 |
Cambridge, UKApril 29, 2026 /PRNewswire/ — Myrtle.ai, a recognized leader in accelerating machine learning inference, today announced that the technology stack built around its VOLLO® product has been audited by STAC®, an authoritative benchmark testing organization for the financial industry.[1] Results unveiled at today’s STAC Summit in London clearly demonstrate that FPGA-based solutions offer significant low-latency advantages in machine learning inference for financial trading and related applications.
Myrtle.ai Halves Latency in Financial Machine Learning Inference Benchmark Record with VOLLO
STAC-ML (Markets) Inference is a technical benchmark standard used to evaluate solutions capable of performing inference operations on real-time market data. Designed collaboratively by quantitative researchers and technology experts from leading global financial institutions, it assesses and reports on the performance, resource efficiency, and overall quality of any technology stack performing inference calculations using specified models.
VOLLO achieves latency as low as 2 microseconds (99th percentile), while also excelling in throughput and efficiency. Across all three benchmark models, VOLLO’s inference latency (99th percentile) is lower than all previously audited systems, cutting the historical best record in half. This low and deterministic latency enables users to leverage more complex models at faster speeds, leading to smarter decisions and a competitive edge in trading, risk analysis, quoting, and various other trading-related activities.
With hundreds of thousands of hours of production-grade trading experience, VOLLO is already generating consistent excess returns for multiple leading trading institutions worldwide. These institutions typically develop and train various models using standard machine learning toolchains, then compile them into VOLLO and deploy them on their chosen FPGA hardware platforms.
In the tested system, VOLLO runs on a Silicom standard form factor FBAP4@VP18-2L0S PCIe accelerator card, equipped with an AMD Versal™ Premium series VP1802 adaptive SoC, installed in a Supermicro AS-2015CS-TNR server. The AMD Versal Premium series adaptive SoC offers PCIe Gen5x8 interfaces and over 3.3 million programmable LUTs, making it highly suitable for low-latency inference applications.
Peter Baldwin, CEO of Myrtle.ai, said: “Since VOLLO first fully leveraged the potential of FPGAs in STAC benchmarks in 2023, we have worked closely with customers to further reduce latency, expand the types and sizes of models VOLLO can run, and broaden the range of platforms it supports. We are delighted to collaborate with AMD, Silicom, and Supermicro on this benchmark, demonstrating how our combined technologies enable ultra-low-latency AI inference in quantitative trading.“
Girish Malipeddi, Director of Data Center FPGA Business at AMD, said: “The future of financial markets will be shaped by AI systems capable of interpreting data and taking action in near real-time. Based on the AMD Versal™ Premium series adaptive SoC, myrtle.ai’s VOLLO showcases how advanced low-latency inference can help unlock a new generation of intelligent trading infrastructure.“
Michael McNerney, Senior Vice President of Marketing and Cybersecurity at Supermicro, said: “We consistently cover a wide range of markets with systems powered by AMD, including the system used in this STAC-ML benchmark. Our servers address the most challenging workloads in the financial services industry, and by partnering with collaborators, we can deliver high-end performance for machine learning workloads with extremely low latency.“
Anders Poulsen, Vice President of Danish Solutions at Silicom, said: “We are pleased that myrtle.ai chose the Silicom Artena accelerator card based on AMD Versal Premium for these tests. Built around one of the largest FPGAs in a PCIe form factor, Artena is an ideal platform for VOLLO. Together, VOLLO and our low-latency hardware deliver deterministic microsecond-level inference for demanding trading workloads.“
The full benchmark results are available in the STAC report (SUT ID MRTL260323) at http://www.STACresearch.com/MRTL260323.