pretty cool

More fun stuff to run when I set up my SNOOP testbench. Although, do you think there's any point trying this with PCI cards?

maxtherabbit wrote on 2023-08-18, 23:38:

pretty cool

Heh, yeah, not sure why I got to geeking out over this, I guess it's a good little weekend project if nothing else.

Updated the code to measure generated RNG kbits/second performance, a coin flip run length test (frequency of runs of consecutive heads or tails), and added a visualization of the Monte Carlo pi search. The visualization should converge to display the top-right quadrant of a quarter-circle.

That visual image makes me think that maybe things are not quite perfectly random, those small horizontal clumps that occassionally appear feel a bit suspect. Oh well, better not look too deep into it 😜

Tried out an Acumos AVGA1 card and a Trident 8900C, and at least neither of those supported this. So it could be that this was just Headland engineers being super nerdy. Still would be cool to know if the original IBM VGA card had this, and if this works on EGA cards as well. (although now that I went and added Mode-X visualization into the program, running the EXE on EGA might not be all that happy)

Pierre32 wrote on 2023-08-19, 04:17:

More fun stuff to run when I set up my SNOOP testbench. Although, do you think there's any point trying this with PCI cards?

I am starting to get a feeling that this support would be pretty rare. But maybe there's another unicorn chip out there 😀

clb wrote on 2023-08-19, 09:20:

I am starting to get a feeling that this support would be pretty rare. But maybe there's another unicorn chip out there 😀

Well I will add my data to the research! I do have a few ISA cards to try while I'm at it too.

Oops, looks like the horizontal streaks came from a bug in the visualization code. I forgot to update the VGA Read Map Select when doing a readback from video memory. Updated that, fixed visualization does not have those streaks any more.

Googled up two different NIST Randomness test battery based testing frameworks,
- https://github.com/stevenang/randomness_testsuite and
- https://mzsoltmolnar.github.io/random-bitstream-tester/

and generated 8*1024*1024 random bits from three different generators:
- Borland Turbo C++ 3.0 rand() function,
- Headland's Light Pen hardware generator, and
- my own hardware random number generator that I dabbled together into CRT Terminator.

First as reference, Turbo C++'s standard random number generator. rand() test fails on the Runs Test on one of the suite, but passes on the other. To my understanding the two test suites should implement identical tests, since they are based on the same NIST test battery. So one of the test suites is buggy. For some reason, the online test suite gives an internal error on the last two tests as well. From what I gather out of all this, the rand() implementation in Borland Turbo C++ 3.0 is pretty good.

Then the Headland HT208 Light Pen Address based hardware RNG.

Too bad, it does not quite get through the test suites. It fails on 8/16 of the tests in randomness_testsuite, so passes 50% of them, random.. (that does make it more random than rand(), right? 😁 )

Then finally, my own hardware RNG implementation.

It passes all the tests in randomness_testsuite, as well as all the tests in random-bitstream-tester, except that the two last tests give a "Error", rather than "Failed". This gives an impression that those tests would internally fail, and not result in the random number generator not producing random values.

I have some noob questions:

Is it ok if i test my cards an post the results here?
Is there a range where it is most likely implemented? Only ISA ? Maybe VLB? Any specific chipset manufacturer?
Is there a real usecase for this (not only at the moment but later on)?

DerBaum wrote on 2023-08-19, 17:14:

Is it ok if i test my cards an post the results here?

Please do. What I like about the forum is that it is well searchable by both Vogons search and Google search, so accumulating a collective knowledge base here is useful to extend the information spread well.

DerBaum wrote on 2023-08-19, 17:14:

Is there a range where it is most likely implemented? Only ISA ? Maybe VLB? Any specific chipset manufacturer?

Gut feeling says that if any, the older ISA cards might be more likely to have it, since they would be closer to the era when EGA cards were still also in their prime. It could well be that only the IBM VGA card and Headland only ever had this.

DerBaum wrote on 2023-08-19, 17:14:

Is there a real usecase for this (not only at the moment but later on)?

Typically, such a hardware RNG can be used as a mechanism to seed a PRNG for game development, for game types where the game seed would benefit from being a high number of bits. For example, shuffling a card deck for poker or solitaire games - Freecell, Minesweeper, etc. A famous bug with competitive Minesweeper was that they used a narrow number of bits in the seeding of the rng, which limited the number of possible games, and generated memorizable cycles. https://minesweepergame.com/website/usman-lat … mineident1.html is a collection of blog posts that illustrate the limitation.

I am proud to present, very likely the first ever developed game that utilizes this hardware RNG feature: the EGA/VGA Hardware RNG Space Adventure! 😁

It provides about few trillion trillion unique game boards to play, you will never encounter the same game board twice.

Download it at https://github.com/juj/crt_terminator/raw/mai … DOS/bin/RNG.zip

It requires a graphics card that supports this Light Pen RNG. Absolutely wouldn't be able to be made to run without that kind of RNG source. Nope. Not. No way.

Assuming Headland's comment in their technical manual is correct, the game should only run on authentic IBM EGA & VGA graphics cards, or fully 100% IBM Compatible VGA adapters like Headland. It won't work on any of them Cheap Clones(TM) that did not bother to implement such a mission critical part of the IBM spec like Light Pen support. (or hmm, if I interpret Headland's technical manual a little differently, there is also a chance that it will only run on IBM VGA cards, but not on EGA cards because those actually do implement Light Pen support.. not sure which. Well, either it works or doesn't work on EGA)

Sorry in advance if anyone manages to actually get it to work. It is 2AM and it was made in a day. 😀

You maniac 😁 Assembling my test bench today.

Good luck with that :>

Is it supposed to not work on hardware that doesnt support it?
Or will it work without the RNG part?

Because it works on my SBC with integrated CHIPS graphics...

Ooh cool, you got it to work! Absolutely Awesome 😀

The game will not start up at all if the RNG is not supported, but it will print a message about no hardware RNG being available. So since it started up, it means it did find the RNG. Awesome, that is the second card that I find this feature being available on.

Something that would be cool if you can run the RNG.zip on that card and compare the text logs and the Monte Carlo picture against the ones I posted. I.e. do the results look as similarly random for you?

Also, it would be cool if you could run the SNOOP.EXE program on that card and post the results. Is it a CHIPS EGA or a VGA integrated graphics card? Do you have a picture of the graphics IC chip from the motherboard?

Hmm looks like that is a VGA card, given that I count there are 80x50 characters on the display. If I understand correctly, an EGA card would produce a 80x43 text mode display only.

Ok... Now i am confused...

Both CHIPS F65545 and F65548 seem to support this...

Thanks! Oww, the results are not so random. Awesome patterns though. That picture is looking directly into the heart of the running memory address counter inside the card. Visually glancing, looks good enough for the game though. That does make me wonder if the generation code could be modified to improve the randomization.

clb wrote on 2023-08-20, 23:12:

Thanks! Oww, the results are not so random. Awesome patterns though. That picture is looking directly into the heart of the running memory address counter inside the card. Visually glancing, looks good enough for the game though.

Yes. I actually love the "visuals". Awesome work.

On one card they are always "straight" and on the other not (but always similar shaped like in my pictures)... Really interesting.