noshutdown wrote on 2026-03-04, 04:16:

i am not a big fan of monochrome, but remembered that in the 1980s there was some hack getting cga games running on hgc cards.
can someone explain how it works, and is it possible to merge it into the Herc9x windows drivers to get 320*200 or 320*240 fullscreen mode for games?

I vaguely remembering using hercules.com to play one of those. There's a review of some TSRs [1]. I don't know how they work; I only can theorise that they wrap around BIOS calls. Maybe they even periodically grab the CGA memory address space, if there is anything, and copy/transform the contents over to Hercules address space, but that wouldn't work if the CGA memory address space is empty.

[1] https://www.dosdays.co.uk/topics/cga_simulato … or_hercules.php

CGA/MDA/HGC test program

Intercepts writes to the CGA space instead writing to MDA space

rmay635703 wrote on 2026-03-04, 23:32:

Intercepts writes to the CGA space instead writing to MDA space

You can't intercept memory access with the tools you have on an IBM XT. There is no on-processor MMU (which appeared in the 386 an can be programmed by a virtualization manager like EMM386), and there is no external MMU as well. Instead, the Hercules card has 64KB of memory, and there is a control bit that enables the second 32K being visible on the ISA bus exactly in the address range of the CGA card. Furthermore, there is another control bit that chooses whether the first 32K (in MDA range on the ISA bus) or the second 32K (in CGA range on the ISA bus) should be displayed. So if you configure the card into a graphics mode that uses 80 bytes per scanline (that is 640 monochrome pixel, as CGA does), has banks of 8K (as CGA does, and yeah, Hercules banks are always 8K as well, so no need to configure anything), and displays the contents of the "upper half" of Hercules memory, this is directly CGA compatible without any kind of "interception".

mkarcher wrote on 2026-03-05, 00:38:

You can't intercept memory access with the tools you have on an IBM XT. There is no on-processor MMU (which appeared in the 386 an can be programmed by a virtualization manager like EMM386), and there is no external MMU as well. Instead, the Hercules card has 64KB of memory, and there is a control bit that enables the second 32K being visible on the ISA bus exactly in the address range of the CGA card. Furthermore, there is another control bit that chooses whether the first 32K (in MDA range on the ISA bus) or the second 32K (in CGA range on the ISA bus) should be displayed. So if you configure the card into a graphics mode that uses 80 bytes per scanline (that is 640 monochrome pixel, as CGA does), has banks of 8K (as CGA does, and yeah, Hercules banks are always 8K as well, so no need to configure anything), and displays the contents of the "upper half" of Hercules memory, this is directly CGA compatible without any kind of "interception".

so is it possible to implement such a hack in that herc9x driver, to provide a 320*200 mode faking vga for games?
this is not expected to work in dos, but windows games are supposed to send rendered data to video driver rather than hardware, so there is room for hacking as long as the resolution can be displayed.

You would have to configure the CRTC for 640x400 line mode with the same horizontal line rate as the monitor expects, which some googling suggests can be done. There's some discussion in the comments on the post here: https://nerdlypleasures.blogspot.com/2014/02/ … aspect.html?m=1

Then it's a matter of dithering the 256 color image down to monochrome at 2x the vertical and horizontal resolution in some way (accounting for the palette colors) and copying into the Hercules buffer. Apparently the DOS port of Super Mario 64 of all things does this. There are several approaches you can take for dithering, ordered dither or error diffusion, but ordered dither is probably more appropriate so one pixel changing doesn't shift the whole pattern down the screen.

Might want to look at the dithering methods used for Wolfenstein 3D CGA port for fast routines. Just an idea that pops into my head for such stuff.

onethirdxcubed wrote on 2026-03-05, 03:28:

Apparently the DOS port of Super Mario 64 of all things does this.

https://www.youtube.com/watch?v=4qXImwCfZco

But note that the CPU is Athlon XP 2600+.
No way for real-time dithering with typical Hercules-era CPUs, ie. 8088..286.

onethirdxcubed wrote on 2026-03-05, 03:28:

You would have to configure the CRTC for 640x400 line mode with the same horizontal line rate as the monitor expects, which some googling suggests can be done. There's some discussion in the comments on the post here: https://nerdlypleasures.blogspot.com/2014/02/ … aspect.html?m=1

You can't get to 400 visible lines at standard MDA/Hercules frequencies. That thread suggests a CRTC configuration that runs horizontal at around 19kHz (instead of 18.4kHz) and a vertical at around 45Hz (instead of 50Hz), which might be OKish for the original 5151. Contrary to popular belief, the 5151 does not blow up if driven at an excessively high horizontal rate, but if driven at an excessively low rate. The is because a too low horizontal rate will cause the energy stored in the flyback transformer core each line to be bigger, resulting in a higher flyback voltage (which might kill things), and in the worst case saturate the flyback transformer, causing an uncontrolled current rise in the horizontal output circuit (which again might kill things). Trying a 5151 at 19kHz thus is not as dangerous at it might initially sound.

In an earlier VOGONs discussion, I actually tested a basic C64 emulator that uses dithering from 320*200 to 640*400 on a Hercules card, and it worked perfectly on my multisync TTL monitor, see Re: So many choices so little time - Hercules graphics card decision (cloned title :-)) for a photo.

mkarcher wrote on 2026-03-05, 13:59:

You can't get to 400 visible lines at standard MDA/Hercules frequencies.

Yes you can - even to about 700 lines - using interlace.
That's how the 640 x 400 tweak works, and there was at least one card providing 720 x 700 mode on MDA monitors - Tecmar Graphics Master.

Grzyb wrote on 2026-03-05, 14:23:

mkarcher wrote on 2026-03-05, 13:59:

You can't get to 400 visible lines at standard MDA/Hercules frequencies.

Yes you can - even to about 700 lines - using interlace.
That's how the 640 x 400 tweak works, and there was at least one card providing 720 x 700 mode on MDA monitors - Tecmar Graphics Master.

I forget the brand but one off brand Hercules display claimed 120 columns of text on your ibm 5151.

That is likely around 1080x350

rmay635703 wrote on 2026-03-05, 20:08:

I forget the brand but one off brand Hercules display claimed 120 columns of text on your ibm 5151.

That is likely around 1080x350

I know about Eizo MD-B05 - it can do 132 columns with 8-pixel-wide characters, therefore 1056 x 350.
With 9-pixel-wide characters, it should approach 120 columns.

Anyway, with analog signal, the monitor in theory has unlimited horizontal resolution.
Vertical resolution is strictly limited by HSYNC and VSYNC frequencies, and can only be doubled using interlace.

Grzyb wrote on 2026-03-05, 20:27:

Anyway, with analog signal, the monitor in theory has unlimited horizontal resolution.

True. Yet the bandwidth of the video amplifier and the sharpness of the beam is limited, turning high frequency horizontal structure into uniform gray at some point, so no unlimited number of discernable vertical line pairs. The limited number of line pairs limits the amount of characters per line that are readable, so I expect it doesn't make sense to try to display considerably more than 132 characters per line on a 5151.

On the other hand, you do get nearly infinite resolution in positioning where a line with a given sharpness and width can be located. This is especially true on monochrome monitors that don't have a dot mask.

Depending on what you are trying to do, that might be desirable. (Smudgy uniform grey).

If you do that with interleaved modes, you can get something approximating grey shading on odd scanline, with larger, sharper dots on even, and rely on phosphor persistence to avoid flicker.

In the changelog of FastDoom there's a note that 640*400 Progressive works on a HGC and original IBM 5151 while 640*200 (I'm not sure if interlaced or progressive) only worked in emulators and was removed because it would not sync on the 5151.

My bigger question though is do Windows 9x display drivers have any control over what DOS does with the display when in full screen mode? I thought it just saved the states of all the registers and then let the DOS program do what it will, which is why full screen DOS doesn't work on a secondary monitor.

e: Using VxD services you can probably map the VGA region to somewhere else in memory and capture all the IO Port writes, but there's no built-in support for it and you will have to make a VGA register simulator (or copy one from DOSBox or 86box or something) to handle anything more complicated than standard Mode 13h. I don't have a Hercules card and haven't tested Herc9x so I don't know if DOS games which already have Hercules support work normally in that yet.

mkarcher wrote on 2026-03-05, 21:15:

Grzyb wrote on 2026-03-05, 20:27:

Anyway, with analog signal, the monitor in theory has unlimited horizontal resolution.

True. Yet the bandwidth of the video amplifier and the sharpness of the beam is limited, turning high frequency horizontal structure into uniform gray at some point, so no unlimited number of discernable vertical line pairs. The limited number of line pairs limits the amount of characters per line that are readable, so I expect it doesn't make sense to try to display considerably more than 132 characters per line on a 5151.

On the other hand, you do get nearly infinite resolution in positioning where a line with a given sharpness and width can be located. This is especially true on monochrome monitors that don't have a dot mask.

On an actual 5151, I think no matter how sharply focused the beam is, there's gonna be phosphor bleed, where the glow of the phosphor itself has a radius in which it will cause the surrounding phosphor to light. Thus reaching a point where a 101010101 pattern looks no different than 111111111 no matter if the amp and sharpness result in a clean signal still. Lower persistence monitors might go a bit further, but would eventually hit this limit. Though whether you hit the bandwidth limits of the amp or the phosphor first is open to speculation.

Edit: If there is minimal capacitance and inductance filtering or ringing effects, I think the absolute worst transistors might do 120mhz, average general purpose might do 150-175mhz , if actually selected specifically for high bandwidth video amplification by a belt and braces engineer, then 175 to 250, but I doubt anything higher would be cost justified.

Edit2: annnnnd transistor performance is probably irrelevant unless you're trying to do 1024x768 at 72hz or 4k pixels per line or something.