Quote from: kTimesG on Today at 01:09:08 PM

Quote from: FrozenThroneGuy on Today at 10:11:40 AM

Hello guys!
My implementation of Pollard-rho algo:
https://github.com/Dookoo2/Mark1
38 minutes for solving 80 bits puzzle with half billion of DP, 14 minutes for solving 70 bits with 200 millions of dp.
AVX2 bloom filter, compact DP table, Brent loop detection and others features.
May be useful for somebody.
Have a nice day:)

Thanks for the code

I don't want to sound mean or anything, but let's do a quick and simple math exercise:

Device | TDP | Speed | Keys/Watt
Ryzen 9 7945HX | 55W | 160 M/s | 2.9 M/W
RTX 4090 | 450W |. 10800 M/s | 24 M/W

So it's more than eight times more costly to solve any puzzle on that CPU.

Moreover, RC claimed he got up to 13.8 GK/s on a RTX 4090, which, if true, would mean it's ten times cheaper, than running a Ryzen 9 7945HX at 100% peak power.

I would also say that solving 80 bits, with some proper Tame precomputed database (with way less than half a billion DPs) takes much less than 20 seconds, even on a budget GPU. There is a precise balance where having too many DPs (of lower bits) is not beneficial, and having fewer DPs (of higher bits) is better overall.

Of course, but:
1. I do not participate in the puzzle race:) For fun, not more.
2. Optimal amount of DP (for 135 and higher) is much more than we can collect and save in RAM.
3. You can start Mark1 with different DP bits, amount of DP, different jump size (k parameter) and others.
4. And I have gpu code too, but it is boring.

And by the way, 135 and higher are like a lottery ticket.