Gonka AI Computing Power Network Adjusts PoC Mechanism: Activate in 5 Seconds, GPU Efficiency Doubles, Small Miners Still Profit

【Chain Wen】Gonka, a decentralized AI computing power network, recently shared some technical adjustment details within the community, with several core changes.

In simple terms, there are three directions: first, PoC and inference now run on the same large model; second, the activation method has shifted from delayed switching to near real-time triggering; third, the calculation of computing power weights has been optimized to better reflect actual hardware costs.

Co-founder David explained why these changes are necessary. Basically, it’s not about catering to short-term miners, but because the network’s computing power scale is growing too rapidly, requiring adjustments to the consensus and verification structures to adapt. The key goals are twofold: to ensure stability and security under high load, and to prepare for running larger-scale AI tasks in the future.

Regarding the community’s complaint that “small models produce too many tokens,” the team responded: for the same number of tokens, small and large models consume vastly different GPU computing power. As the network upgrades to higher density and more complex tasks, they are gradually calibrating the relationship between computing power weights and actual computational costs to prevent long-term structural imbalances from affecting network scalability.

Under the new PoC mechanism, activation time has been reduced to within 5 seconds—this significantly decreases resource waste caused by model switching and waiting, allowing GPUs to allocate a higher proportion of resources to truly effective AI computations. Additionally, unified model operation reduces system overhead caused by nodes switching between consensus and inference, greatly improving overall computing power utilization efficiency.

For single-card and small-to-medium GPU miners, the team specifically mentioned: through mining pool cooperation, flexible participation by Epoch, and engaging in inference tasks, they can still continue to earn rewards and participate in network governance. So, it’s not that small miners have no way out.