Solving ECDLP with Kangaroos: Part 1 + 2 + RCKangaroo

[kTimesG]

Where is any multiplication used here?

Field multiplication is heavily used during any group operations. I never mentioned "scalar multiplication".

- output file(s) should be in the format of (Y value) space (offset from given startpoint) all values should

Geez. It's enough that the storage for 135 is in the order of hundreds of gigabytes when actually using a database, now you want it in hex text format, one line after the other. Great, I hope you have a really good and spacious NAS with many many TB of available storage. Good luck at parsing such files as well.

[n1ghtowl]

I know it's a large amount of storage. The text files just allow me to import the data  for  further  analysis. With other software. How about converting current output file to said layout, I'm sure someone even myself could then edit to my preferred storage method. I'm sorry for my ignorance. As I said I am very inexperienced with both Cuda and c÷+.

At what DP would 135 be expected to be how much storage? ie

  I also forgot my current system allows the negative of the DP (IE Y point (mod 0x100000) == 0xFFFC2F). I hope DP =6 is understood correctly or is it DP= 24 . I understand DP 6 is not sustainable, I am using it as a starting point.

[castonrecovery]

got it working with AMD Radeon RX 9070 RDNA 4. Used hipify-perl to port and had to use x-thread to stop wilds corrupting.

[ostap1706]

That's the main reason why people fail: they don't like what they do and just want money.

Hello. Would you be willing to help search for a private key? I partially lost my data and some characters of the private key are missing. The range where my private key is located is 112. In case of a successful recovery, there will be a good reward.

If you are willing to take on the recovery, I’m ready to discuss all the details in private messages, including the payment.

Thank you for your attention. I will be waiting for your reply.

[RetiredCoder]

Hello. Would you be willing to help search for a private key?

No.
Any offers, opportunities, collaborations, projects, etc - no.

[ostap1706]

Hello. Would you be willing to help search for a private key?

No.
Any offers, opportunities, collaborations, projects, etc - no.

Thank you for your reply.

[Ykra]

I've been on quite the adventure into the hellscape lovely world of SASS.
I've made many minor performance improvements (still early days but they are adding up) focusing on rescheduling field multiplication instructions to hide latency (getting these TRUE latency values was a feat in itself).

For those also seeking the challenge there are some good tools out there to assist with assembling / dissasembling. (Half convinced some of the tools I've looked into might be made by you RC XD)
It's not for the faint of heart and RC was right it could easily drive you insane, rescheduling enough IMAD's you start believing them.

[Cricktor]

...

Hello, I see you mentioned that part of your private key is missing. I believe I can assist you. Please contact me via Telegram: @padamp1

@ostap1706
I would be very careful whom you might trust with your recovery problem. Especially Newbie accounts who invite you to wander off to Telegram, scammer's paradise. I'm not saying that cwrising has bad intentions, I'm only saying, be careful whom you trust. cwrising's post history is mostly about trading some wallets and we commonly know how "real" traded wallets are.

You have no reason to trust me or my judgement. Here's my recommendation:
Contact Dave from the service he announced years ago in this thread https://bitcointalk.org/index.php?topic=240779.0. This guy knows what he's doing, has very likely quite potent gear for cracking and recovery and won't scam you. Get his opinion on your recovery case and see if he can help you. AFAIR, Dave will only charge you a fair percentage if he succeeds, at least that's how I remember his business.
(Pay attention to have no typo in Dave's email address or his website. There are scammers waiting on typo domains!)

Apologies, if this is a bit off-topic, but I felt I should give ostap1706 a better option than some random dudes lurking him away from forum's public space to Telegram.

[kTimesG]

I've made many minor performance improvements (still early days but they are adding up) focusing on rescheduling field multiplication instructions to hide latency (getting these TRUE latency values was a feat in itself).

How many raw FE mul/second are you getting on a stock RTX 4090 @450 W? Like, simply multiplying a value by itself numerous times, and saving it to DRAM after many loops?

My best version gave out 85 billion mul/s (40 registers, 1024 threads / block), but I haven't messed around with altering SASS, just read-only to analyze what instructions get used. The key is to make good use of both FP32 and ALU units during the operations (without getting insane) but also the pseudo-code planning matters a lot, since it affects that balance. Anyway, I think improving the inversion would be the #1 gainer.

[Ykra]

How many raw FE mul/second are you getting on a stock RTX 4090 @450 W? Like, simply multiplying a value by itself numerous times, and saving it to DRAM after many loops?

My best version gave out 85 billion mul/s (40 registers, 1024 threads / block), but I haven't messed around with altering SASS, just read-only to analyze what instructions get used. The key is to make good use of both FP32 and ALU units during the operations (without getting insane) but also the pseudo-code planning matters a lot, since it affects that balance. Anyway, I think improving the inversion would be the #1 gainer.

I don't have a 4090 locally but rented one to test; headless / no DE / nothing utilising the GPU.
This specific 4090 seems to have a lower power ceiling (300W).
SMI outputs (during load):
|  0  NVIDIA GeForce RTX 4090       On  |
| 37%   60C    P2          299W /  300W |

Clock rates during tests:
Graphics: 2415 MHz, SM: 2415 MHz, Memory: 10251 MHz
Graphics: 2400 MHz, SM: 2400 MHz, Memory: 10251 MHz
Graphics: 2385 MHz, SM: 2385 MHz, Memory: 10251 MHz

Tried a few t/b variations, this looked best:
RawFE:: regs=40 threads=1024 block=256 throughput= 87487.5 Mmul/s

This is pre altering SASS.
I'll play around and update after some SASS alterations.