"Even perfection has room for improvement." - Ty Warner
The Status Quo
The Large Bitcoin Collider has been up and running for over a year and is churning away several hundred million addresses per second. It had a peak at 2.5 Gkeys/s and overall server parameters indicate that the architecture as-is can sustain a keyrate of up to 50 Tkeys/s without any significant changes to server software or hardware. It is the largest public BTC privkey collision attempt known and during its operational time it has found several interesting privkeys already.
While LBC performance is good (as of now there is no other known faster BTC address generator and checker), some other programs - namely hashcat - indicate that more performance could be achievable.
As of now, the LBC is Linux-only, which probably restricts the user base more than it should. There are various crutches and workarounds to make it run on Windows, but still a native Windows version with GPU support would be really nice to have.
The Improvement Aim(s)
- Get faster and more scalable GPU generators.
- Make changes to the code so a native Windows version can be built.
While LBC generators have excellent parallelization on GPU, they use little to no vectorization capabilities even if they are present in the GPU. Also, the on-GPU bloom filter check runs in GPU global memory which probably imposes a performance penalty. Because all ECC computations are done on CPU, the final keyrate is also depending on CPU performance which makes scaling tricky on multi-GPU systems.
If we hope to improve LBC generators significantly, we have to assign the task to professionals. After quite some research, the company StreamHPC seems to be the go-to address when it comes to OpenCL KnowHow and advanced GPU compute development. It's understandable that competence doesn't come for free. Their - quote - "development/improvement rate is €760 per day
As a preliminary plan, we would like to set up a 1-week hackathon at the StreamHPC facility with the LBC authors present and improving - hopefully significantly - on the generator performance as described above.
Given all costs and expenses, we assume a minimum total of 7500€ for a 5-day hackathon, which should address all requirements above. Currently, this would be 0.8 BTC. We do, however, plea for a crowdfunding goal of 1 BTC. This will allow us to compensate for any volatility in exchange rate (StreamHPC is paid in EUR) and should we have excessive funds from this initiative, we will define stretch goals so that all money raised will go into LBC development.