Hey everyone,
I wanna share a new tool called VanitySearch. It's kinda like Vanitygen, but way faster on my end. This one doesn’t use OpenSSL for heavy lifting and runs on Cuda, fully taking advantage of PTX assembly.
On my Intel i7, it’s about 4x quicker than Vanitygen. Crazy speeds!
Sounds cool. Are there plans for a Linux version?
I’d like to give it a shot since my laptop is old. Might be interesting to see how it stacks up against Vanitygen.
Totally!
The CPU part should work fine on Linux, just need to check for any missing stuff with gcc.
For CUDA, I've got to install the SDK on Linux and try compiling it. Shouldn't be too tough.
Awesome news!
Just dropped a new release, it includes a makefile for Linux, but it's only for CPU right now.
Still working on the CUDA version for Linux. Would love to hear how it performs on your setups.
Thanks for the update!
So, those two 'Check' fields are just for debugging, right? And if you use the -gpu flag, it slows everything down if all CPU cores are engaged?
Exactly! Those fields help in checking addresses.
Yeah, if you add the -gpu option, it hogs resources since CPU can't manage GPU/CPU transfers well. I’ll make sure to leave one core free when GPU is selected.
Appreciate your interest!
Just released v1.2 which fixes a few issues and boosts performance by around 10%.
Updated the probability calculation for large prefixes too.
Hey all,
Just put out v1.3 with a ~15% overall performance boost and ~20% for GPU.
On a side note, it’s now twice as fast on my machine compared to oclvanitygen. Just need to hit that 8-character prefix goal!
I’m using affine coordinates for speed.
We perform p = startP + i*G, and the inversion only happens once per group. Makes it quicker.
Plus I’ve got a fast modular inversion method.
A bigger group size doesn’t yield significant gains, maybe less than 1%.
My modular inversion method is quick, and with 256, it keeps things manageable.