The Graphics Gremlin – An Open Source Retro ISA Video Card | Hacker News Reader

The Graphics Gremlin – An Open Source Retro ISA Video Card

github.com

130 pointszdw29 comments

I love these retro graphics card projects. This one is especially cool because it can do multiple standards on one card. Of course with an FPGA it is a bit hardware heavy. On the other extreme there are projects that provide VGA output with no special hardware at all - just using timers and GPIO pins.

Linus Akesson's bit-banger comes to mind, which does so with an Atmel ATtiny15. This microcontroller is - well - tiny, about so tiny:

Bit banger is built around an ATtiny15 microcontroller, which runs at 1.6 MHz and has 1 kB of flash ROM and a claustrophobic 32 bytes of RAM. In fact, those 32 bytes are the CPU registers. Only the most basic AVR instructions are supported; they occupy at least two bytes each, and can obviously not be compressed since they are executing from ROM, so a maximum of 512 instructions will fit inside the chip (fewer if static data is needed). [1]

Another cool project I came about recently, but haven't had time to try out, is bit banging VGA Framebuffer written in Rust. [2]

[1] https://www.linusakesson.net/scene/bitbanger/index.php

[2] https://github.com/thejpster/vga-framebuffer-rs

P.S. To give a bit historic context. I remember having an electronics book with a pattern generator made with only a PIC. Must have been the late 90s, so bit banging graphics is definitely not a recent development but it is still incredible to be able to generate these signals fast enough without special hardware and only in software.

EDIT: Thinking about it, I'm not sure anymore if the PIC project was VGA or PAL and I couldn't find the book quickly.

For designs that must interface with existing hardware with precise timings, using any kind of microcontroller bitbanging often becomes unfeasible. Linus Akesson's bit-banger and craft[1] only work because they drive all the the timings, and VGA is really permissive with timings.

And once you are past cheap Atmel or ARM micro-controllers, FPGA designs are often the most cost-effective.

This FPGA comes in at a unit cost of just $5.12 and allows a design of just the FPGA, the SRAM and a few level shifters.

Any design around a CPLD and/or raw 74 series logic would end up with a much higher BOM, and you would struggle to implement the scan-doubler functionality for 31khz output at a reasonable price point.

[1] Seriously impressive, check it out: https://www.linusakesson.net/scene/craft/index.php

The Parallax Propeller is one microcontroller that could (in theory) work for bitbanging precise timings even when something else controls the timing. Since each task can be put on a dedicated core, you could clock-count your assembly instructions to sync up with the right time windows, and offload any variable-time tasks to a different routine on another core.

The only caveat is access to the central memory store inside the microcontroller is round-robin (80MHz / 8 cores), so if the timing requirement was close to that limit (1/10MHz), you might have to do some gymnastics to make sure your routine grabs the right distribution of cores to get auspicious timing, and possibly reset and try again if that can't be achieved.

No the concept is not entirely new, especially since historically, wasting CPU cycles was often cheaper than adding framebuffer RAM.

The c. 1977 Atari 2600 worked in a very similar way, wasting 80+% of the CPU time just to draw the screen. That was seen as a good trade-off back then since the games were simple and not typically doing much else with those CPU cycles.

I think the trade-off was more cost than “we don’t know what we would do with these cycles” (https://en.wikipedia.org/wiki/Television_Interface_Adaptor#B...)

If you look at https://www.jcmit.net/memoryprice.htm, 1kB RAM cost about $25-$50 at the time. The Atari 2600 sold for $199. The 6502 CPU, at $25 was too expensive for that (https://en.wikipedia.org/wiki/Atari_2600#MOS_Technology_6502...)

Of course this means the game had to be simple because there simply wasn't enough CPU time available to do anything complex.

Even further to that end of the spectrum is "The World's Worst Graphics Card"[1] on a breadboard with mostly discrete logic. Fascinating.

[1] https://www.youtube.com/watch?v=l7rce6IQDWs&vl=en

This is great, I've been wanting an ISA card with an FPGA on it for a long time so that I can experiment. I've ordered a few of these PCBs to build some myself. Its very kind of the author to give the design away for free. I also really like the Lattice FPGAs for their FOSS tool-chain, which is a breeze to work with.

Different issue, but related to video: if you build a server around high-end Ryzen chips, there is no onboard video. I could not find a no-gpu card that was a good fit for this scenario. There may be a market for a low-feature, low-tdp PCIe graphics card.

Sure, they exist, I believe you can still buy a geforce GT 710 for about 50 quid. I have a ryzen server at home, my old one had a 710 in it, but I upgraded to an asrock rack mobo that has an integrated IP kvm / ipmi bmc, which is a pretty good option for a real server (mine is just for home though).

Funnily enough, I have a GT 710 kicking around for this purpose. But my point - a hardware firm should be able to retail a no-feature, no-gpu PCIe VGA card with no passive power draw for a good deal less than 50 quid.

Those asrock rack units look like quality, and do have onboard video and IPMI. But they cost a lot more than a B450m pro4. Part of the magic of Ryzen is that you can now get ECC-capable systems at consumer prices, apart from the cost of the RAM itself. This has not been the case since the 486 era.

This would be quite a niche item, which gets in the way of it being a cheap commodity. Might be some chinese manufacturer that does one around an aspeed chip (or a clone of it) or something, but even there it seems like old mainstream chipsets are more common in the cheap offerings.

And a low-end GPU that's not doing anything is also quite power-efficient nowadays, so you don't save much in that regard.

ASrock sells an m.2 graphics card witha a Silicon Motion VGA chip with only 16MB VRAM[0], this should be enough for your needs.

[0]https://www.tomshardware.com/news/asrock-m2-graphics-card

I bought a pair of passively cooled Geforce 210 cards about a decade ago. I put one in my basement server (the other one went to my parents PC). On the rare occasion I'm in front of it, it puts text and GUIs on a CRT, and if I want to push it, video.

That's why Nvidia just launched GT 1010: https://www.tomshardware.com/news/nvidia-gt-1010-released

30w TDP, probably lot less idle consumption, can fit to low profile designs, can be cooled passively, and costs probably around $50.

I'm giggling like an idiot at the annotations on that PCB. Very clever.

It's lovely.

SPI connector: "I SPI WITH MY LITTLE I"

Resistor bank: "YES IT'S A RESISTOR LADDER DAC"

Voltage regulators: "POWER HOUSE"

ISA connector: "SLOW, OLD COMPUTER" (arrow)

Clock modules: "THE TICK" and "THE TOCK"

DIP switches: "MYSTERY SWITCHES"

Composite connector (not populated): "BLURRY PICTURE"

VGA connector: "REGULAR OL' VGA"

TTL monitor connector: "TTL MONITOR"

Did I get all of them?

"MADE WITH <3<3" (but with graphical hearts) "FYOOZE"

I have a stupid idea: a video card, not necessarily EGA/CGA, that instead of a VGA connector it has a ethernet port and it can be accessed through a VNC type protocol. Does this weird thing exist?

Lots of server IPMI interfaces do this. IDK of an add-on card for old computers though...

Not a video card, and HDMI instead of VGA, but I came across these recently:

- https://pikvm.org/

- https://mtlynch.io/tinypilot/

For PCIe there are such things for remote control of servers. (A lot of them are specific to manufacturers and won't work in other systems, but not all) E.g. Asrock Rack makes one called "PAUL".

Typically when they do, they're built in to server chassis as a remote management/LOM tool. A few such add-on cards have been made over the years, though -- the PC Weasel 2000 was quite a nice card back in the day: https://web.archive.org/web/20110406024209/http://www.realwe...

Others like the Enidan J1 were specifically targeted toward people building clusters out of commodity hardware: https://web.archive.org/web/20010511215150/http://www.enidan...

If you like this project and are on Twitter, you might want to consider following @TubeTimeUS who besides this posts all manner of retrocomputing projects through all stages of the building, testing, and design tweaking process.

Very welcome. It is nice it can provided 31KHz+ output.

If I find something's missing and could likely be easily added, it'd be a digital (e.g. HDMI) output.

It'd save many people the step of having to run the output through an OSSC.

Had to look up

>Creative Commons BY-SA 4.0 declared one-way compatible with GNU GPL version 3

>Compatibility means that a person can now take a work they received under the terms of CC BY-SA 4.0 and then distribute adaptations of that work under the terms of GPLv3.

https://www.fsf.org/blogs/licensing/creative-commons-by-sa-4...

Kind of strange to go through all this work to allow CGA and EGA to use the VGA port, but not support the VGA modes.

Did the FPGA just not have enough block ram to do this?

Duplicate: https://news.ycombinator.com/item?id=27192334

wrong way around, this is the older submission. (and at least right now the one with the better ranking)

Reply on news.ycombinator.com