GatlingX, a project led by Oxford alumni specializing in machine learning and reinforcement learning, has unveiled ‘GPU-EVM’, which is claimed to be the best performing Ethereum Virtual Machine (EVM) according to internal benchmark scores.
GPU-EVM is an EVM scaling solution that the development team says is capable of training state-of-the-art reinforcement learning (RL)-based AI agents. Train AI agents to find security bugs by leveraging parallel execution across a variety of Ethereum apps.
GPU-EVM leverages graphics processing units (GPUs) to scale transaction throughput by executing tasks in parallel. The team claims that GPU-EVM can process tasks nearly 100 times faster than current high-performance EVMs, including evmone and revm. This is primarily due to the GPU’s ability to process multiple tasks simultaneously, leveraging an architecture inherently suited to parallel processing.
“Modern GPUs, designed with thousands of cores, can handle multiple tasks simultaneously, making them ideal for parallel processing tasks. These unique architectural advantages allow GPU-EVM to execute vast numbers of EVM instructions in parallel, dramatically accelerating computation speed and efficiency,” the GatlingX team said.
This comes amid rapidly growing interest in parallel EVMs due to their potential to solve blockchain scalability problems. Existing EVM implementations process transactions sequentially as they arrive, which increases processing time and cost when transaction volume is high. However, a parallelized EVM has the ability to process multiple transactions simultaneously as long as they are independent of each other.
AI agent training for smart contract security
GPU-EVM is designed to support the training of AI agents within a parallel simulation environment, according to co-founder Eito Miyamura, who spoke to The Block. These agents are trained to detect and exploit vulnerabilities in smart contracts, which the team compares to the strategic gameplay required to defeat a Go world champion, reminiscent of an AlphaGo scenario from the mid-2010s.
The technology, which is comparable to Nvidia’s Isaac Gym and Google’s Brax, enables parallel simulations on GPUs for accelerated reinforcement learning training and has a wide range of applications, Miyamura added.
The initial phase of the GPU-EVM rollout focuses on creating a hardware-scale EVM infrastructure to facilitate training of AI and RL models. These models interact with a variety of elements, including accelerated layer 2 solutions, maximum extractable value (MEV) operations, and backtest scenarios.
Subsequent steps, expected within a year, will include providing API access to high-performance computing applications, with the ultimate goal of surpassing human capabilities in securing smart contracts and distributed applications.
Disclaimer: The Block is an independent media outlet delivering news, research and data. As of November 2023, Foresight Ventures is a majority investor in The Block. Foresight Ventures invests in other companies in the cryptocurrency space. Cryptocurrency exchange Bitget is an anchor LP of Foresight Ventures. The Block continues to operate independently to provide objective, impactful and timely information about the cryptocurrency industry. Below are our current financial disclosures.
© 2023 The Block. All rights reserved. This article is provided for informational purposes only. It is not provided or intended to be used as legal, tax, investment, financial or other advice.