VOGONS


First post, by dionb

User metadata
Rank l33t++
Rank
l33t++

One of the more obscure VGA chipsets - UMC's UM85C418F. Surprisingly it has extremely good raw DOS performance, coming very close to Ark 1000VL. I'm working on an all UMC build, and have two VLB cards with this chipset.

The oddest thing about the chipset is that it also supports I/O functions. Cards with it invariably have an IDE interface and sometimes full I/O (IDE, floppy, game, serial, parallell) as well as VGA. Gerwin posted a topic about a system with one of those cards a few years back:
486 VLB UMC-Chipset, what is it?

I have the same card and a different revision of the same motherboard - but I also have a second card, a Puretek PT-2037. Information about this card seems non-existent apart from an FCC registration (ID: H52PT-2037). It "only" has VGA and IDE, and 1MB of RAM in two SMD SOJ chips. However it has space for 8x 44256 (1Mb) chips, which would add up to 1MB. When I see empty PCB like that I start itching to grab my soldering iron. But... 2MB RAM on old VLB chipsets doesn't always make sense (CL GD542x is a case in point), so before I start I'd like a second opinion on what if anything this might do.

IMG_20200301_130917.jpg
Filename
IMG_20200301_130917.jpg
File size
1.56 MiB
Views
1441 views
File license
CC-BY-4.0
IMG_20200301_131720.jpg
Filename
IMG_20200301_131720.jpg
File size
1.73 MiB
Views
1441 views
File license
CC-BY-4.0

Now, in his 2014 topic gerwin already mentioned he couldn't find any datasheets for the 85C418F. I've given it another try but not found anything either. The only hint comes from the FCC registration of the PT-2037. Its title (really...) states: Puretek Industrial Co Ltd VL Bus VGA Interface Card, Maximum Resolution 1280 X 1024 (interlaced). PT-2037. That suggests that it's just as DAC-limited as the CL chips. However unlike them it doesn't have an integrated DAC. Instead it has a UMC UM70C188. Now, finding info on that is almost as hard as on the 85C418F, but a single line here gives a lead: UM70C188 (Color palette with triple DAC, same as UM70C171). And datasheets for the UM70C171 are available: pdf.datasheetcatalog.com/datasheet/UMC/mXyztwts.pdf

Unfortunately I'm not able to figure out resolution or refresh rate limits from that datasheet, all I can interpret is that it's limited to 256 concurrent colours from a 256k palette, courtesy of 6bit operation.

So, anyone want to hazard a guess what if anything might happen if I were to solder in the extra RAM (and filter caps)?

Reply 1 of 24, by mpe

User metadata
Rank Oldbie
Rank
Oldbie

I have no idea if the upgrade can make this more useful. I'd say no, even if actually unlocking more non-interlaced or true-color modes. There are better cards for Windows and for DOS probably doesn't matter.

But given the PCB layout, It looks like it is something that's not too hard. I'd do it just for the science. Makes a nice weekend project and self-satisfaction if nothing else.

Definitely report back if you end up doing it.

Blog|NexGen 586|S4

Reply 2 of 24, by dionb

User metadata
Rank l33t++
Rank
l33t++

Next time I'm ordering components I'll get the necessary sockets and caps, and see about the chips. I'd love high colour at 1024x768 non-interlaced, but that's probably out of the question with this RAMDAC anyway. Dead right that this is so easy there's no real reason not to do it.

Bigger challenge: finding the missing JP7 on the card. According to silkscreen on the back that would let me set the IDE to a different I/O address so that this could work as secondary interface. Unlike the RAM that would actually be useful...

Reply 3 of 24, by gerwin

User metadata
Rank l33t
Rank
l33t

Nice card there. So the IDE functionality is actually in that UM85C418 chip.
Just a crazy thought: Can it be solder spots for memory to use instead of the two SMD SOJ chips? My card also has solder spots for two types of memory, but they overlap, so that only one type can actually be used.

--> ISA Soundcard Overview // Doom MBF 2.04 // SetMul

Reply 4 of 24, by dionb

User metadata
Rank l33t++
Rank
l33t++
gerwin wrote on 2020-03-04, 22:06:

Nice card there. So the IDE functionality is actually in that UM85C418 chip.

It certainly looks that way. Can't figure out what the UM9502 is doing, but given most of the traces go directly into the UM85C418 it must be the active component.

Just a crazy thought: Can it be solder spots for memory to use instead of the two SMD SOJ chips? My card also has solder spots for two types of memory, but they overlap, so that only one type can actually be used.

Not crazy at all, that was one of my concerns. I can see some of the traces overlap - but not all. However on a multilayer board that doesn't tell the whole picture anyway. Any idea how I could determine that?

Reply 5 of 24, by mpe

User metadata
Rank Oldbie
Rank
Oldbie

I assume this is a chip with 32bit interface without bank interleaving. So most of the bus will be shared. But there should be at least one connection ( OE or CAS line) that will be wired to the main chip differently when comparing 2x 256k×16 and 8x 256k×4 banks if this is really 2M upgrade and not just an alternate memory configuration

Blog|NexGen 586|S4

Reply 6 of 24, by douglar

User metadata
Rank Oldbie
Rank
Oldbie

I found one that looks like it is from June 1993 with 12 ram chips, so 1.5MB of video ram?

IMG_2088.jpg
Filename
IMG_2088.jpg
File size
767.32 KiB
Views
873 views
File license
Public domain

It has an oversized rom that goes from C000 to CBFF.

                                 HWiNFO v6.2.3 
º Video Chipset: UMC 85C418 º
º Video Card: Genoa 6400/6600 (GVGA) º
º Video Memory Size: 512 KBytes of DRAM, 1,024 KBytes º
º Video Card Bus: 16-bit º
º Video RAMDAC: ICS5341 GENDAC TrueColor º
º Video BIOS Manufacturer: Unknown º
º Video BIOS Version: 1.05, º
º VESA Support: Not Present º

Edit^^^ I think that Genoa identification is a mistake. I'll follow up with Mumak.

As an IDE controller:

  • it's picky about storage > 512MB
  • does not work with EZ Drive 9.09.
  • Curiously, the Winbond driver loads for it and says it enables 32bit access
  • Performs at ISA speeds for me so far: 1400KB/s for read and for write
Last edited by douglar on 2024-01-14, 02:10. Edited 1 time in total.

Reply 7 of 24, by mkarcher

User metadata
Rank l33t
Rank
l33t
dionb wrote on 2020-03-04, 23:28:

Can't figure out what the UM9502 is doing,

That's the clock synthesizer for the VGA pixel clocks. That's why it is located next to the 14.318 MHz reference crystal.

Reply 8 of 24, by douglar

User metadata
Rank Oldbie
Rank
Oldbie
douglar wrote on 2024-01-13, 17:09:

I found one that looks like it is from June 1993 with 12 ram chips, so 1.5MB of video ram?

I put the BIOS here:
https://www.vogonsdrivers.com/getfile.php?fileid=2141

It has a VGA BIOS

**** UMC SuperVGA UM85C4X8  70Hz - 1991, 1992 ****
BIOS VERSION 1.05, BOARD VERSION VL-1AV 11/22/93
(C) COPYRIGHT AWARD SOFTWARE, INC. 1991
(C) COPYRIGHT UNITED MICRO. CORP. 1989

And an IDE BIOS:

º   VESA SUPER I/O Card Installed   º
º ROM BIOS FOR UM82C418 º
º VERSION 1.02 º

And I added the IDE driver from this bundle:
https://www.vogonsdrivers.com/getfile.php?fileid=1471

º     UM82C418 VL BUS IDE DRIVER    º
º COPYRIGHT 1993,6,10 º
º VERSION 1.07 º

Reply 9 of 24, by CoffeeOne

User metadata
Rank Oldbie
Rank
Oldbie
dionb wrote on 2020-03-01, 13:54:

One of the more obscure VGA chipsets - UMC's UM85C418F. Surprisingly it has extremely good raw DOS performance, coming very close to Ark 1000VL. I'm working on an all UMC build, and have two VLB cards with this chipset.
....

Sorry for a "necro reply".
But I am interested in that.
What about the extremely good raw DOS performance? Any benchmarks? Doom, Wolf3D? If possible a comparison using Am5x86 would be appreciated, because with a DX2-66 differences are always small.

Reply 10 of 24, by dionb

User metadata
Rank l33t++
Rank
l33t++
CoffeeOne wrote on 2024-01-14, 11:02:
Sorry for a "necro reply". But I am interested in that. What about the extremely good raw DOS performance? Any benchmarks? Doom, […]
Show full quote
dionb wrote on 2020-03-01, 13:54:

One of the more obscure VGA chipsets - UMC's UM85C418F. Surprisingly it has extremely good raw DOS performance, coming very close to Ark 1000VL. I'm working on an all UMC build, and have two VLB cards with this chipset.
....

Sorry for a "necro reply".
But I am interested in that.
What about the extremely good raw DOS performance? Any benchmarks? Doom, Wolf3D? If possible a comparison using Am5x86 would be appreciated, because with a DX2-66 differences are always small.

My VLB 5x86 system is dead atm, so might not immediately be able to comply, but will try asap on either that or my Pentium 100 VLB system.

Reply 11 of 24, by douglar

User metadata
Rank Oldbie
Rank
Oldbie
CoffeeOne wrote on 2024-01-14, 11:02:

Sorry for a "necro reply".
But I am interested in that.
What about the extremely good raw DOS performance? Any benchmarks? Doom, Wolf3D? If possible a comparison using Am5x86 would be appreciated, because with a DX2-66 differences are always small.

I see some here:

Re: 3 (+3 more) retro battle stations

Re: 3 (+3 more) retro battle stations

Reply 12 of 24, by CoffeeOne

User metadata
Rank Oldbie
Rank
Oldbie
douglar wrote on 2024-01-14, 12:19:
I see some here: […]
Show full quote
CoffeeOne wrote on 2024-01-14, 11:02:

Sorry for a "necro reply".
But I am interested in that.
What about the extremely good raw DOS performance? Any benchmarks? Doom, Wolf3D? If possible a comparison using Am5x86 would be appreciated, because with a DX2-66 differences are always small.

I see some here:

Re: 3 (+3 more) retro battle stations

Re: 3 (+3 more) retro battle stations

Thx for the link. That does not look too impressive for the UMC graphics. Moreover both tables do not have values for Quake, so it always crashes with Quake?

Last edited by CoffeeOne on 2024-01-15, 19:12. Edited 1 time in total.

Reply 13 of 24, by dionb

User metadata
Rank l33t++
Rank
l33t++

Afraid I won't be able to give any answers soon - tried to get a system running with an untested Biostar VLB motherboard and my UMC 82C418F card, but after a few succesful POSTS the machine hung and magic smoke started emanating from both motherboard and VLB card. Haven't figured out exactly what happened (would have been typical for tantalum cap failure, but regardless, it's unlikely this card is doing anything again anytime soon, at least not until I've replaced the 74LS245 chip that failed on me.

Last edited by dionb on 2024-01-15, 13:33. Edited 1 time in total.

Reply 14 of 24, by douglar

User metadata
Rank Oldbie
Rank
Oldbie
dionb wrote on 2024-01-14, 23:25:

Afraid I won't be able to give any answers soon - tried to get a system running with an untested Biostar VLB motherboard and my UMC 82C418F card, but after a few succesful POSTS the machine hung and magic smoke started emanating from both motherboard and VLB card. Haven't figured out exactly what happened (would have been typical for tantalum cap failure, but regardless, it's unlikely this card is doing anything again anytime soon, at least not until I've replaced the 72LS245 chip that failed on me.

I’d pour one out for you if that was allowed on this fourum

IMG_2751.gif
Filename
IMG_2751.gif
File size
97.8 KiB
Views
591 views
File license
Public domain

Reply 15 of 24, by dionb

User metadata
Rank l33t++
Rank
l33t++

I'd suggest getting out the candles and dubious red crayons - I fully intend to resurrect that card 😜

If it's just that one chip that's dead, I can fix it in five minutes once I source a replacement (in fact I might have one knocking around somewhere). I'm more concerned with the motherboard; whatever shorted that chip out probably did something very nasty there too. Oddly enough, I've not found any shorted tantalum caps on it - which is bad news, because anything else causing this kind of behaviour will be more difficult to find and fix.

But on-topic:

Good news - I recalled I had two 85C418F cards, and found the other one in my all-UMC system (UMC VGA, motherboard chipset, but also I/O, NIC and even CPU)
Bad news - with yesterday's incident and my usual 486 VLB system on the blink, no 5x86.

But I also have a DataExpert OPTi-based 386/486 board with Cx486DLC on it, and it has VLB slots, so...

Only had time to do Doom2 timedemo 3, but did stick every working VLB card I had in (and one ISA card for comparison):

4471 gametics in realtics -
Ara VGA 3071 (ISA Tseng ET3000AX): 28492
P-9000 (Weitek 5186 / P9000): 20998
Puretek PT-2072 (Xtec AGX016A): 18897
Miro Crystal 24S (S3 928): 17231
Unknown (Cirrus Logic GD5424): 17096
Aview Aview2 (S3 805): 17041
FIC (?) TK85C418GVIO (UMC 85C418F): 17031
Diamond Speedstar Pro VLB (Cirrus Logic GD5428): 17018
DTC WG-6000VL (WD WD90C33): 17011
Ark Logic 1000VL2M (Ark 1000VL): 17006

I ran the first and last test a couple of times to see about variation. Difference was max 1 realtic, so only ran the rest once.

So, despite the slow CPU, there are definite differences between the cards. ISA is much slower than VLB. I strongly suspect that the Weitek 5186 used by DOS on the Power 9000 card is only connected to ISA, not VLB. Or it's just supremely crap. The AGX016A is a very old and obscure VLB chipset, but from the S3 928 onward scores are very close, just 1.3% difference. If the variation had been bigger, I would say it was all CPU limited, but the fact the variation is negligible and the differences while small are reproducible suggests the differences are real and related to card performance. Conclusion is that this 85C419 card falls short of WD90C33 or ARK1000 - but the difference is less than 0.2%, at least in Doom2.

Now, I distinctly remember my other card outperforming the ARK1000, but I can't exactly prove it. It will only have been a tiny difference in performance either way. One possible reason could be different performance with faster CPUs. Or it could be difference between the cards. The live card tested here has 80ns RAM, the dead card has 60ns RAM. Or I'm just wrong, of course. If I manage to resurrect it, I'll find that one out.

Reply 16 of 24, by CoffeeOne

User metadata
Rank Oldbie
Rank
Oldbie
dionb wrote on 2024-01-15, 20:56:
I'd suggest getting out the candles and dubious red crayons - I fully intend to resurrect that card :P […]
Show full quote

I'd suggest getting out the candles and dubious red crayons - I fully intend to resurrect that card 😜

If it's just that one chip that's dead, I can fix it in five minutes once I source a replacement (in fact I might have one knocking around somewhere). I'm more concerned with the motherboard; whatever shorted that chip out probably did something very nasty there too. Oddly enough, I've not found any shorted tantalum caps on it - which is bad news, because anything else causing this kind of behaviour will be more difficult to find and fix.

But on-topic:

Good news - I recalled I had two 85C418F cards, and found the other one in my all-UMC system (UMC VGA, motherboard chipset, but also I/O, NIC and even CPU)
Bad news - with yesterday's incident and my usual 486 VLB system on the blink, no 5x86.

But I also have a DataExpert OPTi-based 386/486 board with Cx486DLC on it, and it has VLB slots, so...

Only had time to do Doom2 timedemo 3, but did stick every working VLB card I had in (and one ISA card for comparison):

4471 gametics in realtics -
Ara VGA 3071 (ISA Tseng ET3000AX): 28492
P-9000 (Weitek 5186 / P9000): 20998
Puretek PT-2072 (Xtec AGX016A): 18897
Miro Crystal 24S (S3 928): 17231
Unknown (Cirrus Logic GD5424): 17096
Aview Aview2 (S3 805): 17041
FIC (?) TK85C418GVIO (UMC 85C418F): 17031
Diamond Speedstar Pro VLB (Cirrus Logic GD5428): 17018
DTC WG-6000VL (WD WD90C33): 17011
Ark Logic 1000VL2M (Ark 1000VL): 17006

I ran the first and last test a couple of times to see about variation. Difference was max 1 realtic, so only ran the rest once.

So, despite the slow CPU, there are definite differences between the cards. ISA is much slower than VLB. I strongly suspect that the Weitek 5186 used by DOS on the Power 9000 card is only connected to ISA, not VLB. Or it's just supremely crap. The AGX016A is a very old and obscure VLB chipset, but from the S3 928 onward scores are very close, just 1.3% difference. If the variation had been bigger, I would say it was all CPU limited, but the fact the variation is negligible and the differences while small are reproducible suggests the differences are real and related to card performance. Conclusion is that this 85C419 card falls short of WD90C33 or ARK1000 - but the difference is less than 0.2%, at least in Doom2.

Now, I distinctly remember my other card outperforming the ARK1000, but I can't exactly prove it. It will only have been a tiny difference in performance either way. One possible reason could be different performance with faster CPUs. Or it could be difference between the cards. The live card tested here has 80ns RAM, the dead card has 60ns RAM. Or I'm just wrong, of course. If I manage to resurrect it, I'll find that one out.

Thx for the benchmark!
But honestly I am more interested in the high range 486 machines (Am5x86@160 is my favourite).
So for that range your comparison does not give me a new hint. In your comparison a Cirrus 5428 is on par with an ARK 1000. That is clearly not the case with a fast 486. The Ark is 20 or 30% faster than a Cirrus 542x in Doom (I am now estimating, don't have the exact values now).

Reply 17 of 24, by dionb

User metadata
Rank l33t++
Rank
l33t++
CoffeeOne wrote on 2024-01-15, 21:27:
[...] […]
Show full quote

[...]

Thx for the benchmark!
But honestly I am more interested in the high range 486 machines (Am5x86@160 is my favourite).
So for that range your comparison does not give me a new hint. In your comparison a Cirrus 5428 is on par with an ARK 1000. That is clearly not the case with a fast 486. The Ark is 20 or 30% faster than a Cirrus 542x in Doom (I am now estimating, don't have the exact values now).

Once I get a faster VLB system working without emitting too much smoke, I will re-test. Have just ordered some 74HCT245 chips to attempt resurrection.

Reply 18 of 24, by douglar

User metadata
Rank Oldbie
Rank
Oldbie

I did a win98se install.

Notes:
a) Win98se doesn't know my card out of the box and the Win98se generic SVGA driver didn't work with my card
b) The floppy works fine under DOS but in Win98se, my gotek powers off if it is accessed under Win98se and needs a hard reboot. (one time Win98se was able to Ctrl-alt-del to restart but the floppy needed a power cycle)
c) I could manually install The old Win31 drivers under Win98se but they didn't work. Maybe they only did interlaced modes or something. My LCD wouldn't show an image.
d) All of the modes listed on the old win 3.1 video drivers use <= 1MB of ram.
e) I have not found any sysinfo tool that identifies the card as having more than 1MB of RAM even though there's 1.5MB of chips

I could do in a win 3.1 install and try the drivers again, but yeah, still doesn't look like it is going to use more than 1 MB with the drivers that I have.

So I'm out of ideas for how to use more than 1MB ram.

Reply 19 of 24, by dionb

User metadata
Rank l33t++
Rank
l33t++

It's probably only for an obscure interlaced mode like 1280x1024@8b interlaced, like with Cirrus Logic GD-542x chips.

Bad news here - my 74HCT245N chips have arrived, I've removed the dead chip from my burnt-out card and soldered in the new one. It does now let the system boot - no more shorts - and even seems (looking at POST codes) to be recognized as a working card, but zero output.

I've been able to test the VGA BIOS (chip reads fine on my TL866), so that's not the problem - and it would interfere with boot if it was. So the only component left is the RAMDAC, right next to the dead chip. That could be a more challenging thing to fix. I probably have one on my other card, so I can at least rule this out. But that requires digging out the other one again. MIght do it one of these days...