Pollard's Kangaroo Pool for solving the #130 Puzzle

[bitcurve]

I'm happy to announce that I have completed my work based on JeanLucPons' original source code and the results are at least 1.5x and up to 2x faster.
My implementation supports CPU only, Windows 64 bits, and any normal Unix-like OS should work. (Note: tested with ubuntu/Linux only)
Also support for arm64 CPUs is in the works.

You can verify the results using the following demo executable provided in our telegram group:
https://t.me/puzzlesearch

The full program that connects to the pool is also available on our telegram group.

The program is currently closed source - for one reason - I want to create a single pool for the 130th puzzle. Once I see people join, I will open-source the code on this github account: https://github.com/puzzlesearch.

When the reward is found, the rewards will be distributed according to the work done by each user. We will use IPs logged in our server as a proof of identity.

Thanks!

[AB de Royse777]

I was checking your Telegram group and found that you are suggesting to download the .exe file. I don't think anyone will be interested to download a .exe file from a brand new member of this forum knowing that there are people who are waiting for an opportunity to scam some of us members.



The Git page is empty too.

[bitcurve]

You are right; this is not recommended, due to the obvious reasons you mentioned - but you can use a VM or use VirusTotal or any anti-virus to check and see it's nothing bad. Anyways, I am going to release the source code of the client now.
https://github.com/puzzlesearch/kangaroo

For anyone who has doubts about the binaries - just compile yourself.

[bjpark]

I'm happy to announce that I have completed my work based on JeanLucPons' original source code and the results are at least 1.5x and up to 2x faster.
My implementation supports CPU only, Windows 64 bits, and any normal Unix-like OS should work. (Note: tested with ubuntu/Linux only)
Also support for arm64 CPUs is in the works.

You can verify the results using the following demo executable provided in our telegram group:
https://t.me/puzzlesearch

The full program that connects to the pool is also available on our telegram group.

The program is currently closed source - for one reason - I want to create a single pool for the 130th puzzle. Once I see people join, I will open-source the code on this github account: https://github.com/puzzlesearch.

When the reward is found, the rewards will be distributed according to the work done by each user. We will use IPs logged in our server as a proof of identity.

Thanks!

If you want to upload the file, please check it a few times before uploading it
Don't waste your time...
It doesn't work at all.

[NotATether]

Why would anybody be interested in a CPU version of Kangaroo now that Moore's Law has basically died out and NVIDIA RTX gpus are getting ever faster?

[ABCbits]

Also support for arm64 CPUs is in the works.

Now it's interesting. I know ARM generally is slower than x86 CPU, but people usually claim it has better energy/performance ratio. I wonder if has better energy/performance ratio than GPU.

[bitcurve]

Why would anybody be interested in a CPU version of Kangaroo now that Moore's Law has basically died out and NVIDIA RTX gpus are getting ever faster?

My CPU implementation produces on an i7-13700f the same performance as my GTX 1080 with JLP's code. CPUs are way more popular, and thus we can use many more computers if we focus on the CPUs and get better outcome.

[NotATether]

My CPU implementation produces on an i7-13700f the same performance as my GTX 1080 with JLP's code. CPUs are way more popular, and thus we can use many more computers if we focus on the CPUs and get better outcome.

Yes that's great and all, but the puzzles are getting harder.

Now that the previously considered "hard" puzzles like #64 have been solved, the difficulty of the next puzzle increases by a multiplication factor. This, I'm sure you already know.

I don't see anyone trying to attack #66 or #130 with the older GPU generations nowadays. Maybe it's because people are upgrading setups but it also must be understood that if you want to be able to crack the puzzle at the same rate as before, you will need more powerful hardware, otherwise it will take longer.

That's why they're mostly using 30xx and 40xx GPUs nowadays.

Your CPU may give you a good performance but you will need significantly more of them in order to get the same cracking time.

It helps that CPUs are getting cheaper but still you'd actually need to buy a new computer for each additional CPU unless there are multi-socket Xeons out there that people can benchmark and show similar performance to the 13700f.

[bitcurve]

My CPU implementation produces on an i7-13700f the same performance as my GTX 1080 with JLP's code. CPUs are way more popular, and thus we can use many more computers if we focus on the CPUs and get better outcome.

Yes that's great and all, but the puzzles are getting harder.

Now that the previously considered "hard" puzzles like #64 have been solved, the difficulty of the next puzzle increases by a multiplication factor. This, I'm sure you already know.

I don't see anyone trying to attack #66 or #130 with the older GPU generations nowadays. Maybe it's because people are upgrading setups but it also must be understood that if you want to be able to crack the puzzle at the same rate as before, you will need more powerful hardware, otherwise it will take longer.

That's why they're mostly using 30xx and 40xx GPUs nowadays.

Your CPU may give you a good performance but you will need significantly more of them in order to get the same cracking time.

It helps that CPUs are getting cheaper but still you'd actually need to buy a new computer for each additional CPU unless there are multi-socket Xeons out there that people can benchmark and show similar performance to the 13700f.

I was actually thinking about running this on a lot of not-so-strong computers, rather than focusing on running on a small amount of strong ones. For example, I have two old computers who are idle, I can use them for this project.

[bitcurve]

Also support for arm64 CPUs is in the works.

Now it's interesting. I know ARM generally is slower than x86 CPU, but people usually claim it has better energy/performance ratio. I wonder if has better energy/performance ratio than GPU.

Indeed. ARM is generally much more efficient than GPU, but for this task, i can't tell exactly, because i didn't benchmark it yet. remember that a GPU must be connected to a computer with a CPU to run properly. hence we need energy for the CPU and CPU memory as well.

[NotATether]

I was actually thinking about running this on a lot of not-so-strong computers, rather than focusing on running on a small amount of strong ones. For example, I have two old computers who are idle, I can use them for this project.

Alright, but it would help if you encrypted all the pool traffic with TLS, just so that you could run this on computers in different areas without having to worry about LAN headaches or about someone eavesdropping on the traffic and stealing he private key after your pool solves the puzzle.

Indeed. ARM is generally much more efficient than GPU, but for this task, i can't tell exactly, because i didn't benchmark it yet. remember that a GPU must be connected to a computer with a CPU to run properly. hence we need energy for the CPU and CPU memory as well.

I honk all the M1/2/3/4 chips in particular might be faster than all RTX GPUs at compute - I don't know. There's no device-to-host copy overhead at least.

[Akito S. M. Hosana]

Why would anybody be interested in a CPU version of Kangaroo now that Moore's Law has basically died out and NVIDIA RTX gpus are getting ever faster?

My CPU implementation produces on an i7-13700f the same performance as my GTX 1080 with JLP's code. CPUs are way more popular, and thus we can use many more computers if we focus on the CPUs and get better outcome.

I have a version from here:

https://github.com/mikorist/Kangaroo-256-bit

Puzzle 65 for ~50 seconds on AMD Ryzen 9 7950X