Ok, looks like I've routed everything. Now it's time to correct silkscreen, optimize and double-check everything else. And I've probably have lots of time for this - Utsource still can't send me IC's I've ordered. Few notes:
1) It's 2mb only for now. 1mb won't work due to schematic modifications I've made to simplify routing, 4mb was considered risky due to increase routing complexity and, possibly, bios limitations. Although it's still possible to add additional 2mb through double-stacking - RAS0 is available at TP1.
2) I've opted for 39SF010A instead of 27c256 EPROM. Due to it's capacity you can use two ROM images up to 64K (32K VGA + 32K extra, XT-IDE for example) - selectable with a jumper. 32K-64K selection needs some soldering though.

dirkmirk wrote on 2020-06-27, 14:18:

https://www.youtube.com/watch?v=laMzPYhFSHE&t=5s
Also have the STB Nitro 64 2meg.

Can you please share it's bios? I've found Diamond's v2.02, but not STB's V1.1.

That looks already an awesome card! Also what about the sw drivers? In my previous posts I talked about the problem I've found with the ISA version of the GD5429 (at least on the UMC chipsets) and the generic drivers probably built for the VLB ones; I suspect that also the GD5434 being a PCI oriented chip may have different drivers config for ISA and PCI version.

For what its worth I've got driver disks for the Diamond Speedstar 64 & STB Nitro 64 (5434, windows 3.1 maybe dos?), if anyone wants a copy let me know.

I know a 4MB version has been written off (and perhaps not all that useful), but wouldn't it still be nice to leave a few solder pads to allow the end user to add 4MB if they want?

Anonymous Coward wrote on 2020-07-01, 00:08:

I know a 4MB version has been written off (and perhaps not all that useful), but wouldn't it still be nice to leave a few solder pads to allow the end user to add 4MB if they want?

he just said it made the routing horrible, so probably no

Ack, didn't see that.

Oh well.

Anonymous Coward wrote on 2020-07-01, 00:08:

I know a 4MB version has been written off (and perhaps not all that useful), but wouldn't it still be nice to leave a few solder pads to allow the end user to add 4MB if they want?

It's not written off, just not in the first revision.
If you can solder SOJ IC-s you'll be able to double-stack them. Place second memory IC on top, solder everything pin-to-pin except for RAS (pin 14). Connect RAS pins of the additional ICs and run a wire to a test point on the PCB. Done!

maxtherabbit wrote on 2020-07-01, 02:24:

he just said it made the routing horrible, so probably no

Not horrible, but that'll increase error chance, and Diamond Speedstar64 BIOS simply has no "4MB" string in it. Haven't looked into this yet.

Also I imagine additional costs and it's good until it will stay as cheap as it can be. The point that make this awesome it's not only a 2020 late generation ISA card, but also that it could have a reasonable price.
I wonder how the test of such card might be difficult or not.. I suppose even the lenght of the traces may or not improve or decrese stability or such things. I remember from the Arduino sbc experience projects that people had problem just designing the trace on the pcb and the time the signal would have, being close to other components, etc... I suppose the debug of a self made card might be not easy. 😀

Finally my ICs were shipped, now I need to order PCBs and search for STB Nitro64 BIOS.
Uploaded to github, if you see errors - please PM me. https://github.com/Leshak-pcb/Pine-CL-GD5434

Great update.

I was looking for some info for the STB Nitro64 ISA and for what it's worth I found something interesting about the drivers part.. looking at a list of drivers related files it seems there were specific 386 drivers for it?

NITRO64 VS95120I.ZIP STB NITRO64 ISA V.1.20I WIN95 DRVS.
NITRO64 N64V210D.ZIP STB NITRO64 ISA V.2.10D WIN3X DRVS.
NITRO64 N64IV128.ZIP STB NITRO64 ISA (386) V.1.28 WIN3X DRVS.

EDIT:
Yes I think it's right:

ISA BUS
Windows 95 Drivers
VS95120I.ZIP 02-12-96 STB Vision 95 drivers release 1.20i
Windows NT Drivers
NT 3.51 - Use embedded Cirrus Logic 5434 Drivers
NT 4.0 - Use embedded Cirrus Logic 5434 Drivers
Windows 3.1 Drivers
N64V210D.ZIP 07-21-95 Nitro 64 STB Vision Windows drivers release 2.10d
N64IV128.ZIP 08-08-95 Nitro 64 ISA STB Vision release 1.28 (for 386's)
Utilities/Drivers
NTRIUT12.ZIP 03-07-95 Nitro ISA Utils/Drivers release 1.2

Since low end 386sx motherboards often prove to be inappropriate for (and sometimes incompatible with) Win95 and NT,
16-bit drivers are of greater importance to 16-bit ISA cards.
The 16-bit ISA bus predates 32-bit windows by nearly ten years.

Cirrus Logic's AGP hardware might be more appropriate for 32-bit operating systems.

wow, this looks a great project.
Im definitely interested in grabbing one of these cards!

Ok, finally I've assembled one. And, surprise, it works! Few notes:
1) Performance on UMC 486SX is the same as AVGA2 - as I expected. (Note - my AVGA2 runs at maximum possible MCLK & agressive timings, i.e. as fast as it can)
2) I've tested it with Diamond Speedstar 64 & generic PCI BIOS (still searching for STB & Orchid). The only difference so far - PCI BIOS supports DDC monitor detection, while Diamond BIOS allows to use their drivers and utilities.
3) Haven't tested in Windows - but I'm 100% sure it'll work just fine.
4) It takes several seconds to fill the screen at high resolutions, so modes that require 4mb probably will be painfully slow, even with GUI acceleration.
5) Got some flickering in NSSI - not sure why. Other apps with the same 720x400 mode work fine.
6) It took me whole day to assemble one. Eventually I'll assemble 4 more, plus I've got 5 spare boards w/o parts for them. Or you're free to order and assemble everything yourself - just wait few days while I'll check if everything is up to date on github.

very nice!!!!!!!!!!!!!!!!!!!
yes it is expected that screens that need 4mb vram will be a pain with the ISA bus

I found a couple of GD5434 PCI cards in my stash and made BIOS dumps for you:

Also, please check your PM.

Count me in ! How much will be unpopulated PCB alone ?
I need something that won't generate vertical stripes on LCD (in 320x200) - my Trident TVGA9000c unfortunately does this.

Interested in assembled card.

I'm also interested in one either assembled or in kit form depending on availability and price. The bracket is not included in any case right?

zyga64 wrote on 2020-09-04, 05:58:

Count me in ! How much will be unpopulated PCB alone ?
I need something that won't generate vertical stripes on LCD (in 320x200) - my Trident TVGA9000c unfortunately does this.

Don't count on this fixing that problem. Vertical banding can be caused by high frequency noise yes, but since you said 320x200 (640x400) I think it's more likely a sampling error. Most LCDs sample 400-line modes at 900px/line, which is the appropriate sampling rate for VGA text mode (720x400.) Since they are not capable of distinguishing between the two common 400 lines modes, they sample both of them the same, and 640x400 ends up with vertical banding.

I have vgas that use the same chipset and on the same monitor some have banding some do not.
Most of the times branded ones like Diamond/STB do not have this issue where no-name ones with the same chipset have it.
There are some exceptions that behave the opposite but it's rare occurrence with the bunch of cards I have.