🇬🇧 Introducing GPU-EVM: The Most performant EVM in the world

Eito Miyamura
8 min readApr 22, 2024

--

Setting the new state of the art on parallel-EVM execution

♨️ Announcing: GPU-EVM by GatlingX ♨️

The most performant parallel-EVM in the world by 100x. Faster. Cheaper. Easier.

GPU-EVM harnesses the power of Graphical Processing Units (GPUs) to run Ethereum Virtual Machine (EVM) operations in parallel. This means that instead of executing tasks one after the other, GPU-EVM can process many tasks at the same time, significantly speeding up computations. This breakthrough by our team of University of Oxford CS/AI Alumni drastically improves the unit economics of EVM computations/second/$.

Ethereum Virtual Machines (EVMs) are the industry standard virtual machines which run smart contracts, the foundation of modern blockchain technologies. EVMs are like an operating system of blockchains, facilitating transactions without a trusted third party, across many distributed computers on their CPU-based client software.

With GPU-EVM and the performance enhancements it provides, it gives massive unlock to ambitious engineering teams downstream: foundational infra for training AI/RL models that interact with the EVM, accelerated L2s, MEV, backtesting, and more. (details below)

GPU-EVM: New Paradigm of EVM Computation

NVIDIA started as a niche gaming-centric company but is now a pivotal player in computing and at the forefront of the AI revolution. This evolution reflects a move from Moore’s Law, which predicted the doubling of computing power every two years, to Huang’s Law, named after Nvidia’s CEO, Jensen Huang. Huang’s Law posits that GPU performance will more than double biennially, outpacing CPUs, thanks to the integration of hardware, software, and AI, making GPUs central to accelerating complex tasks.

As we hit the limits of Moore’s Law, the reliance on GPU parallelism heralds a new computing era, transitioning from CPU dominance to GPU-driven advancements (cf. end of Dennard scaling, Amdahl’s law). This shift is like moving from a single-lane road to a multi-lane highway, not only speeding up processes but also enabling more simultaneous activities, thus broadening what’s technologically possible.

Jevon’s Paradox illustrates this effect well: just as LED bulbs’ efficiency led to broader, not reduced, usage due to their cost-effectiveness, GPU-EVM’s enhanced efficiency and lower costs open up a plethora of new possibilities. Rather than merely saving resources, it catalyzes innovation and adoption in blockchain technology and beyond, promising a future where GPU computing’s efficiency drives exponential growth in computational applications.

GPU-EVM Performance

Leveraging the significant advances in the general computing capabilities of modern GPUs, we have engineered GPU-EVM to surpass the performance of traditional EVMs by an astounding 100x. Modern GPUs, designed with thousands of cores, are capable of handling multiple operations simultaneously, making them exceptionally suited for parallel processing tasks. This inherent architectural advantage allows GPU-EVM to execute a vast number of EVM instructions in parallel, dramatically accelerating computation speeds and efficiency.

To objectively measure the performance enhancements brought by GPU-EVM, we conducted comprehensive benchmarks using the open-source tool available at [EVM Bench](https://github.com/ziyadedher/evm-bench). This tool allowed us to simulate a wide range of EVM operations and compare the execution times between traditional CPU-based EVMs and our GPU-EVM.

In comparison to traditional computing paradigms, GPU-EVM blows all of these out of the water by leveraging the unparalleled processing power of GPUs, setting a new benchmark for EVM performance and efficiency.

With this technological foundation, let’s explore how GPU-EVM revolutionizes areas like AI training and DeFi simulations, opening new frontiers for blockchain applications.

Training AI Agents with the EVM

AI is changing the world, led by ChatGPT and other LLM chatbots, who were trained with Reinforcement Learning from Human Feedback, applying learnings from Reinforcement Learning (RL). At its core, RL embodies the process of training AI agents to make decisions by interacting with an environment where correct actions are rewarded. This method of learning is pivotal as it mirrors the fundamental way humans and animals learn from their surroundings, making it a cornerstone in the development of intelligent systems that can adapt and optimize their behavior autonomously.

The landmark victory of AlphaGo over world champions in Go is a testament to the transformative power of RL. It wasn’t just a game; it was a showcase of how AI, through RL, could discover strategies and solutions that had remained beyond human insight, by simulating and interacting with the complex environment of the Go board. This breakthrough highlighted the essence of RL: enabling AI agents to autonomously navigate and learn from their environment to achieve specific goals, guided by the reward system.

However, the journey to achieving such breakthroughs in AI through RL is fraught with computational challenges. Simulating environments for AI to interact with requires significant computational resources. The advent of GPU-parallelized simulation environments like NVIDIA’s Isaac Gym, Google’s Brax, and JAX-LOB has been instrumental in overcoming these hurdles. By leveraging GPU-parallelized simulation environments, these platforms have achieved performance improvements ranging from 100 to 250,000 times, making the computational aspect of RL more feasible and efficient. GPU-Parallelization achieves these speed improvements, due to the bottleneck in AI training often being the CPU-GPU communication bandwidth of passing data between each other, and have become an industry staple in the RL research community.

In the rapidly evolving world of AI, GPU-EVM serves as a GPU-parallelized simulation environment, facilitating the training of AI agents directly within the blockchain ecosystem. A compelling application of this is in the finance industry, where GPU-EVM can revolutionize real-time fraud detection systems. History shows the importance of these systems, where Max Levchin developed the first fraud prevention mechanism in PayPal to prevent the company from going bankrupt. By enabling a financial AI to simulate and analyze millions of transactions in mere seconds, it can identify anomalous patterns indicative of fraudulent activity with unprecedented speed and accuracy. This capability, which would have previously taken days to achieve, represents a monumental shift in how financial institutions can safeguard against fraud. By integrating AI agents with the EVM through GPU-EVM, it opens new avenues for applying Reinforcement Learning (RL) principles within the blockchain domain. Here, AI agents learn and improve by being rewarded for identifying fraudulent transactions accurately, based on predefined reward functions.

L2 Acceleration / Simulation

The advent of Layer 2 solutions, or L2s, has been instrumental in scaling Ethereum’s throughput, thereby facilitating its adoption for mainstream applications, notably payments. By processing transactions off the main Ethereum blockchain (Layer 1), L2s significantly enhance the network’s capacity while maintaining its foundational principles of security and decentralization. Distinct from conventional CPU-based systems, GPU-EVM operates independently, enabling the seamless integration and acceleration of existing L2 solutions. This acceleration can be achieved through a variety of methods, including the optimization of view functions and the application of Monte Carlo tree search-like algorithms for more efficient block building and transaction ordering.

The role of parallel-EVMs in the context of L2 acceleration, however, is complex and requires a thoughtful approach. Direct acceleration of L2s through parallel-EVMs is not as straightforward as it might seem. To truly leverage the capabilities of parallel-EVMs, there must be a concerted effort to innovate both the design of L2 solutions and their databases. This point is underscored by works like:

https://x.com/Eito_Miyamura/status/1769778763053826380?s=20

While the nuances of integrating GPU-EVM with L2 solutions are extremely promising, it is important to note that other challenges need to be addressed. The primary bottlenecks in this endeavor include addressing the limitations related to storage, managing long chains of interdependent transactions, and mitigating the state bloat cost. GPU-EVM alone cannot resolve all of these issues. Therefore, a collaborative effort to innovate in the design of L2 solutions and the databases that underpin them is essential for overcoming these obstacles and fully realizing the benefits of GPU-EVM in the context of L2 acceleration.

DeFi Simulation / Fuzzing

The base 100x increase in throughput of GPU-EVM provides a transformational change in DeFi simulation and fuzzing. This significant enhancement in data processing capabilities enables the discovery of previously unaccounted edge cases in DeFi strategies and protocol designs, uncovering novel vulnerabilities that could have remained hidden. To illustrate the magnitude of this advancement, one could compare the traditional CPU-based methods to a water gun, whereas the GPU-EVM operates more like a powerful water hose, offering a much more effective means of extinguishing bugs.

Thanks to the GPU-EVM’s base performance boost, fuzzers operating on this platform can delve deep and operate at remarkable speeds, identifying edge cases in a matter of seconds. This is a stark contrast to CPU-based fuzzers, which could take weeks or even months to uncover the same issues. The ability to run these advanced fuzzers on top of GPU-EVM allows for continuous monitoring of smart contracts, especially those in live production. These automated systems are designed to relentlessly challenge the smart contract, attempting to foresee potential vulnerabilities several moves ahead, much like a strategic game of chess, with the ultimate goal of ensuring the highest level of security.

We are on the brink of introducing a product that embodies this cutting-edge approach to DeFi simulation and fuzzing. Stay tuned for an innovation that will redefine the standards of smart contract security and resilience.

About GatlingX

GatlingX is an applied infrastructure and AI Lab, with a focus on developing heavy technical infrastructure. Our mission involves creating products that empower a wide range of blockchain applications, operating at the deep infrastructure layer.

We believe that there are some classes of extremely difficult technical problems the blockchain industry is unwilling to address, due to how difficult they are. Fast and cheap security, computational performance, and speed are necessary prerequisites for a thriving blockchain ecosystem, but equally extremely difficult problems, with a lot of suffering. We believe that no one will solve these, unless we get together the best problem solvers in the world to solve it.

Our commitment lies in advancing the state of the art (SoTA) in fields such as AI, GPUs, Blockchains, and Distributed Computing, all of which are pivotal in driving global technological progress.

We are hardcore: If we can get something off-the-shelf, we do it. If it doesn’t, we just build it.

Use GPU-EVM

GPU-EVM is currently in private early access as we ramp up GPU capacity. If you’re interested in using GPU-EVM for your engineering work, please fill out this form here to get on the waitlist: https://noteforms.com/forms/signups-hcgrrj

Join Us

Our team is small and extremely talented. Our founding team consists of University of Oxford Alumni who have worked at the cutting edge of infra, applied AI at companies like Crowdstrike, Wayve, Citadel Securities, as well as created impactful projects like ZKMicrophone and Graphite.

Building GPU-EVM is just the first step — our hardest challenges still lie ahead. If you’re excited to solve some of the world’s toughest problems on the cutting edge of AI, GPU-simulation, RL, security, parallel-computing and blockchains, learn more about our team and apply to join us here.

https://gatlingx.notion.site/GatlingX-Job-Board-14151db0dfd04bbc8b142c01956a6006?pvs=4

--

--