Files
zeekay abf99d05fc papers: reorganize into per-paper dirs, add Antje Worring as co-author
Restructured from flat .tex/.pdf files into per-paper subdirectories.
Added Antje Worring as co-author on all papers.
Synced PDFs to pdfs/ for download site.
2026-04-18 21:50:07 -07:00

38 lines
1.7 KiB
TeX

\section{Introduction}
The Ethereum Virtual Machine (EVM) is the dominant smart contract execution environment,
powering over \$500B in total value locked across Ethereum and its L2/L3 ecosystem.
Despite its ubiquity, the EVM's sequential execution model limits single-chain throughput
to approximately 30--50\,Mgas/s on modern hardware~\cite{pevm,reth}.
This paper reports measured performance of four acceleration strategies applied to Lux
Network's EVM implementations:
\begin{enumerate}
\item \textbf{GPU opcode execution} via Metal compute shaders on Apple M1 Max, achieving
85.9\,M opcodes/sec (4.85$\times$ over CPU).
\item \textbf{Native C++ EVM} (evmone++ fork) with full state, achieving 20.9\,Ggas/s
(5.1$\times$ over Go EVM).
\item \textbf{Parallel ecrecover} on 10 CPU cores, achieving 7.1$\times$ speedup on the
operation consuming 87\% of block processing time.
\item \textbf{Block-STM parallel execution} for conflict-free transaction batches.
\end{enumerate}
All numbers in this paper are measured on real hardware with real workloads. Projected
values are labeled explicitly.
\subsection{Contributions}
\begin{enumerate}
\item First published GPU EVM kernel benchmarks on Apple Metal, with crossover analysis
showing the exact transaction complexity threshold where GPU wins.
\item Measured C++ vs Go EVM comparison at 5.1$\times$ with full state layer, not just
interpreter-only micro-benchmarks.
\item Profiling-driven analysis proving ecrecover is 87\% of block time, redirecting
optimization from interpretation to cryptography.
\item Complete GPU memory scaling model from 10k to 100k transactions.
\item 1{,}183-test validation suite spanning C++, Rust, Go, Metal, and WGSL.
\end{enumerate}