3 (+3 more) retro battle stations \ VOGONS

Getting them up and running was quite a journey - going back through my memories, identifying and finding components, restoring them, putting everything together, breaking and fixing stuff (nested loops).
During my research phase i realized that many of us often mix and match components from different generations to achieve best/worst performance, better stability, other more obscure goals, or because that's what we had on the shelf at the time.
Instead, i wanted to stay true to the given platform and its time period. So, i avoided any DLC/POD upgrades and made sure that no piece of hardware/software is from later year than the one marked on the calendar.
Well, with one exception - after couple of setbacks with failing old HDDs, i ended up substituting them with CF cards. For now ...

What else ?

Any overclocking and other tweaks had to be long-term stable. End goal was to build computers that work for real.

No ATI video cards.
Looking back, i realized that somehow i never owned ATI hardware.
Well, except for a FirePro v7900 couple of years ago, but it was donated by ATI, so it kind of does not count. Wonderful card btw !
To keep the tradition alive, i passed on Wonder/Edge-16, Mach32 and 3D Rage II for the 3 rigs, respectively.

No Pro 3D accelerators.
I guess my choice of timings was less than ideal considering the situation with such cards back then.
In 1996 - Intergraph, 3DLabs, Dynamic Pictures, etc., were targeting Pentium/WinNT systems with their second generation 3D accelerators, but not a 486/Win95 one. They didn't even have proper (or any) drivers for Win95. At some point i considered strapping GLINT 500TX/DELTA, or Oxygen, but then common sense prevailed and i went with cards that are more natural fit for the system.
In 1993 - Everything interesting related to 3D accelerators for the PC was happening on the first Pentium/PCI systems, but not a 386/ISA one.
In 1990 - What professional PC 3D cards ? The low/mid-end pro graphics was Amiga territory. There were handful of CADs and pretty much no graphics software for the 286.

SoundBlaster fanboy ?
Not really, just being practical with the sound setup, with focus on compatibility.
Never owned a GUS for example. Curiosity is significant, but prices are chilling. 😀

Computer cases:
I didn't want to hide the carefully selected and assembled hardware inside standard PC cases.
Instead, i modified 3 ATX acrylic ones to fit the AT components. This took a lot more work than expected, but the result was worth it, i think.

286 from 1990

Some of the best 286 designs emerged around that time. The platform was in its prime, but with 386 rapidly growing in popularity and the first 486 systems introduced to market the year before - writing was on the wall already.
Still, highly integrated 286 chipsets kept showing in the following 2-3 years, but they didn't introduce anything more interesting than compact form factors.

Piece of sweet memory:
I clearly remember the moment when i saw for the first time VGA graphics on my 16MHz 286.
Shaky fingers swap old CGA card and monitor with fresh VGA ones (Trident + 14" 0.36 dpi CRT). PC is on, first game shows-up on screen. The young individual in front of it is deeply moved.
Today people think that Apple keynotes are magical.
Sure. 😜

HARDWARE

motherboard: Dunson Electronics 286-16 (VLSI VL82CPCAT-16QC)
cpu: Intersil CS80C286-25 25/30MHz
fpu: IIT 2C87-20 22MHz
ram: 4Mb 60ns parity FPM
vga: Diamond SpeedSTAR 24 1Mb (Tseng Labs ET4000AX, 24-bit DAC) (primary)
- Diamond Flower VG-1000 rev 2 512Kb (Cirrus Logic CL-GD510/520)
  Trident TVGA8800CS rev 2.0 512Kb
  Western Digital WD90C00-JK 512Kb
  OTI-037 512Kb
  Ahead V5000-50PC-B 1Mb
  Headland GC208-PC 1Mb
i/o: UMC 82C862F
audio: Sound Blaster 1.5 CT1320C with C/MS
lan: Tiara Lancard/E-PC 16
psu: 300W AT
input: BTC 5121, MS Mouse
os: DOS 4.01 + Windows 3.0
partitions: 1x504Mb

NOTES

The system works really well at 25MHz with 0 wait states and 30MHz with increased wait states, but it took quite an effort to "climb that mountain". For example, finding the right memory modules was a memorable experience by itself.

No SCSI ?
The top SCSI controller at the time AHA-1542B is actually slower than what this motherboard does with IDE.
SCSI's max data transfer rate is below 2.8Mb/s (all settings on max), while the IDE goes way over 3Mb/s.

[EDIT] The story evolved with:
Improved ventilation, complete long-term stability: Re: 3 (+3 more) retro battle stations
and 27.5MHz + Cyrix FPU FTW: Re: 3 (+3 more) retro battle stations, probably the fastest 286 class PC around here.
Upgrade with TI 486SXL2-50 @66MHz CPU for extra cool (or weird).
And the latest development - evolution to 30MHz and 0 wait states.

THE CULPRITS

286 at 25MHz

286 at 30MHz

386 from 1993

Until then a good 386 system was able to hold its ground (well, more or less) against an average 486SX/DX one, especially if one considers the price-performance ratio, but the newer 486DX2 designs from 1992 changed that for good. Around that time 486 became the mainstream and the first Pentium CPUs were introduced to market. This indicated the end of the platform.

Another melodramatic insert:
Summer job paid for a shiny Sound Blaster 2.
Bought it. Back to the bus stop. Every minute is eternity. Bus finally coming.
Back home. Installing. Connecting the sound system with beefy speakers. Starting Dune 2. Watching the intro.
Transformative experience.

HARDWARE

motherboard: PC Chips M321 rev 2.3, 256Kb 12ns L2 cache (* read update notes below)
cpu: AMD Am386DX/DXL-40 40/45MHz
fpu: Cyrix FasMath CX-83D87-33-GP 40/45MHz
ram: 32Mb 60ns parity FPM
vga: Genoa Systems Windows VGA 24 8500 1Mb (Cirrus Logic GD-5426) with Diamond Speedstar BIOS (primary)
- Cirrus Logic GD-5420 512Kb
  Tseng Labs ET4000/W32i 2Mb
  Tseng Labs ET4000AX 1Mb
  Artist Graphics WinSprint 1000i (5075) 2Mb + 8Mb 60ns for applications / Chips F82C450D 512Kb
  Trident TVGA9000C 512Kb
  Trident TVGA8900CL-C 1Mb
  Realtek RTG3105 512Kb
  ELSA XHR Winner 1000 2Mb (S3 928)
  UM85C408AF (512Kb)
  WDC90C30-LR (1Mb)
i/o: Adaptec AHA-1542CF
- PTI-227 (W83757F)
audio: Sound Blaster 16 CT1750 v4.05 (* read update notes below)
lan: Intel Ethernet 16-Bit LAN Adapter FA82595TX
psu: 300W AT
input: BTC 5121, MS Mouse 2.0A
os: DOS 6.2 + Windows 3.11
partitions: 1x504Mb

NOTES

Speedsys gives lower score for 386 CPUs if FPU is present. All other tests indicate consistent CPU performance.

Increasing the L2 cache to 256Kb and adding WB chip improves performance by couple of %.

The system works really well at both 40Mhz and 45MHz with 0 wait states.
When at 45MHz it requires "cool-down" time between cold restarts. Significant effort went into identifying the source of the problem. At the end, it turned out to be the cache chips being very sensitive to heat. Will slap active cooling on them at some point soon.

Cyrix's FasMath CX-83D87-40-GP (black top) is actually slower than CX-83D87-33-GP (gray top) by 2-3%. This is confirmed by the rendering tests.
A hint in this thread A brief comparison of 386 FPUs sparked my interest to compare the two more carefully.

S3 928 does not cooperate when the CPU is running at 45MHz.
Also, despite the 2Mb of installed memory, the driver maxes out at 800x600x8bit in Windows 3.1x, which is disappointing, because the Vibrant driver tells different story.

A lot more disappointing was the fact that WinSprint's Windows 3.1 driver does not support High and True colors either. Real shame for such an amazing GUI accelerator.

*UPDATE 2020-01-26:
The PC-Chips M321 motherboard has been replaced first with Forex FX-3000 and then with DTK-PEM-4036Y - probably the fastest clean 386 class PC around here.
*UPDATE 2020-05-03:
While the initially used Sound Blaster 16 CT1750 v4.05 is period correct for year 1993, it just didn't feel right for 386 class computer, so Sound Blaster Pro 2 CT1600 took its place. It is a more natural fit for this PC.

THE CULPRITS

386 at 40MHz

386 at 45MHz

486DX5 from 1996

486DX5 was introduced at the end of 1995.
Pentium took-off for real in the mid 90-ies.
Some of the best 486 designs emerged around that time, but also, these were the last golden days of the platform.

Last one:
From early age i was feeling artsy ... or something, so i had this urge to visualize my ideas. This was bad for the trees.
Once i got into computers, the understanding of what they can do in regards to art quickly trickled-down my head and in the mid '90-ies i found myself with a massive interest in computer graphics and software development.
Of course, the results i produced at the time were not exactly impressive, so the only real measurable effect was my desire for powerful hardware.
That second part was unpopular at home and the 486DX5 had to be the popular hardware of choice in my room.

HARDWARE

motherboard: ASUS PVI-486SP3 rev 1.22, 512Kb 10ns L2 cache
cpu: AMD-X5-133ADZ 160MHz (primary)
- AM486DX5-133V16BGC 160Mhz (1999)
  Intel A80486DX4-100 100MHz
ram: 128Mb 60ns parity FPM
vga: Matrox Millennium 590-05 REV B (Matrox MGA-2064W-R3), 8Mb (with Viking memory expansion card), or 4Mb (with Phillips 7110A video capture card)
3d: Diamond Monster 3D (3DFX Voodoo), 4Mb
audio: Sound Blaster AWE32 CT3980 (OPL3 MIDI) 32Mb
lan: 3Com 10/100 3C905-TX rev A
psu: 450W AT
input: BTC 5121, MS IntelliMouse 1.1A
os: DOS 6.22, Windows 95 OSR2
partitions: 4x504Mb

NOTES

My favorite motherboard.
It smells and feels 486 (socket 3, VESA, FPM, PIO 4), mixed with the right set of "next-gen" tech (PCI, PnP, on-board IDE), all wrapped in a fast and reliable package that just works.
This was the product that made me notice the ASUS brand back then - a gift for them that keeps giving.

Increasing the Level 2 cache size to 512Kb (from the default 256Kb) improves performance by 1-3%.
For example, Quake 1 went from 16.9 to 17.3 FPS.

The system works so reliably at 160MHz with the AMD CPU, that i don't even consider it being overclocked.

It looks like the 1996 ADZ version of the AMD chip is a tiny bit slower than the BGC one from 1999.
For example, Speedsys shows score difference of 0.04 (59.94 vs 59.98). This is confirmed by the rendering tests as well.

The Intel CPU running at 120MHz completes all DOS games and other benchmarks, but is unable to finish, or sometimes even start, any 3D rendering tests.

Why no SCSI ?
Not convinced that the SCSI will outdo the current IDE performance of about 10 Mb/s. Will verify at some point later, but for now is low priority.

486DX5 at 160MHz

The board has clock generator that can go past 50MHz, but the system is unstable at 4x50 MHz.
Used trusted components that handle these frequencies just fine with other motherboards.

BENCHMARKS

As you noticed, each PC has multiple video cards in its specs.
I wanted to see how they stack against each other.

286 30MHz

286 25MHz

386DX 40MHz

386DX 45MHz

486DX5 160MHz

286 30MHz

286 25MHz

386DX 40MHz

386DX 45MHz

486DX5 160MHz

286 30MHz

286 25MHz

386DX 40MHz

386DX 45MHz

486DX5 160MHz

386DX 40MHz

386DX 45MHz

486DX5 160MHz

386DX 40MHz

386DX 45MHz

486DX5 160MHz

386DX 40MHz

386DX 45MHz

486DX5 160MHz

RENDERING

LightWave3D

3D Studio

Chaos

Last edited by pshipkov on 2025-05-12, 05:14. Edited 361 times in total.

retro bits and bytes

Reply 1 of 2374, by Almoststew1990

Posted on 2019-02-18, 13:05

Almoststew1990 Offline

Rank Oldbie

Rank: Oldbie
Posts: 880
Joined: 2017-03-14, 19:48
Location: Southampton, UK

What a fantastic case! I wouldn't mind owning a transparent case but it would show off my untidy builds 😁

Your 386 is sort of the opposite to mine, mine is an AMD 40MHz SX chip, soldered onto the board, no cache, 4MB Ram... It can barely run Wolfenstein 3D 😁 I do have a ct3600 sound card though.

Reply 2 of 2374, by keropi

Posted on 2019-02-18, 13:23

keropi Offline

Rank l33t++

Rank: l33t++
Posts: 9660
Joined: 2003-09-08, 06:45
Location: Greece

excellent first post and builds! the 286/25 is really uncommon 😎
kudos!

🎵 🎧 ⏳ ⚡ MK1869, PCMIDI MPU , OrpheusII , Action Rewind , Megacard and 🎶GoldLib soundcard website

Reply 3 of 2374, by oeuvre

Posted on 2019-02-18, 22:53

oeuvre Offline

Rank l33t

Rank: l33t
Posts: 3231
Joined: 2015-03-31, 13:31
Location: USA

woah that's one of the most impressive first posts

HP Z420 Workstation Intel Xeon E5-1620, 32GB, RADEON HD7850 2GB, SSD + HD, XP/7

Reply 4 of 2374, by pshipkov

Posted on 2019-02-20, 00:43

pshipkov Offline

Rank l33t

Rank: l33t
Posts: 2200
Joined: 2018-10-11, 05:08

Thank you for the positive feedback guys.

Yes, the 286 turned out be a beast.
Making "ultimate" PCs was never the goal, but i may snowballed a bit in that direction along the way. 😀

@almoststew1990
I hear you.
I have few more rigs laying around and some of them are baffling slow for their class.

retro bits and bytes

Reply 5 of 2374, by rasz_pl

Posted on 2019-02-20, 05:19

rasz_pl Offline

Rank l33t

Rank: l33t
Posts: 4208
Joined: 2017-06-04, 00:57

something weird with those Doom realticks, I had a (mis)fortune of playing Doom on 386dx40 back in the day, and there is no way it ran faster than 10-15 fps at low details. Did you mean post stamp sized window and low details?

https://github.com/raszpl/FIC-486-GAC-2-Cache-Module for AT&T Globalyst
https://github.com/raszpl/386RC-16 memory board
https://github.com/raszpl/440BX Reference Design adapted to Kicad
https://github.com/raszpl/Zenith_ZBIOS MFM-300 Monitor

Reply 6 of 2374, by snorg

Posted on 2019-02-20, 16:24

snorg Offline

Rank Oldbie

Rank: Oldbie
Posts: 1029
Joined: 2011-05-29, 15:32

OP, really like your builds, especially the 286. That is a real beast of a 286. Where did you get the acrylic case? I've been thinking about doing a home-build case but with ABS instead of acrylic. Looks really sharp, though.

Reply 7 of 2374, by Intel486dx33

Posted on 2019-02-20, 17:58

Intel486dx33 Offline

Rank l33t++

Rank: l33t++
Posts: 5344
Joined: 2018-05-17, 01:17
Location: U.S.A.

Nice way to show case your hardware.
Is there any advantage to owning these 3 types of computers or can you do with just one ?
If so , Which is your favorite and why ?

Reply 8 of 2374, by pshipkov

Posted on 2019-02-20, 23:17

pshipkov Offline

Rank l33t

Rank: l33t
Posts: 2200
Joined: 2018-10-11, 05:08

@rasz_pl
You are correct - min settings mean low detail and smallest possible playable screen area.

@snorg
Acrylic cases: https://www.aliexpress.com/item/QDIY-PC-D777X … .2c8a4c4dHXbBKK
The cases turned out well, but i had to order second batch of front/back panels. Let's say that my initial attempt to modify the cases produced underwhelming results. Luckilly the seller is flexible enough and can send you any part separatelly.

@Intel486dx33
If i have to be practical - there is no point of bothering with vintage hardware.
But we are humans and practicality is not one of our primary superpowers. 😁
For me it was about the excersise itself and remembering the good old days.
I like the 3 machines equally, but in a different way.
I was all over arcades and 8 bit XTs, but didn't have the understanding of this whole computing thing.
286 was the real entry point - very exciting experience where i started grasping things.
386 was the time where i started getting reaaaaly into it.
The DX5 i associate with the peak of a period of drastic personal evolution, where i transitioned from "consumer" to "creator".
This resonates strongly with me to present day, so i guess this rig is a bit more special.

Last edited by pshipkov on 2020-04-05, 04:28. Edited 2 times in total.

retro bits and bytes

Reply 9 of 2374, by rasz_pl

Posted on 2019-02-21, 01:55

rasz_pl Offline

Rank l33t

Rank: l33t
Posts: 4208
Joined: 2017-06-04, 00:57

pshipkov wrote:
@rasz_pl
You are correct - min settings mean low detail and smallest possible playable screen area.

so you tested all those different graphic cards by setting game up to output as little graphics as possible? 😀

Reply 10 of 2374, by pshipkov

Posted on 2019-02-21, 02:56

pshipkov Offline

Rank l33t

Rank: l33t
Posts: 2200
Joined: 2018-10-11, 05:08

The 386 was tested that way. The thinking behind it:
When the frame rate is too low, the measuring step is too big and we cannot register the more granular differences in performance between the video cards.
I can run and post the full screen low and high settings for doom for completeness, if you think that's missing.

The 486DX5 was tested with both low and high settings:
- the low settings were used to compare how the two class machines stack against each other
- the high settings are to show how the DX5 class hardware actually behaves

retro bits and bytes

Reply 11 of 2374, by rasz_pl

Posted on 2019-02-21, 04:30

rasz_pl Offline

Rank l33t

Rank: l33t
Posts: 4208
Joined: 2017-06-04, 00:57

pshipkov wrote:
The 386 was tested that way. The thinking behind it:
When the frame rate is too low, the measuring step is too big and we cannot register the more granular differences in performance between the video cards.

which would mean differences were meaningless at the time (~1992), this would be surprising, Id expect really terrible ones (rtg, trident 9000) to be significantly slower

pshipkov wrote:

I can run and post the full screen low and high settings for doom for completeness, if you think that's missing.

was there more than 1-2 fps between the cards?

Reply 12 of 2374, by pshipkov

Posted on 2019-02-21, 06:02

pshipkov Offline

Rank l33t

Rank: l33t
Posts: 2200
Joined: 2018-10-11, 05:08

Tbh cannot remember, it was a while ago, so let me run some bmarks and post the numbers here.

retro bits and bytes

Reply 13 of 2374, by mrau

Posted on 2019-02-23, 22:39

mrau Offline

Rank Oldbie

Rank: Oldbie
Posts: 1145
Joined: 2015-11-28, 12:43

very nice bench, that 286 is an intersting one especially

one thing that makes me wonder here - that matrox tool showing almost a second required to draw lines? how should this line be understood? i believe the matrox was rather snappy, which this number seems to contradict;

i also see You did test at respective max bitdepths - but some cards did slow down massively at 24bit due to alignment - would You know if that is the case for these cards?

Reply 14 of 2374, by pshipkov

Posted on 2019-02-24, 18:20

pshipkov Offline

Rank l33t

Rank: l33t
Posts: 2200
Joined: 2018-10-11, 05:08

The comments about the 286 rig made me look up benchmark results posted on the forums and i am realizing now that the other two builds are quite of performers as well. 😊

That's a WinTune 2 screenshot with detailed breakdown of the video card test.
The test is composed of multiple tasks and what you see there is the time it took to execute each one of them.
The heaviest task is drawing lines on screen. If you think that MM is slow there - you have to see the rest - double or tripple that.
I didn't want to litter the post with each and every bit of information captured from the tests, but i still have the screenshots somewhere around here and can post them eventually.

About the bitdepts:
My thinking was that if a video card is not able to hit at least 16-bit color depth - that's a disqualification by itself.
I actually remember testing the Cirrus Logic cards in 8 bits and the differences to 16 bit were negligible. This discouraged me to do full run for all adapters.

retro bits and bytes

Reply 15 of 2374, by Disruptor

Posted on 2019-02-25, 08:10

Disruptor Offline

Rank Oldbie

Rank: Oldbie
Posts: 1760
Joined: 2018-03-22, 18:31
Location: European Union

Well done.
Too bad you haven't tested any VL graphic cards in your ASUS 486 PVI SP3.
What cache timing parameters do you use on this board at 40 MHz FSB?
What cache modules do you use?

Reply 16 of 2374, by pshipkov

Posted on 2019-02-25, 21:43

pshipkov Offline

Rank l33t

Rank: l33t
Posts: 2200
Joined: 2018-10-11, 05:08

I have couple of VL cards from the '93/'94 period - cooking a "traditional" 486DX2 build with them atm.
Updated the post with the VLB tests as well.

All BIOS parameters are at their lowest timings (2-1-1-1, 0 waits, etc).
Using 10ns ISSI chips. Had to sort through a bunch to find the right ones, otherwise Doom and Win95 complained at timings below 3-2-2-2.

Thanks.

Last edited by pshipkov on 2020-03-15, 08:02. Edited 3 times in total.

retro bits and bytes

Reply 17 of 2374, by Disruptor

Posted on 2019-02-26, 15:20

Disruptor Offline

Rank Oldbie

Rank: Oldbie
Posts: 1760
Joined: 2018-03-22, 18:31
Location: European Union

pshipkov wrote:

Are you interested in how VL compare against PCI ones, or just testing bunch of VL against each other ?

PCI vs VL. Especially ET4000/W32 VL

pshipkov wrote:

All memory and cache timing parameters are at their minimum (2-1-1-1, lowest waits, etc).
Using 10ns ISSI chips. Had to sort through a bunch to find the right ones, otherwise Doom and Win95 complained at timings below 3-2-2-2.

Same problem here. But i haven't found a combo that runs on 2-1-1-1 @ 40 MHz FSB yet.

Reply 18 of 2374, by pshipkov

Posted on 2019-02-27, 03:49

pshipkov Offline

Rank l33t

Rank: l33t
Posts: 2200
Joined: 2018-10-11, 05:08

@mrau
Tested 486DX5 Matrox and Cirrus Logics at 8 bits in Win95. WT2 reports no performance difference compared to their max bit depths. A bit suspicious i admit.

@rasz_pl
Updated the 386 Doom tests according to your suggestions.
Also, converted all Doom charts to FPS.

@disruptor
I hear you about the Doom/timings. Pain in the rear.
Will try to "smoke" the ET4000 VL/PCI at some point soon.

retro bits and bytes

Reply 19 of 2374, by pshipkov

Posted on 2019-03-04, 00:41

pshipkov Offline

Rank l33t

Rank: l33t
Posts: 2200
Joined: 2018-10-11, 05:08

Installed the Adaptec SCSI adapter to the 386 PC.
Updated the main post with few more screenshots for Speedsys and Coretest tests.
In a nutshell - it boosts the disk speed by about 1Mb when the system is running at 40MHz and with half a megabyte when running at 45MHz.

retro bits and bytes

Go to top of page Go to top of page

Back to System Specs