Solving ECDLP with Kangaroos: Part 1 + 2 + RCKangaroo

[Ykra]

I updated the first part:
SOTA v2: This method has same complexity but uses only two groups of kangaroos.
So we can get same K=1.15 with only two herds: one wild and one tame.

Interesting, tested this method in my GPU solution and can confirm it gets the same K as my 3 group implementation based on your original SOTA.
Thanks for the continued contribution.

I've been (for fun) trying to push my 5090's GKeys/s, currently sat at a peak record (for me) of 11.6 GKeys/s.
I'm aware you've focused mostly on 4090's but have you tried to push a 5090 yourself and if so did you achieve a lot higher? (understand it's less bang for buck so was maybe classed as a waste of your time ^^)

[bebejj]

What are the reported numbers for a single GPU? I'd like to know the estimated maximum performance per GPU and see if there are any bottlenecks in the upgrades I'm making, since the highest I've been able to count is a single RTX 5080 GPU at 6000 MK/s.

[RetiredCoder]

Interesting, tested this method in my GPU solution and can confirm it gets the same K as my 3 group implementation based on your original SOTA.
Thanks for the continued contribution.
I've been (for fun) trying to push my 5090's GKeys/s, currently sat at a peak record (for me) of 11.6 GKeys/s.
I'm aware you've focused mostly on 4090's but have you tried to push a 5090 yourself and if so did you achieve a lot higher? (understand it's less bang for buck so was maybe classed as a waste of your time ^^)

Currently I have code that has about 18G/s on 5090 for SOTA. Sources are difficult for understanding a bit, also there are some additional requirements to build them, but I will prepare everything for publishing as open source with explanations because I want to make my ideas public.

[mjojo]

Interesting, tested this method in my GPU solution and can confirm it gets the same K as my 3 group implementation based on your original SOTA.
Thanks for the continued contribution.
I've been (for fun) trying to push my 5090's GKeys/s, currently sat at a peak record (for me) of 11.6 GKeys/s.
I'm aware you've focused mostly on 4090's but have you tried to push a 5090 yourself and if so did you achieve a lot higher? (understand it's less bang for buck so was maybe classed as a waste of your time ^^)

Currently I have code that has about 18G/s on 5090 for SOTA. Sources are difficult for understanding a bit, also there are some additional requirements to build them, but I will prepare everything for publishing as open source with explanations because I want to make my ideas public.

I feel this is a sign if P135 bits already solved by Mr. RC.. Congrats Sir

[Cricktor]

I feel this is a sign if P135 bits already solved by Mr. RC.. Congrats Sir

Please use your brain.

If RetiredCoder had already solved puzzle 135, why would he wait to claim the coins? There's no need to use hidden mempool service like slipstream.mara.com in that case, because the public key is of course already known. Any solver of puzzle 135 could broadcast the withdrawal transaction publicly. It would be stupid to wait to withdraw the coins. And I don't locate RetiredCoder in the stupid camp.

For puzzles like #71 and moderately up, where the public keys are unknown, a solver should definitely not withdraw by public transaction, only via above mentioned Slipstream service and non-public mempool!


Even with impressive performance on a 5090 as RetiredCoder claims for his code, you need a heck of a lot cards, time and energy and thus a hefty investment to solve #135.

[Ykra]

Currently I have code that has about 18G/s on 5090 for SOTA. Sources are difficult for understanding a bit, also there are some additional requirements to build them, but I will prepare everything for publishing as open source with explanations because I want to make my ideas public.

Very impressive, I was quite happy with my baby steps hitting 11.6 G/s (while keeping SOTA) haha.

Really curious to see where the largest gains stem from and particularly interested by "also there are some additional requirements to build them", this sparks some ideas that I will toy around with.