mirror of
https://github.com/lux-apps/papers.git
synced 2026-10-30 18:15:21 +00:00
zeekay
abf99d05fc
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.
59 lines
2.5 KiB
TeX
59 lines
2.5 KiB
TeX
\section{Updated Benchmark Results}
|
|
|
|
\label{sec:updated}
|
|
%% ============================================================================
|
|
|
|
Recent benchmarks with larger block sizes (47K transactions = 1\,B gas) and pipeline
|
|
optimizations yield significantly improved numbers:
|
|
|
|
\begin{table}[h]
|
|
\centering
|
|
\small
|
|
\begin{tabular}{lrr}
|
|
\toprule
|
|
\textbf{Benchmark} & \textbf{Result} & \textbf{Notes} \\
|
|
\midrule
|
|
CPU evmone sequential & \SI{534}{Ggas/s} & 100K txs, \SI{3.9}{ms} total \\
|
|
GPU Metal kernel V1 & \SI{352}{Mops/s} & 50K txs, 2.2$\times$ on contract bytecode \\
|
|
GPU unified pipeline & \SI{41}{ms} total & Consensus + EVM + FHE, same Metal device \\
|
|
C++ state layer & \SI{22965}{Mgas/s} & 4.9$\times$ faster than Go \\
|
|
GPU batch ecrecover & eliminates \SI{1613}{ms} & 87\% of block time removed \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\caption{Updated cevm benchmarks on Apple M1 Max. The GPU unified pipeline demonstrates
|
|
that consensus, EVM execution, and FHE can share a single Metal device with \SI{41}{ms}
|
|
total latency. GPU batch ecrecover eliminates the \SI{1613}{ms} signature bottleneck that
|
|
dominated block processing at 47K transaction block sizes.}
|
|
\label{tab:updated}
|
|
\end{table}
|
|
|
|
The C++ state layer at \SI{22965}{Mgas/s} represents a 4.9$\times$ improvement over the
|
|
Go EVM, consistent with the earlier 5.1$\times$ ratio measured on smaller blocks. The
|
|
GPU Metal kernel V1 at \SI{352}{Mops/s} achieves a 2.2$\times$ speedup on contract
|
|
bytecode execution at 50K transactions.
|
|
|
|
\subsection{Consensus Pipeline Results}
|
|
|
|
\begin{table}[h]
|
|
\centering
|
|
\small
|
|
\begin{tabular}{lrl}
|
|
\toprule
|
|
\textbf{Metric} & \textbf{Result} & \textbf{Configuration} \\
|
|
\midrule
|
|
Block throughput & 107K blocks/sec & M1 Max, $K{=}1$ SoloGPU \\
|
|
Theoretical TPS ceiling & 100M TPS & BurstParams: 2.1B gas, 1ms blocks \\
|
|
GPU sig recovery speedup & 32$\times$ & \SI{1613}{ms} $\to$ $\sim$\SI{50}{ms} for 47K sigs \\
|
|
Build tag requirement & Zero & Auto-detect on darwin \\
|
|
\bottomrule
|
|
\end{tabular}
|
|
\caption{Consensus pipeline integration. 107K blocks/sec measured on Apple M1 Max
|
|
with $K{=}1$ SoloGPU configuration. BurstParams define the theoretical ceiling at
|
|
100M TPS (2.1B gas limit, 1ms block time). GPU signature recovery achieves
|
|
32$\times$ speedup, reducing 47K-signature batches from \SI{1613}{ms} to
|
|
$\sim$\SI{50}{ms}. Zero build tags required---Metal backend auto-detected on darwin.}
|
|
\label{tab:consensus}
|
|
\end{table}
|
|
|
|
%% ============================================================================
|