"Hashwar": A Battle Royale for individual GOL configurations

[hash0]

Here is the distribution of Conway Glyphs according to their runtime (i.e. number of unique, non-repeating generations) that I have been looking into with the above code:

Total Attempts 368.258.500
Current Rate [patterns/s] 36.980
Elapsed Time [sec] 10.161
Generational Distribution
0-99 171.888.388
100-199 120.803.301
200-299 47.494.082
300-399 17.635.430
400-499 6.560.706
500-599 2.557.073
600-699 897.673
700-799 288.564
800-899 93.790
900-999 27.060
1000-1099 8.171
1100-1199 2.577
1200-1299 924
1300-1399 499
1400-1499 162
1500-1599 86
1600-1699 7
1700-1799 3
1800-1899 2
1900-1999 0
2000-4999 0
5000+ 14


So, on my Ryzen 5 9600X (6 Core) CPU, this code generates and explores (i.e. runs the Game of Lifes until they reach a halting state) roughly 37k "soups" PER SECOND.

[hash0]

I've started vlogging our discoveries here: https://www.youtube.com/watch?v=1AAYNnk6jps

[hash0]

Here is an updated histogram (the distribution of generational count for Conway Glyphs) after having looked at over 1 billion soups:

Total Attempts 1.094.912.000
Current Rate [patterns/s] 36.980
Elapsed Time [sec] 31.568
Generational Distribution
0-99 511.025.037
100-199 359.144.261
200-299 141.203.481
300-399 52.421.109
400-499 19.502.420
500-599 7.604.929
600-699 2.665.820
700-799 858.616
800-899 278.944
900-999 80.103
1000-1099 24.196
1100-1199 7.440
1200-1299 2.770
1300-1399 1.434
1400-1499 500
1500-1599 215
1600-1699 38
1700-1799 11
1800-1899 4
1900-1999 0
2000-4999 0
5000+ 31

The "new frontier" seems to be the 1800-1899 class (the longest-lived one running for 1,884 unique generations). I refuse to believe that this is the "End of Longevity", though, for the simple fact that the realm above 1,299 generations was completely invisible to us before we switched to more performant search algorithms. Why should we believe that ever rarer patterns do not hide in the "Conway Deep Field"?

Of the subset of 31 known patterns with more than 5,000 generations I looked at, all evolve from soups to the p5024 Agar pattern that was identified earlier - the longest running of those has 5,177 generations before it enters a strict loop (not just a shifted pattern). I surmise that we should therefore expect to find such "Agar class" Glyphs that run for 6,000 and more generations because we could argue that a Glyph might conceivably evolve into an Agar on the, say, 1,700th generation or above, running for a total of 1,700 + 5,024 = 6,724 unique generations.

I am proposing the following questions for active research:
1) Are there other Agar patterns apart from the p5024 with (7,7) shift on the 16x16 torus?
2) Is the histogram "continuous" from 1,900+ all the way up to 5,000+ or is there a natural "cut-off" for non-Agar Glyphs?

A billion patterns is a nanoscopically small fraction of all possible patterns on the 16x16 torus - I therefore chose to entertain the dream that there is a multitude of interesting and long-lived patterns out there that we have no hope of ever finding, even if we devoted all of humanity's computational resources from now until the heat death of the universe to the task. We are but wanderers on a beach, looking at individual grains of sand in a galaxy full of planets entirely made of beaches...

[NNlk05]

I see you had broken my record.
As soon as I get home I will use my gpu porgram to, hopefully, break it again.

[hash0]

Begun the Hash War has...

No, seriously: I sincerely hope that you DO break my record (Tännhauser Gate with 5,172 unique iterations) which would most likely mean you discovered another Agar pattern for the Conway Glyphs!

Have a look at the current Glyph Registry:
https://lifehashes.net/Y176/catalog.php ... C&limit=24

[NNlk05]

It's running!
I have 1100000 soups per second
I am pass 2 billion soups in 3 minuets
Will update once I get something

Update:
New agar?

Code: Select all

x = 16, y = 16, rule = B3/S23:T16,16
bbbobobbooobbobo$obobbobbbobbooob$boooooobobooobob$bobboobbbboboooo$obobbobbooobbbbb$boobbobobobobbob$bbbobbbboooobooo$boobboboboobbboo$oboboooboooooobb$oobbobooobbobobb$boooooobbbboobbb$obboobbboobobbob$obobbbbbboooooob$booooboobooobbob$oobbobboooboooob$bbobbbbobbbbbboo!

[hash0]

Congratulations on your monster of a GPU search engine! With this, we are surely going to see what's further out there in the Conway Deep field.

Yes, you appear to have discovered another Agar pattern for the Conway Glyphs: It evolves into a pattern with periodicity 63 at generation #46 which repeats exactly (no shifting of the pattern with each period, hence the Glyph evolution terminates by generation #109).

[bobinou]

Hello everyone
I am new here and just found a pattern that lasts for 5,237 generations before stabilizing.
String:

Code: Select all

x = 16, y = 16, rule = B3/S23:T16,16
2o2bo3bobo2b3o$bobobo4b6o$3o2bo2b3o$bo2b3o3bob4o$obobobobobob3o$bo5bob
obo$obob2o2bobo4bo$2b6o2b2ob2o$3bo4bobobo2bo$2o2bob3ob2o2b2o$4o2bo2b2o
2b3o$2ob7o4bo$o2bobo8bo$bo3b3obobobobo$o6bobobob3o$bo3b4ob4o!

Binary:

Code: Select all

1100100010100111010101000011111111100100111000000100111000101111101010101010111001000001010100001010110010100001001111110011011000010000101010011100101110110011111100100110011111011111110000101001010000000010010001110101010110000001010101110100011110111100

[NNlk05]

Hello everyone
I am new here and just found a pattern that lasts for 5,237 generations before stabilizing.
String:

Code: Select all

x = 16, y = 16, rule = B3/S23:T16,16
2o2bo3bobo2b3o$bobobo4b6o$3o2bo2b3o$bo2b3o3bob4o$obobobobobob3o$bo5bob
obo$obob2o2bobo4bo$2b6o2b2ob2o$3bo4bobobo2bo$2o2bob3ob2o2b2o$4o2bo2b2o
2b3o$2ob7o4bo$o2bobo8bo$bo3b3obobobobo$o6bobobob3o$bo3b4ob4o!

Binary:

Code: Select all

1100100010100111010101000011111111100100111000000100111000101111101010101010111001000001010100001010110010100001001111110011011000010000101010011100101110110011111100100110011111011111110000101001010000000010010001110101010110000001010101110100011110111100

Hello, and welcome to the forums!
This stabilises into the same agar as Tännhauser Gate and others.
I suggest checking out the wiki to learn more.

[hash0]

Welcome to the show!
I confirm Bobinou's entry to be another example of the Agar pattern we discovered in the Conway Deep field:

ITERATION: [5237]

ATTRIBUTES
85e666d52de85aba3b1a11bf145262d76c7e84802f8a63398da16aec9bb8abbf
a0702c1e1f270a3b455b1ae26499f9f17640bb4f386b4aca3848bac969726595
[STEPS:5237][MIN:14][MAX:126][PEAK:1258][TERMINUS:5024-ADIC]

Of note is the fact that this is the longest-lived known Conway Glyph to date, surpassing my own record holder "Tännhauser Gate" of 5,172 unique generations.

Kudos, Bobinou!

[hash0]

Care to elaborate on your custom-built code, Bobinou? I think your engine is even more performant than Vort's (who also took a shot at cracking my daily challenge website and, to date, still holds the all-time record)...

[bobinou]

I can open source it after a bit of polishing for anyone who is interested
That record will be broken ^^

[Vort]

... I think your engine is even more performant than Vort's ...

It is possible, sure, especially if code runs on GPU.
However, I want to note that my code calculates lifespan in a different way.
I thought it would be unfair to add oscillation duration to the lifespan.
Especially considering that on torus spaceships are also oscillators.
So my engine throws away results like this one: lifespan: 213, period: 5024.

[hash0]

Don't throw away oscillators! My Conway Glyph engine filters out periodic patterns, anyway!

[Vort]

Don't throw away oscillators!

This is what Methuselah article on LifeWiki says by the way:
"The time when a pattern is considered to have stabilized is commonly agreed upon to be the first generation such that the pattern can be resolved into still lifes, oscillators and escaping spaceships, ..."

My Conway Glyph engine filters out periodic patterns, anyway!

Do you mean that it will reject single "escaping" glider, but will accept it with extra dot added somewhere?

[hash0]

It means that my engine accepts ANY configuration as long as it does not repeat exactly - that's why the Glyphs based on the Agar pattern will eventually terminate under my rules even if the pattern could keep replicating itself infinitely.

[hash0]

Update on the state of the search for ultra-long-lived Conway Glyphs: Bobinou discovered today a pattern that runs for 2,021 ("Tantalus Harvest") unique generations. This significantly improves on my previous record holder 'Oklo' of 1,884 generations.

I shall give Bobinou access to my collection to register his Glyph:
https://lifehashes.net/Y176/catalog.php ... C&limit=24

These questions are still open, though:
1. Is the space between the "Conway Deep Field" (Glyphs of 1,000+ generations) and the "Ultra Deep Field" (5,000+ generations where the Agar-based patterns live) continuously inhabited? There is currently a gap between the known longest-lived non-Agar (2,021 generations) and the shortest-lived Agar (~5,000 generations).
2. All known Glyphs with 5,000+ generations derive from the same Agar pattern. Are there other Agars on the 16x16 torus?

[bobinou]

I made a quick, simple website to visualize the distribution and findings of my miner for anyone who is interested!
https://hallucinat.in/
Some interesting patterns are showing up - either Agar patterns above 5400 or quite a few in the 2000s range.
If needed i can also provide a open API access to all the datapoints.

Let me know what you guys think about this.

[hash0]

How many patterns have you been looking at now? My own Glyph Hunter engine looekd at over a billion Glyphs but I could not find any non-Agars that live for more than 1,884 generatiosn, so your discovery of 2,000+ Glyphs has significantly expanded into the Deep Field.

[bobinou]

Found the 2000+ patters after about 7 billion. In total it has been about 32 billion, 20 of them are visible on the website. I have a miner running 24/7 on a small raspberry pi like pc which runs at 100k iterations per second. From time to time it runs on my pc at about 2M/s, which will be apparent on the live counter.

[NNlk05]

Wow. I really need to work on my GPU based one...

[bobinou]

NLK05, I'm very interested on hearing more about your approach, because my current miner is a CPU only cpp implementation. I made a CUDA-based version for hash0's daily conway challenge, as hashing Suite 8+ scales better on GPUs (hitting about 700M hashes/s). However, I haven't managed to compute the games on it yet.
When disregarding the hash challenge my approach results in immense backpressure by the gpu, making the entire system slower than a pure cpu approach.

[NNlk05]

NLK05, I'm very interested on hearing more about your approach, because my current miner is a CPU only cpp implementation. I made a CUDA-based version for hash0's daily conway challenge, as hashing Suite 8+ scales better on GPUs (hitting about 700M hashes/s). However, I haven't managed to compute the games on it yet.
When disregarding the hash challenge my approach results in immense backpressure by the gpu, making the entire system slower than a pure cpu approach.

I think I published a older version of my script to this thread.

[hash0]

I'm currently working on a major overhaul of the Adversarial Conway Conflict engine - and this new version will be open for the public so all you Glyph Hunters out there can submit your longest-running patterns for epic battles.

Here is a preview of what it's going to look like:
https://youtu.be/P5XpHNXA-CA

[hash0]

The new battle engine has a "LIVE CELL PROFILE" that updates in real-time - it displays the number of alive cells at every iteration of the Conway Glyph.

With this it's easy to see that our "Conway Ultra Deep Field" Glyphs are all evolutions of the same Agar pattern we discovered:

Agar_Periodicity.png (886.86 KiB) Viewed 531 times