Myrtle.ai Achieves Breakthroughs in Financial Machine Learning with VOLLO's Latency Reduction

Myrtle.ai Transforms Financial Machine Learning with VOLLO

Myrtle.ai, a prominent name in the realm of machine learning inference, has made significant strides in reducing latency through its cutting-edge VOLLO product. In a recent announcement, the company highlighted its success in achieving unprecedented latencies during STAC-ML benchmark tests, specifically designed for financial applications.

Benchmark Breakthrough

Conducted by STAC, a leading benchmarking authority in finance, the results showcased VOLLO's capabilities. The technology recorded latencies as low as 2 microseconds, solidifying its position in the market by halving the previous benchmark record. This impressive feat means that traders and firms utilizing VOLLO can process complex models and make quicker, more informed decisions during trading, risk assessments, and various other financial operations, drastically enhancing their competitive edge.

The STAC-ML (Markets) benchmark aims to evaluate technology solutions on their ability to run inference on real-time market data efficiently. Developed collaboratively by experts from major financial institutions, the benchmark assesses performance, efficiency, and resource utilization of different technology stacks designed to perform inference work.

Performance Insights

Myrtle.ai's VOLLO demonstrated outstanding performance across three different benchmark models, eclisping all previously tested systems. As a result of its low-latency capabilities, traders are empowered to harness more intricate models with greater efficiency than ever before. This translates into substantial gains in their operations as it allows the processing of larger datasets and refining strategies at remarkable speeds.

A Proven System

With extensive production trading experience, VOLLO is currently deployed by leading trading firms, effectively generating alpha through sophisticated model implementations. These firms have meticulously developed and trained various models using standard machine learning processes, which are then compiled into VOLLO for execution on FPGA-based hardware platforms.

During the testing phase, the VOLLO stack ran on the FBAP4@VP18-2L0S PCIe accelerator card from Silicom, featuring the AMD Versal™ Premium series VP1802 Adaptive SoC, within a Supermicro AS -2015CS-TNR server platform. The beta testing has proven advantageous, offering PCIe Gen5x8 performance and remarkable programmable LUT capabilities, tailored specifically for low-latency inference applications.

Looking Forward

Peter Baldwin, CEO of Myrtle.ai, expressed enthusiasm regarding the collaborative efforts with partners, including AMD, Silicom, and Supermicro, to realize these remarkable benchmarks. "Since our initial breakthrough with STAC benchmark in 2023, we've continuously worked to further reduce latencies and expand the diversity of models VOLLO can efficiently process. This allows us to support a wider range of platforms," he stated, indicating a clear path toward future advancements.

Girish Malipeddi from AMD remarked, "The future of financial markets will be heavily influenced by AI systems capable of processing data and executing actions in real-time. With Myrtle.ai's VOLLO backed by AMD technology, we can provide groundbreaking solutions for intelligent trading infrastructures."

Supermicro's Senior Vice President, Michael McNerney, emphasized the company's commitment to addressing diverse markets with efficient systems. He noted that their collaboration is essential for optimizing performance and minimizing latency in machine learning workloads within the financial sector.

Silicom's VP Solutions, Anders Poulsen, highlighted the agility of the Artena accelerator card, developed on AMD's cutting-edge technology, further validating the effectiveness of their joint efforts in achieving notable latency targets for demanding trading environments.

Start Utilizing VOLLO Today

For ML developers keen on experiencing these enhancements, Myrtle.ai invites you to evaluate how your models may perform on the VOLLO platform, requiring no FPGA tools or specialized knowledge. For additional information, you can visit vollo.myrtle.ai or reach out directly to myrtle.ai via email at [email protected].

For those interested in the comprehensive benchmark results, refer to the STAC Report (SUT ID MRTL260323) available at STAC Research.

About Myrtle.ai

Myrtle.ai specializes in AI/ML software solutions, providing advanced inference accelerators on FPGA systems compatible with top-tier FPGA suppliers. The company has garnered expertise in delivering acceleration solutions across various sectors, including financial technology and telecommunications, leveraging deep neural networks for diverse applications.

With their commitment to innovation, Myrtle.ai is paving the way for future advancements in financial technology, making them a key player to watch in the AI-driven landscape.