Gonka reveals PoC mechanism and model evolution direction: aligning real computing power to ensure continuous participation of multi-level GPUs The decentralized AI computing power network Gonka recently explained the phased adjustments of PoC mechanism and model operation mode in the community AMA. The relevant adjustments mainly include: unified use of the same large model for running PoC and inference, adjustment of PoC activation mode from delayed switching to near real-time triggering, and optimization of computing power weight calculation method to make it closer to the actual computing cost of different models and hardware. Co founder David stated that the above adjustments are not aimed at short-term output or individual participants, but rather a necessary evolution of consensus and verification structures as network computing power rapidly expands. The aim is to enhance the stability and security of the network under high load conditions, laying the foundation for carrying larger scale AI workloads in the future. In response to the issue of high token output of small models mentioned in community discussions at the current stage, the team pointed out that there are significant differences in the corresponding real computing power consumption of models of different scales under the same number of tokens. As the network evolves towards higher computing power density and more complex tasks, Gonka is gradually guiding the alignment of computing power weights with actual computing costs to avoid long-term imbalance of computing power structure, which affects the overall scalability of the network. Under the latest PoC mechanism, the network has compressed the PoC activation time to within 5 seconds, reducing the computational waste caused by model switching and waiting, and enabling GPU resources to be used more efficiently for AI computing. At the same time, by running the unified model, the system overhead of node switching between consensus and reasoning is reduced, and the overall computing power utilization efficiency is improved. The team also emphasizes that single card and small to medium-sized GPUs can continue to generate revenue and participate in governance through mining pool collaboration, flexible participation by Epoch, and inference tasks.