1
$\begingroup$

so I have been working as theoretical physicist for my university. In my research i found a way to simulate a system - that yields a bitstream - which I then linearized - and with the use of AMD RocM and a few custom GPU Kernels - I was able to generate a Bitstream of randomness with about 200MB/s. Of course I have an initial seed - but the Bitstream somehow passes all tests i threw at it. And I couldnt find anything comparable in the literature.

For the BigCrunch Test I ran about 4TB - and the results were good. IF I were to sell 4TB worth of random bits - AWS has services that sell RNG bits. And if i were to sell my external HDD with 20TB worth of RNG - that would be equivalent to very much moneys. But I know - most of the price for sure lies in the reliability.

And all of that - without using Von-Neumann correction. So 100% yield. If I use a hardware noise card - and just overlap the signal - one would have mathematical perfect entropy. So the generated bitstream of my generator - is mathematical somewhat perfect. And overlapping it with physical noise - makes any reconstruction impossible.

I tested - NIST - AIS - Bigcrunch/crunch/smallcrunch - Dieharder - everything passed. I made some "Entropy tests" and the bitstream yielded 99.99%

If anyone has the ressources to help me monetize this thing or atleast help me to conclude that it is not as much worth as I am hoping - I would be grateful.

Improve this question
New contributor
REE is a new contributor to this site. Take care in asking for clarification, commenting, and answering. Check out our Code of Conduct.
$\endgroup$
9
  • $\begingroup$ Did you compare your solution with other non-cryptographic GPU-based PRNGs? In my benchmark, CPU-based non-crypto PRNGs can be faster than 10GB/s. The fastest crypto PRNG (ChaCha8) is around 1GB/s. $\endgroup$ Commented 15 hours ago
  • $\begingroup$ Here is the catch - i can also get up to 1GB/s - but i can vary the computation speed. So it is like a strong extractor - which "strength" can somehow scale by some variation of my input - in words of a physicist. And it runs on my GPU - the custom Kernels for video games have gotten so advanced - that you can finally do something more complicated than just matrix multiplication. A bit of research yielded - that the algorithm at runtime operates at a space of (2*N)! - yes - factorial. So for "small Systems" so to say - we can use GPU. At some size - i need the 128MB 3DVcache of r9 9950x3d $\endgroup$ Commented 15 hours ago
  • $\begingroup$ So If you would build a custom System - with a very large V3 Cache - one could take advantage of the fact - that the algorithm scrambles the input - with a key of the input size of (2N)! - yes factorial. But the input space needs to be sampled with another PRNG - that then chooses one element of that (2N)! input space. And this is pure insanity. $\endgroup$ Commented 15 hours ago
  • $\begingroup$ Because all of that - just to decode - one single bit. ONE BIT. And then you sample another version of the PRNG - that chooses one element of the input space - and then this algorithm - which I found in a few Textbooks about Quantum-Field-Theory - so Quantum mechanics x Generalized relativity yield QFT so to say. And the mathematics is a bit special. I was the only physicist of my year - that also voluntarily - did all the theoretical mathematics and computer science courses. $\endgroup$ Commented 15 hours ago
  • $\begingroup$ I used a paper from 2022 - of nvidia - that used an algorithm that used bijection algebra to represent this input space. My prof at my Uni of algebra was Fourier - the second man to become a Professor as a "Fourier" - because the OG. Fourier of which is all the Fourier analysis stuff - is dead since 200 years. But in his family something something relatives - and today another fourier doing a class about Linear Algebra. $\endgroup$ Commented 15 hours ago

3 Answers 3

Reset to default
3
$\begingroup$

... or at least help me to conclude that it is not as much worth as I am hoping - I would be grateful.

That, I can tell you that - it's not worth what you are hoping.

What you have is essentially a pseudorandom number generator - it takes a seed and determisticly generates a long sequence of random looking bits.

We have a number of free good ones that we believe are cryptographically secure (that is, we believe that, even given the design of the system, an adversary cannot distinguish that output from a truly random output), and so what you have would not be considered interesting.

And no, there is no one who is willing to pay someone else for a collection of supposedly random bits.

There are also people who sell physical random sources (these days, generally based on Quantum Mechanics) that can generate output; sometimes at a very impressive pace. Even if you develop something based on a physical system (as opposed to simulating it) and try to sell that, you'd still have a hard road - you would need to compete with those existing competitors (and I suspect they're having a hard time finding customers)

Improve this answer
$\endgroup$
1
  • $\begingroup$ Yeah thats what i thought... Well atleast I can use it to use it on my own simulations. Turns out the observables i am dealing with are so sensitive to the random number generation - that I wanted to build one that conserves coinflip statistics without needing for Von-Neumann correction... I will publish it on GitHub then - for anyone to use. If someone can do something commercial with it - will maybe let me know. $\endgroup$ Commented 19 hours ago
1
$\begingroup$

Interesting post. Could you share a sample bitstream, code, or a reference to the algorithm? Without verifiable data, its hard to evaluate the method. If you have a GitHub repo or a paper, that would help a lot.

Improve this answer
$\endgroup$
2
  • $\begingroup$ A sample bitstream would be useless to evaluate the method (unless the method is hideously bad, which we can assume from his testing is not) $\endgroup$ Commented 10 hours ago
  • $\begingroup$ Fair point. I was just curious about the method. Maybe something new will come up. $\endgroup$ Commented 9 hours ago
1
$\begingroup$

To properly evaluate your RNG, the community would need more than just test results. A sample bitstream, the algorithm name, or a reference to a paper would allow independent verification.

Without such data, its difficult to assess whether this is a new contribution or a variant of an existing PRNG. If you have a GitHub repository or a technical report, please share it.

This would also help address the commercial aspects you mentioned, since reliability and transparency are key for any RNG product.

Improve this answer
$\endgroup$
3
  • $\begingroup$ Yes - I just finished the Amazon AWS implementation. But the technical details are a bit hard to explain. One needs background knowledge in statistics - but also partial differential equations - complex analysis and quantum field theory. $\endgroup$ Commented 9 hours ago
  • $\begingroup$ I worked out many details but my problem I do not have much time because i am borderline homeless. My mom kicked me out when i was 18 - and told me i could stay for a bit - so I worked overtime to get everything in line to monetize a bit so I can move out again. I will try to minimalize the mess of code i have currently. But hey the code works to 100% my CPU and GPU. Then I will upload the results - and create a landing page for what exactly it does. Just for the meme - I bought an RTX 5050. Because each bit - is more or less a coinflip - a 50:50. $\endgroup$ Commented 9 hours ago
  • $\begingroup$ Got it. If you upload code or a paper later, I will take a look. I like non-standard approaches myself. Good luck with everything $\endgroup$ Commented 9 hours ago

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.