papers/zoo-quasar-benchmarks/sections/08-discussion.tex

\section{Discussion}


\subsection{Epoch Length Trade-off}

The epoch length (currently 100 blocks) represents a trade-off:

\begin{itemize}
    \item \textbf{Shorter epochs} (10 blocks): Faster reward feedback, higher overhead (3.3ms/block vs 0.33ms/block)
    \item \textbf{Longer epochs} (1000 blocks): Lower overhead (0.033ms/block), slower reward feedback
\end{itemize}

We chose 100 blocks ($\approx$630ms at 21 validators) as a compromise that provides sub-second reward feedback with $< 1$ms/block overhead.

\subsection{Validator Hardware Requirements}

The dual-layer design requires validators to have GPU hardware for both consensus participation and AI compute. This raises the hardware bar compared to CPU-only validators but ensures that all validators can verify AI attestations firsthand.

Minimum validator requirements for Zoo mainnet:
\begin{itemize}
    \item CPU: 32 cores, 3.0 GHz
    \item RAM: 128 GB
    \item GPU: NVIDIA A100 40GB (or equivalent)
    \item Storage: 2 TB NVMe SSD
    \item Network: 1 Gbps
\end{itemize}