Bitcoin GPU mining: our measured performance gains and how Hybrid Solo rewards work

Disclosure: I operate BTC PoW Lab Pool through Power CM Software. This post presents our own pool and internal research results.

https://btcpowlab-pool.com/

Hi everyone, I wanted to share some concrete GPU results from our research and explain how our reward model works. The attached image summarizes both.

What did we measure?

On NVIDIA, our compatible client delivered 21.67% higher end to end throughput compared with the same client with the optimization disabled. We used the same GPU across four 120 second ABBA test blocks, including connection, preparation, job changes and share submission. All four blocks produced accepted shares, with no rejected shares or fallback activation.

On AMD/OpenCL, receiving prepared schedules produced a 69.11% increase in client throughput. Accounting for central preparation and latency, the separately documented conservative total improvement was 12.91% under warm operating conditions.

The approach moves some preparation work to a central component that supplies compatible clients. These measurements concern specific implementations and baselines. They do not establish the same improvement on every GPU, and connecting an ordinary miner does not automatically reproduce them.

How do rewards work?

Our Hybrid Solo model distributes the reward from a found block as follows:

• 85% goes to the block finder.
• 10% goes to other eligible community miners.
• 5% goes to the lab and operations. (also special rewards like referals)

Community eligibility uses accepted work over a rolling seven day window. Allocation considers contributed work, tier and pool composition. The finder is excluded from the community allocation for their own block.

This means an eligible miner can receive a community allocation when someone else in the pool finds a block. Rewards still depend on actual block finds, so there is no guaranteed payment schedule or income.

For the GPU miners here, what would make these results most useful to evaluate: longer benchmark runs, comparisons with specific mining clients, or power consumption measured at the wall?

Best and thanks! https://btcpowlab-pool.com/