to be that way - for 386 - year 1993 is the upper limit for hardware and software).
While 3D Studio works fine with the tightest timings, for LightWave3D the DRAM wait state has to be increased to 2.
Still, for those who are not into retro 3D rendering (who really is ?), the motherboard will operate completely stable at 50MHz with the tightest BIOS timings at all times.


The "DOOM" 386 build (running at 55MHz with no FPU) showcased in this thread (DooM) achieves 39.5fps and 9.7fps in the Doom test. This mobo outdoes it at 50MHz. Will try to post results for 55MHz and no FPU at some point later.
3D rendering performance is compared against FX-3000 running at 45MHz, which was the best (fully stable OC) i got with 386 hardware until now.

As a side note:
I happen to have another version of this motherboard: DTK-PEM-4036YB (notice the B at the end).
It has proprietary, unusual and terrible DTK BIOS.
Does not overclock *at all*.
The board layout and chip types suggest a later version, but the keyboard controller is older than the PEM-4036Y.
The serial numbers of the Symphony chips are different between the two mobos.
The SIMM slots are awful - pretty much impossible to insert or remove memory sticks (this actually prompted me to notice how nice the memory slots on the other board are).
So not sure what to make out of it.

2 years ago i wanted to build clean 286, 386 and 486 class computers that i am fully satisfied with (as mentioned in the first post of this thread).
The 286 and 486 ones got there relatively quickly with the VLSI and ASUS PVI motherboards.
My 386 effort started with PC-Chips M321 which is impressive board by itself, but didn't overclock well.
Several boards later i was still searching for the right one until stumbled upon this Symphony Haydn guy.
The 386 rig is shiny and sparkly now ... and my 386 journey seems to be finally coming to an end.


Last edited by pshipkov on 2020-04-05, 04:13. Edited 8 times in total.

Reply 55 of 79, by pshipkov

User metadata
Rank Member

Another unnamed 286 VLSI based motherboard. Obviously a late model from '90-'91.
Very fast - runs at 25Mhz, but very picky about RAM quality at this frequency.
Does not cooperate at higher speeds, because it does not provide control over wait states.

The board came with SIPP slots and 4 Mb of 80ns RAM.
I was getting ready for an hour or few of soldering work to swap the SIPP with SIMM slots, but then decided to spin the mobo for a round of early tests and inserted the SIMM slots into the SIPP ones. They snugged so tightly, so i decided to keep them that way.
Soldered fresh battery, crystal oscillator socket and was ready to roll.

The motherboard with 20Mhz Harris CPU, 12Mhz IIT FPU and 4Mb 60ns RAM.
https://www.petershipkov.com/temp/retro_pc_im … motherboard.jpg

The tests were performed with 25MHz Intersil CPU, 20MHz IIT FPU, 4 Mb 60ns RAM and 0-wait-state ET4000AX VGA with 1Mb RAM.


Metrics are a bit lower than the 286 VLSI motherboard from the first page of this thread.
BIOS seems to be exactly the same, so there are some minor differences in the hardware setup.
For some reason NSSI didn't want to work on this system - reports memory corruption - which may, or may not be indicative of potential instability.
Other than that - everything worked as advertised, as far as i can tell.

Brief comparison with the 286 VLSI motherboard from the first page of this thread:

Simply great 286 motherboard - one of the fastest out there.

Last edited by pshipkov on 2020-02-18, 09:32. Edited 1 time in total.

Reply 56 of 79, by pshipkov

User metadata
Rank Member

386 Micronics TK-82C491/386-4N-D02C motherboard.
Late model - probably '93 or even '94.
These small format boards are work horses - very stable.
Most of them are not the best performers out there, but this one is actually pretty good.

The board is very unpretentious - takes whatever RAM, SRAM, IDE, VGA get thrown at it.
BIOS is chock-full of options. Things are stable with the lowest possible wait states.
There are options for DLC - verified - works great with CPU upgrades.
These late models don't use crystal oscillators, which makes it pretty much impossible to overclock.
VGA performance is great, but a bit slower than the Symphony mobo.
What really surprised me was the 3D rendering performance - it is the fastest clock-to-clock one i have seen so far.

https://www.petershipkov.com/temp/retro_pc_im … motherboard.jpg

Here is the usual set of stats:

And the usual set of tests:
There is something wrong with the LW3D rendering test of the DTK motherboard at 40MHz - will need to take a look later.

Simply great motherboard- fast and reliable.
Until now i had the fantasy that mobos with 128Kb cache are inferior to 256Kb ones - especially for memory intensive operations, but i was proven wrong - looking at the 3D rendering perf here.
The board can clearly do more, but the installed clock generator limits it to 40MHz - too bad.

Last edited by pshipkov on 2020-04-12, 19:11. Edited 1 time in total.

Reply 57 of 79, by pshipkov

User metadata
Rank Member

I have captured data for few more 386 motherboards.
Will post it here over time for completeness.
So, two more guys in this post: ALI-1419 and ALI-1429.


Everything started great - pretty agile and nimble mobo. Overclocked to 45Mhz with no problem.
Performed faster than FX-3000 in the basic DOS tests, but then completely failed WinTune2 and the 3D rendering.

https://www.petershipkov.com/temp/retro_pc_im … 6_ali_m1419.jpg

And the usual set of stats and benchmarks:



Obviously very late 386 motherboard. Probably 1994. Very stable. Works as advertized. 128Kb cache only.
Does not use crystal oscillator and i don't feel like figuring out if using another clock generator is possible, so capped at 40MHz.

https://www.petershipkov.com/temp/retro_pc_im … 6_ali_m1429.jpg

The usual set of stats and benchmarks:


Didn't want to wait the 2+ hours for the LW3D test. It wouldn't reveal anything surprising - hardware is stable and not very fast.

All in all - average 386 motherboards. Stable, predictable.

Last edited by pshipkov on 2020-03-21, 02:25. Edited 1 time in total.

Reply 58 of 79, by pshipkov

User metadata
Rank Member

Sharing info about two more 386 motherboards.

Biostar MB-1333/40AEA-Q

Neither good or bad board - from the middle of the pack.
Does not overclock at all. Pretty stable. Does what is supposed to do.
Trying to think of anything unique about it but - nothing comes to mind.

https://www.petershipkov.com/temp/retro_pc_im … 333_40aea_q.jpg

The usual set of stats:

Biostar MB-1340UCQ

Another average representative of the very late 386 class hardware.
Works as advertised.

https://www.petershipkov.com/temp/retro_pc_im … _mb-1340ucq.jpg

The usual set of stats:

The usual set of benchmarks. Didn't run LW3D rendering - takes too long, won't show anything surprising - hardware is stable and performs allright.

Not bad motherboards, very stable, but there are clearly better options out there.

Last edited by pshipkov on 2020-05-04, 06:10. Edited 2 times in total.

Reply 59 of 79, by pshipkov

User metadata
Rank Member

DTK PTM-1632C is an interesting 286 motherboard.

It allows to combine SIMMs along with on-board memory for maximum amount of 5 Mb RAM.
Cannot remember seeing this in other boards.

BIOS is custom DTK software that was super advanced for 1990 - menus and stuff.
It is very similar to the one of the 386 DTK Symphony boards i stumbled upon not far ago.

The board overclocks pretty well to 27.5 MHz, but requires good quality SIMMs to get there.
With on-board memory chips i was not able to get beyond the 20 MHz. Tried few different 70ms and 80ms sets - but no-go.
So i had to choose between 5 Mb of RAM at 20 MHz, or 4 Mb at 27.5 MHz.
4 Mb of RAM is plenty for 286 hardware, so i went with the higher frequency.

One of the SIMM sockets had its clips chipped-off - replaced it with fresh one.
Didn't have black colored socket, so this section is not color coordinated, but who cares.

https://www.petershipkov.com/temp/retro_pc_im … motherboard.jpg

The usual set of stats and benchmarks:



An interesting board, considering its BIOS and memory options.
Overclocks pretty well.
There is no control over wait-states.
Not the fastest piece of 286 hardware.

Updated the first post of this thread with information about two VLB video cards - Diamond Stealth 64 (S3 Trio64) and Paradise (WD90C33-ZZ).

Last edited by pshipkov on 2020-04-05, 04:19. Edited 2 times in total.