I started the hobby with SCSI - the usual nostalgia and desire to try toys that were out of reach before.
Don't have skin in the SCSI-or-not game. Just share findings for what they are.
Over time my retro computing interests became predominantly performance-finding oriented, but i am not blind about the versatility of things and appreciate all other flavors, so your notes are spot on - SCSI is certainly very interesting.

As for being fair or not - i actually don't think much about that.
Just check bunch of cards for a given interface and that's it.

In fact i was surprised about these findings.
Suddenly up became down and down was up from everything i knew until then.
Took me a while to accept it and change my view.

This SL4100/D4100BL "Cobalt board" has a footprint for a SCSI chip and 50 pin, didn't notice a Patriot, Ambra or other model that advertised SCSI while I was trawling through stuff though, but the PCBBS files have SCSI drivers, not sure if for that or an add in card.

Didn't get a whole lot done today, was scraping at the Dallas all damn day, I swear that pin16 contact was buried 2x deeper than anyone else's. Kept having to sanity check it against pics and vids, there was one point it seemed like I was just smooshing a softish piece of solder around, pushing it deeper without finding top edge. Anyhoo, fully exposed and cut off, battery still had a whole 200millivolts, I dunno why it stopped working 🤣 Will have to solder tomorrow, too crosseyed and shaky now.

Found data for the clock oscillator, AV9154-04 https://www.datasheetarchive.com/pdf/download … e=M&term=AV9154 pic of the frequency table below, starred values listed as "5V only", think this board only implements two bits with the high bit tied low. At least there's two jumpers near the clock osc and no others look likely. So can probably get 25, not sure about 40 or 50, would have liked to have tried those at 2x at least. Though it is suggested by your experience that the FasMath isn't going to do 50.

I go with the soldering iron after these RTC modules.
Does not smell good for some time but pads are exposed after few seconds only.

Very nice table, but two jumpers only suggest 25/33/40.
At least that's how it is on the Alaris Leopard and Cougar boards.

As for the FasMath @50MHz - on some boards it is a go, on others not. More often leans towards no with BL3 CPU.
Impossible to say without trying it.

---

@Anonymous Coward

Tried today CTCHIP on the Matra mobo. Results are kind of inconclusive. Still not sure if OPTi on board, or not.
Can you think of a good way to prove it ?

pshipkov wrote on 2023-03-06, 00:08:

.... @CoffeeOne […]
Show full quote

....
@CoffeeOne

Very cool.
Updated the Wolf3D chart accordingly.
So, what exactly made it even faster ?
Spell it out please.

As for the "limited" local storage.
Can you fill even 4Gb with 486 class software ?
........

I don't know exactly what made it faster honestly spoken.
I made small changes:
VLB IDE controller at fastest setting.
Compact Flash Card instead of harddisk (I think that was the tiny change).
complete new installation of Win98 and everything.
added sound card, DVD drive connected to IDE port of sound card (so compact flash is alone on the VLB IDE)
more free DOS memory.
But after all we are speaking of reallly minor improvements.

About size: true 4 or 8GB is plenty for a 486. But when you keep copies of CDs (like the Win98 installation CD), somehow the free space gets less and less.

How are the CTCHIP results inconclusive? I would think either it works or it doesn't.
If you have an actual 495 board, you could use CTCHIP on that to see if the results are the same.

@CoffeeOne

None of that looks like a factor, but hey - we take the better perf. : )

@Anonymous Coward

Will give this another go.
My approach was not very structured in the first round.

So, i am seeing a reoccurring problem with Mr.BIOS microcodes.
This is not a new thing. Been observing it for long time already.
Happens on different motherboards with different chipsets and different Mr.BIOS versions.

If the system hang during POST, boot to DOS, or at runtime there is a small chance this will corrupt something in the BIOS settings and as a result the local storage device will not be recognized anymore.
This cannot be fixed by using a jumper to reset BIOS settings.
EEPROM chips need reprogramming.
EPROMs also get stuck permanently in this state and also require reprogramming - this is very puzzling to me.

Such situations occur most often when imposing tight wait states in combination with overclocking.
Anyone else seeing this ?

Two days of distractions, finally get back to it today, fire up favorite soldering iron, and wait, still not hot, wait, nope... damn, stone frigging cold, things blown out, gah, okay then, next, stupid fat conical tip on this one, struggle through soldering up the Dallas, somehow the bits that should have been hard were easy and vice versa. Anyhoo, DMM still working, phew, tests good 3.05 V at the Dallas... nice, stick it in... no boots, panic... oh right, had the motherboard unplugged from PSU... boot, phew... okay save some basics... restart... yes it kept time and stuff.. off... on... kept time.. good, seems working.. off... hook up floppy... error no boot... dammit is the controller okay.. fiddle fiddle.. well I thought this was a good drive... grab this random crappy looking one I don't even know about, blow dust out of it, hook it up... boots off floppy, yay, no errors... run some of what I've got on floppy... seems between 386 and 486 DX33 when I've got it set to 1x... okay basic function seems good, reboot to set it to full speed, and hammer keys, nope, reset, hammer keys, nope reset, hammer keys, nope reset... can't get back into setup... (Well I could force HW error, but wanna find the normal way) just been looking stuff up... Aptivas of same age with Surepath it says F1 only while the memory counting, and only direct hint for Patriot and/or Ambra twin is this..
https://marc.info/?l=classiccmp&m=115955310423291&w=2
F2 held while powering on... so off to try this now...

edit: so it turns out it's F1 only during the memory count and it won't detect it if you jumped the gun and caused a 301 keyboard error first. Anyway, first runs at 100, the stuff I had on floppy seemed to confirm it's at 100, snooper, norton sysinfo, old version of HWinfo did but did some probing that knocked it back down to 1x/33. Gets about a third more than a 486dx33 on norton, 2x286mhz on snooper (202) and quick run of whatever version of 3dbench was 15 at 1x and 45 ish at 3x . So seems good. Need to prep a HDD for it and load up with all the versions everyone else has got of everything and do a true baseline run now. Fan is a bit unsteady sounding though, so gotta see to that before I hammer it much at 100.

pshipkov wrote on 2023-03-08, 20:08:

So, i am seeing a reoccurring problem with Mr.BIOS microcodes. This is not a new thing. Been observing it for long time already. […]
Show full quote

So, i am seeing a reoccurring problem with Mr.BIOS microcodes.
This is not a new thing. Been observing it for long time already.
Happens on different motherboards with different chipsets and different Mr.BIOS versions.

If the system hang during POST, boot to DOS, or at runtime there is a small chance this will corrupt something in the BIOS settings and as a result the local storage device will not be recognized anymore.
This cannot be fixed by using a jumper to reset BIOS settings.
EEPROM chips need reprogramming.
EPROMs also get stuck permanently in this state and also require reprogramming - this is very puzzling to me.

Such situations occur most often when imposing tight wait states in combination with overclocking.
Anyone else seeing this ?

Is it happening on boards that can't be programmed on board? I remember some machines doing this in socket 7 era, and advice was to make sure you didn't run with programming voltage enabled. That's about when "hot swap" flashing became a thing among hardware experts, because of BIOSes corrupting. It remained a thing, Athlon XP boards started coming with dual BIOSes so you had a backup if one corrupted. I don't know if it got fixed somehow, like it was a glitch in all codebases they finally found a workaround for or not. Maybe that was part of the reason why there was a switch to new type.

pshipkov wrote on 2023-03-08, 20:08:

So, i am seeing a reoccurring problem with Mr.BIOS microcodes. This is not a new thing. Been observing it for long time already. […]
Show full quote

So, i am seeing a reoccurring problem with Mr.BIOS microcodes.
This is not a new thing. Been observing it for long time already.
Happens on different motherboards with different chipsets and different Mr.BIOS versions.

If the system hang during POST, boot to DOS, or at runtime there is a small chance this will corrupt something in the BIOS settings and as a result the local storage device will not be recognized anymore.
This cannot be fixed by using a jumper to reset BIOS settings.
EEPROM chips need reprogramming.
EPROMs also get stuck permanently in this state and also require reprogramming - this is very puzzling to me.

Such situations occur most often when imposing tight wait states in combination with overclocking.
Anyone else seeing this ?

Very perplexing. I don't normally use IDE cards controlled by the MB's BIOS, so I haven't run into this exactly. But does the issue also occur with CPUs that do not contain L1 cache? I've only run into the situation whereby MR BIOS board won't turn on or BIOS data has been changed. I have to clear the data by pulling the battery. Try removing the MR BIOS EEPROM, putting in the AMI BIOS, boot up, then put the MR BIOS EEPROM back in. Will this let the local storage device funciton again?

@BitWrangler
That's how it goes often.
You plan for a productive and enjoyable retro computing session.
Few minutes later - a "disaster" upon a "disaster".
At least you came through.
Ok, so you have now a working 3x33 MHz BL3.
Great.

Do you have control over wait states in the BIOS ?

About the F1 hotkey thing.
I am big fan of standardization. Makes everybody's life easier.
Never understood the big brand and their customized personal computers. Irrational. Anyhow. : D

---

The Mr.BIOS issue appears on the Alaris Cougar boards and at least two other motherboards with Mr.BIOS - one with UMC chipset the other one with OPTi.
I don't think it is related to presence of L1 cache, but also - not sure if it is BL3 only or SXL2 and 386DX as well - will know more soon.
A very cool board is on the test bench. It already pushed Mr.BIOS into the weird zone, which actually was the moment where it clicked in my head that the previous occasions are not completely sporadic, but strongly related to Mr.BIOS doing something wrong.
For reference - i have an Award BIOS for this board - nothing like that happens with it.

pshipkov wrote on 2023-03-10, 07:13:

@BitWrangler That's how it goes often. You plan for a productive and enjoyable retro computing session. Few minutes later - a "d […]
Show full quote

@BitWrangler
That's how it goes often.
You plan for a productive and enjoyable retro computing session.
Few minutes later - a "disaster" upon a "disaster".
At least you came through.
Ok, so you have now a working 3x33 MHz BL3.
Great.

Do you have control over wait states in the BIOS ?

I think there's settings for everything that can be set in BIOS, I keep forgetting to take screenshots. Got the fan lubed up and cleaned more thoroughly. Got bogged down with finding a HDD to use, I swear every used drive I buy ends up creating more storage/archival problems than it solves, interesting old apps and drivers etc. So yah, ended up "going through" a bunch, and still don't have one set up. Then more delays... it is apparently super fussy about memory, I can't get more than the 2 IBM 4MB SIMMs it had in it to work, the other 2x4MB that were sitting in it were EDO. I might end up trying my entire stock of 72pin FPM in it just to find something. I am not quite understanding what it's problems are with them. I had some 70ns IBM PS/1 8MBs with the right SPD coding and it hated those. So far I have no idea if it's gonna like 60ns FPM if I find any. Wanted to have 16MB in it. Also I thought I might be able to run a VLB in the riser slot, but there's too many pins, there's an extra double ground at the back and front of the ISA portion of the slot. Not sure anything on ISA is gonna match up to the VLB onboard 5428, which gave good results on 3dbench, but I will test anyway, maybe have an ET4000AX board to test, but it might not do well, early versions of it are supposed to be crippled and I don't know if this one will be (8 bit mode) Anyway, very slow going. Now today I'm probably not going to get to it, other chores, then "events" tomorrow.

edit: also keep looking at the spare video RAM pad to give it more than 512kB.... might try a socket, chop the bottom out and solder it from inside, risk of melting SIMM or ZIP socket. I have found one kinda nasty/battered looking 40 pin RAM chip, though I also have a SIMM with that style on I might harvest (Two gone off it already)

Got an hour to mess with it, went through a box of RAM, none of the nice matched sets would work of course, nor any 8MB module, some showed 4MB. Anyway, the very dregs of the box provided two 4MB modules that worked, one had IBM chips but module was made in Italy, so probably some IBM/Olivetti bastard, the other had TI modules on. They work with the orig 4MB IBM chips so yay, 16MB functioning. I booted a 3GB drive that had a Win98SE install on it for a stress test, and it seems okay. Possibly I have a nicer matched batch of IBM chip IBM FPM somewhere that it might like... but I'm kinda feeling like it doesn't deserve it 🤣 maybe it will get modules out of my 6571 when that gets gone through. So yah, has large HDD support, probably up to 8GB in case I didn't mention that. I think the watch crystal in the Dallas clock has gone screwy, it's gaining weeks while I leave it off. I don't know if that was prior age related failure or whether I damaged it trying to get the battery wires attached with the inappropriately large and clumsy soldering iron tip.

@BitWrangler
It is surprising there is so much trouble with RAM.
Usually these boards take 4x4Mb for maximum of 16Mb, but troubles with 4Mb sticks is certainly out of the ordinary.
At least you overcame the resistance.
Win98SE on this poor silicon and working ok ?
By ok you mean slideshow upon every user action, or actually ok ok ?

Last question - what will be your next steps with this guy ?

---

Several 386 motherboards based on Unichip U4800 passed through here, but none of them was worth talking about.
They were those compact PCBs with 128Kb level 2 cache only, clock generators limited to 80MHz (40MHz FSB) and very slow BIOS microcodes.
Finally managed to obtain a motherboard with the right characteristics.

Target Micro Typhoon 486-386/ISA based on Unichip U4800 chipset - a hybrid 386/486 assembly.

Had no idea what to expect from it, so curiosity was significant.
Unfortunately it came dead. Or so i thought.
The drama started with me inserting an 80MHz crystal and trusted CPU, RAM, VGA.
Hard no lights.
But i was desperate to get this guy up and running. Such boards with this chipset are extremely rare. Unlikely i will be able to find another.
Probed all over the place with multimeter and oscilloscope - everything came out clean and as expected.
Went over jumper settings countless times. Tried any other configurations, including random ones.
Desperation gradually ramped up. More and more invasive methods for reanimation were applied.
Endless swapping of components, reflowing joints, pins, pads. Bending.
Twice baked in the oven.
Heating with hot air gun sections of the PCB - blindly trying to fix eventual cold joints
At some point legs of the main IC started separating from the pads underneath. This resulted in even more soldering work around them.
4 level 2 cache sockets completely melted.
And so on.
Eventually i gave up and threw the motherboard in the pile of broken components.
...
2 weeks later i impulsively grabbed it, put it back on the test bench, inserted a 66MHz oscillator - it lit up right away and booted to DOS.
My jaw was on the floor.
Inserted an 80MHz oscillator - hard no lights.
No lights with 70, 80, 90, 100.
But worked right away with 110MHz one.
What the heck.
Over time, don't ask me how i arrived at it, i realized that with 70-100MHz crystals i must switch JP3 from 1-2 to 2-3 or back during POST.
Despite this nasty quirk i felt very guilty for all the abuse this motherboard went through for no reason.
The really surprising thing is that the PCB survived my Emergency Room, but hey, i take it.

With all the drama behind us (me and the board) i spent good time to heal it properly and finally moved onto testing it.

16Mb of RAM
256Kb level 2 cache in WriteBack mode
Cirrus Logic GD-5434, 2Mb
Standard IDE controller with CF card.

--- 386DX running at 55MHz, ISA at 18.33MHz

Any FPUs makes the system very unstable when FSB is in the range of 45-50MHz.
At the same time at 55MHz or higher FPUs don't get recognized.
Basically things work well at up to 33MHz FSB, but get weird in different ways past that - the POST issue outlined above which needs jumper swaps and now this FPU stuff.
Testing a 33MHz system is just not interesting, so decided to go with the 55MHz no FPU configuration.

All BIOS settings on max, except:
AT WAIT STATE = 1 WS (best is 0 WS)
AT BUS CLOCK = CLK2/8

Performance is actually pretty ok.
Above average performance in the DOS interactive graphics tests, but very slow in Windows accelerated GUI.

No Quake 1, PC Player Benchmark and offline graphics tests (no FPU).

--- TI 486SXL2

This didn't go well.
Motherboard does not light-up above 35MHz FSB (70MHz crystal).
Tried hard to figure this out, but without success.
Disappointing.
Decided not to waste time testing at such a low frequency (35MHz).

--- IBM BL3 running at 100MHz (2x50), ISA at 16.66MHz

Motherboard loved this processor, but the situation with FPUs still stands.
All BIOS settings on max, except:
AT WAIT STATE = 1 WS (best is 0 WS)

Average performance.

No Quake 1, PC Player Benchmark and offline graphics tests (no FPU).

--- IBM BL3 running at 110MHz (2x55), ISA at 18.33MHz

Not 100% stable, but feels very close to the threshold. Rare, sporadic hangs may occur from time to time.
Same FPU problems.

All BIOS settings on max, except:
AT WAIT STATE = 1 WS (best is 0 WS)
CACHE SCHEME = WRITE-TRHOUGH (required for stability in Windows)

Speedsys hangs.

Performance is pretty good.
Number 4 in Superscape and Doom.
Not great in Windows accelerated GUI.

No Quake 1, PC Player Benchmark and offline graphics tests (no FPU).

---

Looks like this motherboard was designed to handle 386 CPUs at up to 33MHz. The stability at 55Mhz feels like an accident than by design.
Obviously not a great option to build a retro computer around it, but at least it was enough to answer some performance questions about the Unichip U4800 chipset.

benchmark results

What I find most shocking that you want all out on cold joint repair before trying a 50-66 Mhz oscillator. When I have a new board, I usually do my first power-up with a 50 MHz oscillator.

This only increases the surface area of my face-plant.
Series of successful straight to 50MHz FSB cases lowered my guard and led to that.
Lesson relearned.

Yah I've done that a few times over the years, skipped basics and gone hunting for a more sophisticated problem when it was something stupid.

pshipkov wrote on 2023-03-12, 05:27:

@BitWrangler It is surprising there is so much trouble with RAM. Usually these boards take 4x4Mb for maximum of 16Mb, but troubl […]
Show full quote

@BitWrangler
It is surprising there is so much trouble with RAM.
Usually these boards take 4x4Mb for maximum of 16Mb, but troubles with 4Mb sticks is certainly out of the ordinary.
At least you overcame the resistance.
Win98SE on this poor silicon and working ok ?
By ok you mean slideshow upon every user action, or actually ok ok ?

Last question - what will be your next steps with this guy ?

Yes, don't have a RAM table to refer to so maybe was putting double banked RAM in wrong sockets for it, though I tried one of the "sees 4MB" 8MBs in different sockets. What's working is all 1-0-0-0 PD coded. I am wondering if 8MB is so much trouble because IBM didn't distinguish between 80ns and 70ns in their codes, so it thinks "no no no, too slow". If I get REALLY curious I might rig some jumpers to Presence Detect pads and see what happens. Although another thought is that these slots seemed a bit cheap and nasty in the first place and maybe tiny fractions of a millimeter in board thickness are as much a problem as type of ram chip on the module.

Win98, I was meaning that it was okay in that it didn't bluescreen or otherwise glitch. 98 is more sensitive to marginal timings, overheating etc than quake is, slightly less so than Doom but there's not much in it, hard to find a system 1990-2000 era that screws up on Doom but will still run 98. Whereas you can set up something that will run quake but crash 98. For use, 98 seems just about tolerable, a bit laggy, what you'd expect from any mid 486, HDD is in PIO mode, no DMA so that's not helping. You maybe wouldn't want to use it productively for any reason, but if you need something like Internet Connection Sharing, Windows networking, or some PnP setup that doesn't work in DOS then it would chug through it to do what you had to.

Next steps, near future, still working on getting it it's very own HDD to keep and get that loaded with standard benchmarks for starters. Then do a baseline run, then start playing see if we get 40x3, 50x2 etc. Test the 5434 and ET4000 ISA cards in it. ... ... after that though, exploring every single setting I can get out of it, slow and faster, idea being to make it into a 286 through mid 486 variable system for picky games.

I been having a "slow" week, if it isn't 5 minute tasks turning into all day games of phone tag, it's feeling like someone switched out all my coffee for decaf. I don't know if it's the daylight savings "jet lag" or early maple tree pollen or something.

Anyway, discovered one more thing about that board... it doesn't have full LBA support, it's BIGDOS or something up to 4047 cylinders. Worse, if you have BIGDOS translation on, it can't even see the disk if it's bigger, so gotta leave translation off and only get 512 or whatever 1023 cylinders gets you. So anyway, spent my PC time yesterday setting up a nice 7.5GB disk for it... then today figuring out why it don't work....well I can see it on a floppy boot, so halfass have access to the benchmarks on it. It's getting a 420MB Quantum tomorrow, there now, second time you threw something nice back in my fact you little bugger.

Oh also has a millennium rollover bug... I wonder if that got a lot of them trashed, hence low number of survivors. The RTC calmed the heck down, it was galloping weeks ahead, I don't know why. May have been a fresh battery, slight overvoltage effect, and it being in a week or so now trimmed it to nominal.

Ok, sounds like things are going somewhere.
About your HDD issues: I use local storage devices no bigger than 2Gb, or maybe 4Gb in some cases. Greatly reduced time spent on the wrong stuff.
I think this will be a good fast 286 to slow 486 machine.

---

I went through quite a few XT/286/386/486 motherboards based on UMC chipsets, but the only really truly good one so far is Biostar MB-8425UUD-A.
Hope dies last and i keep looking for another unicorn out there, especially in the 386 domain.

Unknown brand/model, based on UMC UM82C482AF, UM82C481BF, UM82C206F chipset.

As mentioned above, couple of boards based on this chipset passed through here and one of them (Elitegroup US 3486) showed sign of real potential, but didnt scale well past the 45MHz.
This unknown motherboard seemed to have the bones to show what UMC 481/2 is capable of, so decided to give it a spin.

Old battery leak damaged 5-6 rails in the upper left corner.
That was fixed promptly. System lit-up right away with the fixes in place.

The assembly was no drama at all. Stuff just works. Nothing more to add really.

For testing used STB Nitro graphics card (Cirrus Logic GD-5434) and standard IDE controller (UMC chipset), 256Kb level 2 cache, 16Mb 50ns RAM (4x4).

--- IBM BL3 at 110MHz (2x55), ISA bus at 13.75MHz

System cannot be fully stable, especially with FPU present, regardless of BIOS settings, even the most conservative configuration.
It was established in previous posts that no 387 grade FPUs scale past the 50MHz (natively, meaning - no clock doubling).

Things are ok in DOS only. Captured few numbers with DOS interactive graphics tests:

1Wolf3D: 66.8 fps
2Superscape: 50 fps
3Doom: 26.5 fps

Not bad at all really, but that's pretty much it.

--- IBM BL3 at 100MHz (2x50), ISA bus at 12.5MHz

Fully stable system.

All BIOS settings on max, except:
DRAM WAIT STATES - READ = 1 WS (best is 0)

Great overall performance, very satisfying experience.

--- TI 486SXL2 at 45MHz, ISA bus at 15MHz

System is functional at up to 60MHz (120MHz oscillator), but there are all sorts of issues, especially IDE failures and obviously no FPU.
55MHz is possible and as far as i can tell stable, but without FPU, also level 2 cache does not get recognized. This configuration is inherently compromised, so didn't spend time to "explore" it.
50MHz cannot be fully stable as well. This was surprising. Tried hard, but without success.
Down to 45MHz we go.

At up to 16Mb RAM everything is good, but with 32Mb some of the heavy offline compute and code compilation tests get flaky.
That's a red flag. Ran with 16Mb only.

All BIOS settings on max, except:
DRAM WAIT STATES - READ = 1 WS (best is 0)
DRAM WAIT STATES - WRITE = 1 WS (best is 0)

Speedsys hangs. This is often the case on many motherboards with SXL2 processor.

Clock-to-clock performance is not bad at all, but at "mere" 45MHz it appears mediocre when compared to the rest.

--- AMD 386DX at 45MHz, ISA bus at 15MHz

Again, i was not able to reach 50MHz with classic 386DX processor which kind of became the norm over time.
Same symptoms like with the SXL2 case above.

System is fully stable.

All BIOS settings on max, except:
DRAM WAIT STATES - READ = 1 WS (best is 0)

Forgot to take a Speedsys screenshot. Will do at some point later.

Slightly above average performance, which is not bad for a 45MHz system.

---

benchmark results
Look for "Unknown UM82C482".

---

One of the best 386 class motherboard based on the UMC 82C481/82C482 chipset that i have seen so far.
But this statement applies only when BL3 processors in use.
Performance is impressive with that processor, even at “mere” 90MHz (2x45):
- number 2 in PC Player Benchmark
- number 1 in Quake 1, along with DTK PEM-4036Y
- number 3 in the complex offline graphics tests
But looking at the 386DX tests, there is a chance that Elitegroup US 3486 is actually the better system. Unfortunately i sold that Elitegroup mobo before gained access to SXL2 and BL3 CPUs, so this is only a presumption at this point.

I am going to leave this here ... for reference.