With the introduction of OpenMM, miners are now required to use GPU hardware for energy minimization simulations. Protein folding is a computationally intensive task. There are no shortcuts, and as such miners are rewarded truly on their ability to find the best configuration for their protein.
With the removal of CPU-based miners, participants are encouraged to explore the trade-off between parallell processing and dedicating all compute power to a single simulation. Limiting the number of active simulations allows miners to focus on finding the best configuration for a single protein without the overhead of managing multiple tasks.
The miner is responsible for running a config.neuron.max_workers number of pdb_id jobs, where these jobs are launched via the miner-specific logic. The flow is below:
The validator rewards miners by checking what each miner's intermediate results are for a specific pdb job. The faster you can find the best solution, the more likely you are to recieve rewards on each query step.