Anyone have luck installing Kangaroo on vast.ai platform? The Linux CUDA installation seems somewhat complex is there a streamlined process of exactly what to install?
I have installed Bitcrack for someone on Vast.ai so that's close enough.
Use the default image that Vast gives you for CUDA. They bundle CUDA Toolkit 10.0 but without nvcc. You need to install cuda-nvcc-10-0 and cuda-cudart-dev-10-0 and build-essential. Don't try to install another toolkit version because the installer doesn't play well with Vast servers and you'll have to destroy and create another server again.
Don't lease any RTX 30 series cards because those need CUDA toolkit 11.0 and as I already mentioned it's a PITA to install.
@WanderingPhilospher I am using one file: "save.work" to search for privkeys for different public keys in one specific range.
Does this decrease the speed with one file: "save.work"? Or do me need to change the file: "save.work" that is, register a separate file for each public key?
My gut feeling says that with the amount of disk I/O involved in writing files, it's faster to use one work file for saving. This only makes a difference if Kangaroo happens to write so many work files at once though, I believe it only writes one so it doesn't make a difference whether you use the same name or a different name.