Bitcoin Forum

I am a total newbie with GPU calculations so please be patient with my questions.

I have Nvidia Quadro P4000. I need to find private key from public key 256bit ecc.
The Kangaroo seems to be unable to cope with this task. I know it can be supported by GPU but here I do not understand how much of GPU is really used by Kangaroo. What I have seen here it uses maximum a few of tens of GPU threads. But gpu has got thousands of them? Also to be sure Kangaroo is bugless is a bit fantasy. Also difficult to debug. Does Nvidia card need monitor to be connected or to have that special dongle virtual monitor from Aliexpress? How is computer creating screen if nvidia is used for calculations?

So I have this idea, please try to understand and estimate if it is feasible? I want to use whole gpu, 1024 threads or more. Each thread would search in a brutal way the private key in very simple manner: instead of multiplying generator point every time with new possible private key I would start every thread on its own ?? -bit space. It would start from initial "private" key and continue with adding generator point, this would be much faster. After adding generator point (+1 gp) gpu will also check X coordinate if it fits already to public key. Easy to debug. So I would have 1024 gpus performing simple task adding generator point and compare just one coordinate, say X. If it would work but one gpu P4000 is not enough maybe GPU cloud rent for 1000euro for 1 month might be enough? At this moment I have no idea how much time would gpu need to perform one iteration step so I welcome any estimations from you.

It's not feasible, not with that GPU, not with the entire cloud running for years with current technology.

"If you could do that, we won't be here"


Maria Olivia Wan

but 1024 gpus is alot. it might be worth a try.

It isn't. It doesn't matter if he has every GPU in the entire world starting at different private keys and adding one each time as he proposes to try to find a specific public key. Even if he had 1 trillion GPUs and each one could check 1 trillion private keys a second with no duplicates, and he had done this every second since the birth of the universe 13.7 billion years ago, he would only have covered somewhere in the region of 0.0000000000000000000000000000000004% of all possible private keys.

Until (and if) sufficiently advanced quantum computers come along, it is impossible to calculate a private key from knowledge of only the public key.

Do you have the formula for what it takes to use Pollard's kangaroo algorithm to find a private key with given bits of entropy x compared to a second private key with x + 1 bits of entropy (the expected value of both)? I have been reading about this algorithm recently, but the math is a bit over my head TBH.

If the answer is 2x, then there is no way using a Kangaroo program is possible to find a private key with 160 bits of entropy. If the answer is something less than 2x, and there is a way for the algorithm to scale (this is unclear), it may be possible.

But he's talking about renting over 1000 GPUs. surely something good would come of that. that's alot of horsepower. have you actually tried what you're talking about because otherwise I'm not sure how you could know for sure.

also this doesn't have anything to do wtih the OP but I did find it noteworthy that there are apparently 2 exascale computers now in china. I bet if they threw those things into brute forcing bitcoin private keys they would get some pretty fast.

Easy way for estimate that should not be over your head:

Expected group operations = avg. est. rangewidth / 2 + 1

120 bit = 119 / 2 + 1 = 60.5
121 bit = 120 / 2 + 1 = 61
122 bit = 121 / 2 + 1 = 61.5
160 bit = 159 / 2 + 1 = 80.5

As all things, cracking a 160 bit or 125 bit or 256 bit private key is possible with kangaroo...it all comes down to time and how many GPUs/CPUs are working together.

Brute force 256 bit key using 1000 GPUs...longer than the universe * universe

Kangaroo a 256 bit key using only 1000 GPUs...longer than the universe

You do not have to try renting 1000 GPUs to get an estimate on run time.

256 bit = 255 / 2 + 1 = 2^128.5 group ops to solve key
Now take 1000 GPUs and let's say their speed is 2,000 MKey/s: 1000 * 2,000,000,000 = 2^40.8

so 2^128.5 / 2 ^40.8 = 2^87.7 seconds to solve.

what about the exascale computer in china and they said there's one that is 100 trillion time faster now. i'm sure that can do it but they got better thing to do than worrying about bitcoin.

CUDA or not is not the issue, that's just a constant factor, a multiplier.
Kangaroo is still an exponential algorithm in the problem space, just stochastic. It won't provably solve any key in a normal lifetime with all the hardware in the world, with current technologies.