Bitcoin puzzle transaction ~32 BTC prize to who solves it

[Denevron]

can the progress made be used so that I don't always have to start the search from the beginning?

Yes.

You removed Xoshiro256plus why ?  

Because the Mersenne Twister performed better on smaller puzzles. Specifically, I tested it on puzzle 40.

================= WORK IN PROGRESS =================
Puzzle/Bits       : 40
Target Hash160: 95a156cd21b4a69...6864f4c8b82a82a
Prefix length     : 9 bytes
Mode                : Random
CPU Threads     : 12
Mkeys/s           : 35.95
Total Checked   : 118821993408
Elapsed Time    : 00:55:05
Start Range      : 8000000000
End Range        : ffffffffff
Progress           : N/A
Progress Save   : 5
Stride               : 1
================== FOUND MATCH! ====================
Private Key    : 000000000000000000000000000000000000000000000000000000E9AE4933D6
Public Key      : 03A2EFA402FD5268400C77C20E574BA86409EDEDEE7C4020E4B9F0EDBEE53DE0D4
Total Checked : 118875620444
Elapsed Time  : 00:55:06
Speed            : 35.9366 Mkeys/s

Try to solve puzzle 40 in Python using random mode in less than an hour. 

Code:

✅ Key found : 0000000000000000000000000000000000000000000000000000004b5f8303e9 -> 0b304f2a79a027270276533fe1ed4eff30910876!      
 
⏳ Full time: 2.70 sec.

here is the 40th puzzle, a python script 

that is not puzzle 40

Sorry, 39 puzzle)

[nomachine]

Elapsed Time  : 00:00:12

12 seconds

Congratulations.

Code:

✅ Key found : 0000000000000000000000000000000000000000000000000000004b5f8303e9 -> 0b304f2a79a027270276533fe1ed4eff30910876!      
 
⏳ Full time: 2.70 sec.

here is the 40th puzzle, a python script  

Yes, but in which context and which script? (I didn't mean a kangaroo.)

When I said RANDOM, I meant a random script:

import sys, secrets, secp256k1 as ice
while True:
    dec = secrets.SystemRandom().randrange(549755813887, 1099511627775)
    h160 = ice.privatekey_to_h160(0, True, dec).hex()
    if h160 == "95a156cd21b4a69de969eb6716864f4c8b82a82a":
        HEX = "%064x" % dec;wifc = ice.btc_pvk_to_wif(HEX)
        with open("KEYFOUNDKEYFOUND.txt", "a") as f:
          f.write(f'Private key (wif) Compressed : {wifc}\n')
        break

You'll be waiting a long time...

Hello everyone. I have published my optimized versions of VanitySearch (CUDA) with speed boost in case anyone is interested  

Thank you.  

[b0dre]

Hello everyone. I have published my optimized versions of VanitySearch (CUDA) with speed boost in case anyone is interested  

The "bitcrack" version is specific to the puzzle and allows searching for addresses and prefixes (compressed) within a given range. The speed is about 6900 MKey/s on a 4090 and 8800 MKey/s on 5090.

The second version, on the other hand, performs a standard search for vanity addresses (not just P2PKH compressed) but with the same optimizations in terms of math and CUDA code. Random searches with endomorphisms.

https://github.com/FixedPaul/VanitySearch-Bitcrack

https://github.com/FixedPaul/VanitySearch

Thanks!
why not multi-gpu?

[nomachine]

i would use....

The Cyclone script processes keys in large batches (POINTS_BATCH_SIZE = 256) for efficiency.
Processes hashes in AVX2-optimized batches (HASH_BATCH_SIZE = 8 ) .
Uses IntGroup for batch modular inversions.
Precomputes points in batches.

The simple division doesn't account for these batch boundaries, potentially causing:
Threads may end their ranges mid-batch.
Some keys near thread boundaries might get processed twice or missed.

But go ahead—open GitHub, clone the repository, and modify it. That’s the goal: getting everyone involved in improving the scripts, even if they seem useless. Why? Because sooner or later, as more people engage, someone will come up with a brilliant idea . Which doesn't even have to be related to solving puzzles.  

[Akito S. M. Hosana]

That’s the goal: getting everyone involved in improving the scripts, even if they seem useless. Why? Because sooner or later, as more people engage, someone will come up with a brilliant idea . Which doesn't even have to be related to solving puzzles.  

You mean this activates brain activity and exercises the brain, plus you learn a programming language. That makes sense. At least my head will grow, even if my wallet doesn't. 

[nomachine]

At least my head will grow

Buddy, unless you're a Marvel superhero or a bobblehead, your head stopped growing a long time ago.

[fixedpaul]

Hello everyone. I have published my optimized versions of VanitySearch (CUDA) with speed boost in case anyone is interested  

The "bitcrack" version is specific to the puzzle and allows searching for addresses and prefixes (compressed) within a given range. The speed is about 6900 MKey/s on a 4090 and 8800 MKey/s on 5090.

The second version, on the other hand, performs a standard search for vanity addresses (not just P2PKH compressed) but with the same optimizations in terms of math and CUDA code. Random searches with endomorphisms.

https://github.com/FixedPaul/VanitySearch-Bitcrack

https://github.com/FixedPaul/VanitySearch

Thanks!
why not multi-gpu?

The "standard" version allows multi GPUs

The "bitcrack" version does not. since the range size is a power of 2 the number of GPUs must be as well in order not to have overlapping keys. It would also be slightly slower because the threads of one GPU have to wait for the threads of the other to finish, which is a problem if the 2 GPUs do not have the same speed.

It was a choice to keep things simpler and a little more efficient

[Akito S. M. Hosana]

Speed a little too slow seems to be.

AMD 9950X with 32 threads, here 24 active.

It would be awesome to make Cyclone work on cloud server with 192 threads!...
I know there are other versions on github but they don't have random mode.

I achieve 40M keys/s, but after 25 minutes, it drops to a constant 37M keys/s. Is this normal on AMD?

I think he already has AVX-512, but he hasn't published it on GitHub yet. There are pictures of his screen showing 128 cores 

[brainless]

Hello everyone. I have published my optimized versions of VanitySearch (CUDA) with speed boost in case anyone is interested  

The "bitcrack" version is specific to the puzzle and allows searching for addresses and prefixes (compressed) within a given range. The speed is about 6900 MKey/s on a 4090 and 8800 MKey/s on 5090.

The second version, on the other hand, performs a standard search for vanity addresses (not just P2PKH compressed) but with the same optimizations in terms of math and CUDA code. Random searches with endomorphisms.

https://github.com/FixedPaul/VanitySearch-Bitcrack

https://github.com/FixedPaul/VanitySearch

Thanks!
why not multi-gpu?

The "standard" version allows multi GPUs

The "bitcrack" version does not. since the range size is a power of 2 the number of GPUs must be as well in order not to have overlapping keys. It would also be slightly slower because the threads of one GPU have to wait for the threads of the other to finish, which is a problem if the 2 GPUs do not have the same speed.

It was a choice to keep things simpler and a little more efficient

Cuda 8 version available?

[nomachine]

Speed a little too slow seems to be.

AMD 9950X with 32 threads, here 24 active.

It would be awesome to make Cyclone work on cloud server with 192 threads!...
I know there are other versions on github but they don't have random mode.

I achieve 40M keys/s, but after 25 minutes, it drops to a constant 37M keys/s. Is this normal on AMD?

I think he already has AVX-512, but he hasn't published it on GitHub yet. There are pictures of his screen showing 128 cores  

The performance drop you're seeing (from 40M keys/s to 37M keys/s) is pretty normal, depending on your system setup and workload.

After 20 minutes of non-stop running, your CPU might heat up and trigger thermal throttling, which causes a slight slowdown to keep things cool.

Yeah, I have the AVX-512 version, but it's even worse. If you're using a dedicated machine (like a root server), this aggressive script could totally fry the CPU because they usually have tiny stock cooler.

That's why I'm kinda iffy about uploading it – I don't wanna be on the hook if someone melts their $7-8K processor.

P.S. Cloud servers usually don't give full physical access to hardware – you’ll never hit max speed. Plus, they'll kick you out of there pretty quickly.

[POD5]

If you ask DeepSeek for

"Is it possible to control hardware temperature when running a script?"

you'll get some interesting answers... 

[Akito S. M. Hosana]

If you ask DeepSeek for

"Is it possible to control hardware temperature when running a script?"

you'll get some interesting answers... 

He told me I gotta use liquid nitrogen, lol.

[Dom1nic]

Code:

✅ Key found : 0000000000000000000000000000000000000000000000000000004b5f8303e9 -> 0b304f2a79a027270276533fe1ed4eff30910876!      
 
⏳ Full time: 2.70 sec.

here is the 40th puzzle, a python script 

that is not puzzle 40

Sorry, 39 puzzle)

did you get some tips from coffee grounds 

[fixedpaul]

Hello everyone. I have published my optimized versions of VanitySearch (CUDA) with speed boost in case anyone is interested  

The "bitcrack" version is specific to the puzzle and allows searching for addresses and prefixes (compressed) within a given range. The speed is about 6900 MKey/s on a 4090 and 8800 MKey/s on 5090.

The second version, on the other hand, performs a standard search for vanity addresses (not just P2PKH compressed) but with the same optimizations in terms of math and CUDA code. Random searches with endomorphisms.

https://github.com/FixedPaul/VanitySearch-Bitcrack

https://github.com/FixedPaul/VanitySearch

Thanks!
why not multi-gpu?

The "standard" version allows multi GPUs

The "bitcrack" version does not. since the range size is a power of 2 the number of GPUs must be as well in order not to have overlapping keys. It would also be slightly slower because the threads of one GPU have to wait for the threads of the other to finish, which is a problem if the 2 GPUs do not have the same speed.

It was a choice to keep things simpler and a little more efficient

Cuda 8 version available?

Nope sorry:(

[nomachine]

Added AVX2-256 vectorized copy for aligned chunks in the Montgomery multiplication routine.

https://github.com/NoMachine1/Cyclone/commit/51f6785ab4cd88df6763dc19bfc8a1df48ea7d97

[Akito S. M. Hosana]

Added AVX2-256 vectorized copy for aligned chunks in the Montgomery multiplication routine.

https://github.com/NoMachine1/Cyclone/commit/51f6785ab4cd88df6763dc19bfc8a1df48ea7d97

42 Mkeys/s here

[JackMazzoni]

What is the limit of target public keys in keyhunt? how many public keys.

[Akito S. M. Hosana]

This latest Cyclone is the most stable one so far. What is it doing? 

[Desyationer]

Hello everyone. I have published my optimized versions of VanitySearch (CUDA) with speed boost in case anyone is interested  

The "bitcrack" version is specific to the puzzle and allows searching for addresses and prefixes (compressed) within a given range. The speed is about 6900 MKey/s on a 4090 and 8800 MKey/s on 5090.

The second version, on the other hand, performs a standard search for vanity addresses (not just P2PKH compressed) but with the same optimizations in terms of math and CUDA code. Random searches with endomorphisms.

https://github.com/FixedPaul/VanitySearch-Bitcrack

https://github.com/FixedPaul/VanitySearch

Thank you for your work – it's truly impressive! The first program achieves a speed higher than any other solution I've seen. Even with a 33% power limit on an RTX 4090, it reaches around 2.3G keys per second. The second program delivers an even more record-breaking speed of about 4G keys per second under the same power limit. However, unfortunately, these impressive numbers are merely theoretical and not useful for solving puzzles, as the program does not support working with ranges.

I wonder if it is possible to implement Bitcoin address prefix searching not only by the starting characters but also by any other positions within the address. For example, searching for characters at the end, in the middle, or even a combined search where part of the characters are at the beginning, part in the middle, and part at the end, and so on.

[fantom06]

How were they able to figure out that this happened? They deliberately  things so that they don’t find the right range. I have been here for a long time, but I am only answering now.


1.01000000000000 * 147573952589676412928 * 1.00400000000000 = 1.49645890884035e20 => Hexadecimal: 0x81cc100e6afcd0000
1.02000000000000 * 147573952589676412928 * 1.00400000000000 = 1.51127533368036e20 => Hexadecimal: 0x83150dae3e6c50000
1.03000000000000 * 147573952589676412928 * 1.00400000000000 = 1.52609175852036e20 => Hexadecimal: 0x845e0b4e11dbc8000
1.04000000000000 * 147573952589676412928 * 1.00400000000000 = 1.54090818336037e20 => Hexadecimal: 0x85a708ede54b48000
1.05000000000000 * 147573952589676412928 * 1.00400000000000 = 1.55572460820037e20 => Hexadecimal: 0x86f0068db8bac8000
1.06000000000000 * 147573952589676412928 * 1.00400000000000 = 1.57054103304037e20 => Hexadecimal: 0x8839042d8c2a48000
1.07000000000000 * 147573952589676412928 * 1.00400000000000 = 1.58535745788038e20 => Hexadecimal: 0x898201cd5f99c8000
1.08000000000000 * 147573952589676412928 * 1.00400000000000 = 1.60017388272038e20 => Hexadecimal: 0x8acaff6d330948000
1.09000000000000 * 147573952589676412928 * 1.00400000000000 = 1.61499030756038e20 => Hexadecimal: 0x8c13fd0d0678c0000
1.10000000000000 * 147573952589676412928 * 1.00400000000000 = 1.62980673240039e20 => Hexadecimal: 0x8d5cfaacd9e840000
1.11000000000000 * 147573952589676412928 * 1.00400000000000 = 1.64462315724039e20 => Hexadecimal: 0x8ea5f84cad57c0000
1.12000000000000 * 147573952589676412928 * 1.00400000000000 = 1.65943958208039e20 => Hexadecimal: 0x8feef5ec80c740000
1.13000000000000 * 147573952589676412928 * 1.00400000000000 = 1.67425600692040e20 => Hexadecimal: 0x9137f38c5436b8000
1.14000000000000 * 147573952589676412928 * 1.00400000000000 = 1.68907243176040e20 => Hexadecimal: 0x9280f12c27a630000
1.15000000000000 * 147573952589676412928 * 1.00400000000000 = 1.70388885660040e20 => Hexadecimal: 0x93c9eecbfb15b0000
1.16000000000000 * 147573952589676412928 * 1.00400000000000 = 1.71870528144041e20 => Hexadecimal: 0x9512ec6bce8530000
1.17000000000000 * 147573952589676412928 * 1.00400000000000 = 1.73352170628041e20 => Hexadecimal: 0x965bea0ba1f4b0000
1.18000000000000 * 147573952589676412928 * 1.00400000000000 = 1.74833813112041e20 => Hexadecimal: 0x97a4e7ab756430000
1.19000000000000 * 147573952589676412928 * 1.00400000000000 = 1.76315455596042e20 => Hexadecimal: 0x98ede54b48d3b0000
1.20000000000000 * 147573952589676412928 * 1.00400000000000 = 1.77797098080042e20 => Hexadecimal: 0x9a36e2eb1c4328000
1.21000000000000 * 147573952589676412928 * 1.00400000000000 = 1.79278740564042e20 => Hexadecimal: 0x9b7fe08aefb2a8000
1.22000000000000 * 147573952589676412928 * 1.00400000000000 = 1.80760383048043e20 => Hexadecimal: 0x9cc8de2ac32228000
1.23000000000000 * 147573952589676412928 * 1.00400000000000 = 1.82242025532043e20 => Hexadecimal: 0x9e11dbca9691a8000
1.24000000000000 * 147573952589676412928 * 1.00400000000000 = 1.83723668016044e20 => Hexadecimal: 0x9f5ad96a6a0128000