Tiny update:
I'm still working with basic PCB layout, progress is slow - I've got lot's of other work. If you want to help & have some knowledge in electronics I can send schematics so you can check it for obvious errors.
Meanwhile I've discovered that VGA filter capacitors on my Matrox Millennium have value markings printed on them. Turns out that it has ~162MHz cut-off frequency - kinda strange since RAMDAC is rated for 220MHz, but perfect for GD5434 (up to 135MHz). Speaking of filters - Radeon x800, while having reasonable 400MHz cut-off frequency, also has strange band-stop filter that blocks 217-317MHz range.
Also I don't understand why RGB traces are so thick on most early SVGA boards, since they should have 75 Omh impedance, i.e. very thin.

watlers_world wrote on 2020-05-19, 18:41:

Would the GD5434 be a good card for a 16mhz 386sx?
What classic DOS software supports the video acceleration features of the GD5434?

1) No, unless you plan running MS Windows on this.
2) AutoCad, maybe something else.

watlers_world wrote on 2020-05-19, 18:41:

Would the GD5434 be a good card for a 16mhz 386sx?
What classic DOS software supports the video acceleration features of the GD5434?

ISA cards based on 5434 are usually the fastest cards for any ISA only systems in DOS. The increase depends on your current VGA card. You should also take into consideration the price - they can be pretty expensive.

Jager wrote on 2020-06-02, 08:21:

Also I don't understand why RGB traces are so thick on most early SVGA boards, since they should have 75 Omh impedance, i.e. very thin.

I imagine the main concern is minimising inductance and with it picking up of noise etc.
Impedance control probably wasn't a huge priority and nobody probably even knew what trace thickness etc. was needed on the particular PCB material etc. to achieve the impedance necessary. The segment from video out pins to connector is fairly short, and is terminated at the connector side anyway, any transmission line effects probably didn't matter at the resolutions everyone used the cards at... There's not even trace length matching done which begins to matter a lot at the higher resolutions.

keropi wrote on 2020-05-16, 17:29:

I am waiting for this to happen - 4mb is certainly not worth it , maybe keep it 2mb and sockets to upgrade to 4?
higher resolutions and depth don't like iSA much but maybe some ISA 486DX2 could use them

I completely disagree. The ability to run higher resolutions and color depths without high frequency noise from poor routing and design would be the only reason I'd consider using a new production ISA video card. That would really be the only feature that would set it apart from the myriad other "authentic" cards out there, GUI acceleration notwithstanding.

maxtherabbit wrote on 2020-06-03, 12:54:

The ability to run higher resolutions and color depths without high frequency noise from poor routing and design would be the only reason I'd consider using a new production ISA video card.

Higher resolutions and color depths that require 4 MB of video RAM (e.g. 1280x1024 @ 24 bit) are unusable with any 16-bit ISA card. Even if you could build a card that would fully use the 16.67 MB/sec. possible with 16-bit ISA, do the maths on how many frames you can get thru. And I've yet to see an ISA graphics card being able to push more than 7 MB/sec. over the bus. There's a reason why VLB and PCI were created.

Don't forget GUI acceleration in Windows actually doesn't use much bandwidth over ISA. The parts that aren't accelerated are slow - like bitmaps and video. And then there is also a perception of the term "unusable"...

For example ISA ATI Mach64 is just as fast as VLB S3 805 in Windows.

Looking at this as a Windows accelerator:
Clip art and Metafiles were supposed to be scalable.
Higher resolutions, would help prove the chips polygon acceleration.

Why would anyone try for an overpriced red herring when they could create a genuine white elephant?
The GD5434 might not be paradise, but it could be galapagos.

Everything is/was/will be over priced...

High value would only suggest that the the item is expensive, however, a new creation is not simply a white rhinoceros.

The implementation of technological advancements in hardware should not be confused with emulation.
Emulation often provides hardware as it never was and never could be. (pink elephants)

Although custom projects might not be practicable, I would think that a fully outfitted GD5434 should perform very well in 8-bit color modes.
Jager, I wish you luck with your project.

I have an IBM PC340 with this 5434, like this one i found here:

IBM PC340 -- an upgrade/restoration project

The card is only 1mb, but it works fine for most things. My PC340 has a 100mhz P1 (256k cache), and i really dont have any video complaints with it. Let's be honest, if I want to run anything too intensive, I wouldn't be using a 100mhz cpu anyway. For 286/386/486 I think it's a simple easy solution. I

Did adding another 1mb of ram to it make a big difference?

Grzyb wrote on 2020-05-14, 17:34:

kixs wrote on 2020-05-14, 11:08:

I doubt any accelerator card works on a 286 - except ATI/S3 with integrated 8514A drivers and not their own drivers.

ATI Mach8 and Mach32 do work with 8514/A drivers, and such drivers are already provided with Windows 2.11 - so I guess totally 286-compatible.
S3 chipsets, however, don't work as 8514/A, they need dedicated drivers.

watlers_world wrote on 2020-05-16, 01:44:

Perhaps you should use 4mb of memory with a gd543-4. […]
Show full quote

Perhaps you should use 4mb of memory with a gd543-4.

Someone told that their ISA GD5424 was too slow to be of help to their 486.
What type of machine would an ISA GD5434 actually accelerate?

I once tried an ISA WD90C33 in my 286 and found that there were no drivers supporting it's accelerated features.
Grzyb, have you benchmarked your mach32 to see what acceleration features are supported?

I know your posts are about a month old and this is perhaps slightly off-topic now, but..

- Windows 2.03 (plain, no /286 or /386) already supported 256c and high resolutions (say 1280x1024) if a sutiable driver was used.
In fact, Windows 2.x had a completely different palette system than 3.x. It treated 256c as a high colour depth.
If you want to run Windows 1.x/2.x applications on 3.0 in colour, you have to use the Windows 2.x drivers and run Windows 3.0 with Real-Mode kernal (win /r).
Here's an example of Windows 2.x in 256c and 640x400 pels using the Paradise driver: https://www.youtube.com/watch?v=-YGx-KeujIU

- Windows 3.1 in Standard-Mode DOES support acceleration, high resolutions and colour depths, IF the driver in question is compatible with Standard-Mode (*.DRV drivers).
Unfortunately, most drivers from the mid-90s are made for 386 Enhanced Mode and use *.386 (VXD) driver model (which in turn is based on OS/2's LE EXE format).
At this time, 286es were already phased out, also. Your best bet will be drivers for veteran graphics chips.
Paradise PVGA1A/PVGA1B (WD90C00 or later), Tseng ET-3000, ET-4000, OAK OTI-67c, Trident 8900D, Realtek RTG3105..

My AST Bravo dx2/66 already has a PCI gd5434 with 2mb onboard.
However, as a PCI card the GD5434 is not particularly fast.

I have a 486/33 machine with 1mb/GD5428 onboard.
Even for my 486/25 with onboard GD5424,
a 2mb 5434 would hardly be an upgrade.

My 386 machines already have 1mb video cards.

Cirrus chips were a great low cost video solution.
After you remove the one greatest selling point,
you have to do better than 2mb, if you want to impress.

Even still, a custom ISA card is a rather impressive project.

derSammler wrote on 2020-06-03, 13:05:

Even if you could build a card that would fully use the 16.67 MB/sec. possible with 16-bit ISA, do the maths on how many frames you can get thru. And I've yet to see an ISA graphics card being able to push more than 7 MB/sec. over the bus.

Actually, the fastest you can put over ISA (not overclocked, so running at 8.33MHz) is not 16.67 MB/sec, but just 8.33MB/s. While it is obviously true that a single bus operation can transfer 16 Bits (thus 2 bytes) over the 16-bit ISA bus, the shortest possible memory write operation on the ISA bus (called 0WS) is two cycles. Cards that are able to push (err, receive, they only ever optimized write operations) 7MB/sec are quite close to the theoretical limit of 8.33MB/s.

The term "GUI accelerator" does not refer to the datarate of a particular BUS.
A card's internal datarate may be much faster than the ISA bus.
Telling a video card to draw a complex shape may be faster than sending it a picture of one.

When a GPU is less capable than the main CPU it can have a negative performance impact.
A lack of memory or proper drivers can also make any GUI accelerator worthless.

The effect of an ISA accelerator is rather parenthetical.

Jo22 wrote on 2020-06-21, 03:59:

I know your posts are about a month old and this is perhaps slightly off-topic now, but.. […]
Show full quote

I know your posts are about a month old and this is perhaps slightly off-topic now, but..

- Windows 2.03 (plain, no /286 or /386) already supported 256c and high resolutions (say 1280x1024) if a sutiable driver was used.
In fact, Windows 2.x had a completely different palette system than 3.x. It treated 256c as a high colour depth.
If you want to run Windows 1.x/2.x applications on 3.0 in colour, you have to use the Windows 2.x drivers and run Windows 3.0 with Real-Mode kernal (win /r).
Here's an example of Windows 2.x in 256c and 640x400 pels using the Paradise driver: https://www.youtube.com/watch?v=-YGx-KeujIU

- Windows 3.1 in Standard-Mode DOES support acceleration, high resolutions and colour depths, IF the driver in question is compatible with Standard-Mode (*.DRV drivers).
Unfortunately, most drivers from the mid-90s are made for 386 Enhanced Mode and use *.386 (VXD) driver model (which in turn is based on OS/2's LE EXE format).
At this time, 286es were already phased out, also. Your best bet will be drivers for veteran graphics chips.
Paradise PVGA1A/PVGA1B (WD90C00 or later), Tseng ET-3000, ET-4000, OAK OTI-67c, Trident 8900D, Realtek RTG3105..

None of those cards have GUI acceleration. I've tried to use Tseng ET4000/w32, ATI Mach64 and similar cards on a 286 and no drivers exist to support them fully. Standard VGA and 256c drivers work but are not accelerated...

The 8514/A driver should be fully compatible with Mach64. That works in standard mode apparently.

It sounds like a nice complex project and impressive. I suppose the chip choice has been made but what about older (than Trio) S3 based chips? The ones that were used on the VLB bus should or not compete with the GD5434?
Anyway I've the GD5429 on ISA with 1MB 60ns DRAM (even if my 386 mainboard seems to not like to run it stable I'm trying to understand why) and sure were fast cards but not that impressive on the vga output quality imho. I've switched it to a Wonder XL 28000-5 ATI card and the vga quality filters are much better on a LCD that usually are more sensible to the analog input signals than CRT were back in those times. Not to mention obviously PCI cards like the Millennium, so I'd imagine a similar design for the vga quality target.
Also maybe I'll say something obvious but I'll look on the PCB final thickness to have a perfect no-force installation on ISA connectors. Many cards I've tried had all different sizes of PCB layers and some were even almost difficult to install like it would maybe risk to "open" the whole connector, while others cards got installed as easily as a PCI one... so I suspect there was a margin of different number of layers and final sizes and I'd not make it too much thick nor too much thin...

Anonymous Coward wrote on 2020-06-22, 00:32:

The 8514/A driver should be fully compatible with Mach64. That works in standard mode apparently.

No, Mach64 isn't compatible with 8514/A.
Only Mach8 and Mach32 are.

That's interesting. I never actually tested it, but it doesn't make sense that ATi would remove 8514/A compatibility from the Mach64, especially considering how similar it is to the Mach32.

Mach64 isn't that similar to Mach32 - not only 8514/A compatibility was removed, but also CGA/Hercules compatiblity.
And some acceleration functions native to Mach32 are absent (or behave diffrently) in Mach64.
I reckon I once tried to run Windows 3.1 with Mach32 drivers on a Mach64 - and it didn't work.