Topic: BitCrack - A tool for brute-forcing private keys - page 53. (Read 76850 times)

Do you try this version already or not yet?

https://github.com/yoyodapro/BitCrack/releases/tag/v11.2-beta

Been following this thread for some time now.

Was running cuBitCrack on Windows with a Geforce GTX 1650 card. Using Nvidia 452.06 and Cuda 10.2.89. It did work and got around 49 Mkeys/s with cuBitCrack-0.32a 64/128/3200 settings and 634000 targets (compressed and uncompressed).
Don't recall how much on a single compressed target and don't want to take the computer apart just for that.

Bought yesterday an Asus RTX 3060 12 Gb. Off course the Nvidia 452.06 didn't work. Tried 457.30 as well with nil luck. Next step was 461.72 together with Cuda 11.2.2 and can confirm it does run with clBitCrack. However cuBitCrack doesn't work with current setup. At least not for me.

So at the moment I get 150 Mkeys/s on a Windows 10 Pro 64 (20H2), Nvidia 461.72 and Cuda 11.2.2 using clBitCrack (Yoydapros version v11.2-beta) settings 224/512/800 with 634000 targets (compressed and uncompressed). Its calculated to 91,750,400 starting points. If I try to push the slightest on any of the -b/-t/-p settings it gets an error.
Single compressed target gets 730 Mkeys/s with same settings.

I guess I should be happy getting three times the higher speed with the 3060 and running clBitCrack. Would like to try a working cuBitCrack to see if any difference. Does anybody know if there is a Cuda version BitCrack that works with RTX 3060 or higher?

ok, here without use both -b and -t

cuBitCrack.exe -d 0 --keyspace 8000000000000000:+1000000000000000 16jY7qLJnxb7CHZyqBP8qca9d51gAjyXQN

GeForce GTX 1050 1436 / 4096MB | 1 target 81.03 MKey/s (1,942,487,040 total) [00:00:22]

I have been a gpu miner dev since 2013 (with 1000 or more opensource commits to github in the ccminer project). All the optimizations in this mod are real, and I didn't change the algorithm so it's pure teqnique. With more work I can make it faster, but there is no money to be earned so I will probobly return to miner development. Check out my vertcoin miner with 1% fee. Vertcoin is halving in 2021, the price is low. Listed on bittrex.
Profitable to mine on old hardware with the sp-mod...

Try to run without the -b and -t parameter, the defaults has been improved in #3.

I believe the opencl bug has been solved

https://github.com/BoGnY/BitCrack/commit/786ea3615a9b4dabae8f399e5db5daca0521e3fd

Build his github and you will be fine.

What program? Is is a bitcrack solver?

Actually pretty easy. Set up a 32 bit (or 36 bit, or whatever your GPU can get through within 5 to 10 minutes) range/keyspace with 3 keys. One up front, one in middle, one at end.

Run sp's and then run original, with same exact settings.  Make sure to pause (pause at end of batch script for Windows) each program once it is done. Compare the times and see if both programs found all 3 keys.

How would that work though? By posting a bunch of speeds reported from different programs, you're still relying on taking their speed at face value.

To see whether sp_ really did speed up the program and not simply multiplying the counter by 4 or 5, we must test it with a small range, such as 0 to 5 billion, which we can expect to finish in about 5 seconds on the regular Bitcrack on some GPU that does 1000MKeys/s.

If the mods are genuine then it will finish in about a second (again, I can't do this myself because no source code => no way to compile on Linux aaaaargh  )

cuBitCrack.exe -d 0 -b 20 -t 256 --keyspace 8000000000000000:+1000000000000000 16jY7qLJnxb7CHZyqBP8qca9d51gAjyXQN

laptop
GeForce GTX 1050 1436 / 4096MB | 1 target 81.70 MKey/s (3,342,336,000 total) [00:00:38]
GeForce GTX 1050 1436 / 4096MB | 1 target 83.18 MKey/s (5,827,461,120 total) [00:01:08]

Care to share the source code? I’m really wondering what’ve been missing all this time.

Not just you but for others as well; it is one thing to talk about the speed shown. How many have actually compared the two actually going through say a 40 bit range and finding a known key?

The shown speed means nothing, anyone can easily tweak code to show and increase, but how fast does it get through a range?

For those of you testing sp's new release(s), show comparison test, not just speed.

1100 MKey/s for a V100 seems slow. I get more than that with a 2080 (non Ti) with a different program. 

Thanks for releasing this. On Tesla V100 it shows around 1100 Mkey with your sp-mod #3. Huge difference vs. the initial release.  

Hi, you made some great changes on the cuda version, I get about 220 Mk / s on gtx 970.

would you be able to improve and repair the opencl version too?

Good deal...the more important issue is to tackle multi gpu use. It makes it so much easier versus multiple instances.

I have ordered new cards.  Will get them next month..

Another +10%

https://github.com/sp-hash/Bitcrack/releases/tag/3

gtx 1070 375Mkeys
gtx 1070ti 445Mkeys
gtx 1080 480Mkeys
gtx 1080ti 650Mkeys
gtx titanx 610Mkeys

It's better to use a network library like this one (https://github.com/Qihoo360/evpp, just a random link from the first page of Google search results) that lets you send packets asynchronously for the size of data you're trying to send, else the blocking nature of network I/O will make the whole program no faster than writing to a disk. And even that can be written to asynchronously using stuff like libevent but the results aren't much better: Instead of stalling the GPU you're now hammering the disk making it unavailable for all other programs.

So it's like you got to optimize for your network speed and only try to send that many packets per second. Since these are local networks we're talking about the speed should be predictable. Cloud platforms like Hetzner also let you network with their other servers since 99% of people reading this don't have 10GBit ports and spare boxes lying around somewhere.

How would you do this? Any code or snippets to look at? Or is it something you write on client side to send all output to another PC on local network?

If you're referring to Bitcrack itself and not a pool, it's pretty straight forward: make BUILD_CUDA=1

Memory usage during compilation is actually very light, should work on an entry-level VPS. I suspect this is not what you're asking though. I'm quite confused.


Now I know what an endomorphism is. By sheer coincidence, I was thinking about implementing something like:


So I would've checked 15 other keys just by flipping each of the 4 64-bit words that make up the Int using NOT, but in the end I scrapped it because I didn't think any of those additional points have any special relationship with the target public key.


Alternatively if you have a 10GBit Ethernet port lying around somewhere you can just send all the points across your local network to some other system that just writes the points in its own storage.

The idea scales too: While the other system is waiting on packets it can send some more points to other systems so that it doesn't have to write all the points on its own disk and fill it up quickly. There are some kernel options that tune the network drivers for uploading.