Myrtle.ai Sets New Benchmark in Financial Machine Learning Inference Latency

Myrtle.ai Revolutionizes Financial Machine Learning Inference

Myrtle.ai, a recognized leader in accelerating machine learning inference, has recently made significant strides in reducing latency for financial machine learning applications. At the STAC Summit held in London, the company announced the results of a recent audit performed on its VOLLO® product, showcasing a remarkable reduction in inference latency.

The benchmark tests, conducted by STAC®, a reputable authority in the financial industry, demonstrated that VOLLO achieved a latency time of just 2 microseconds at the 99th percentile. This score not only highlights the efficiency of VOLLO but also represents a 50% reduction when compared to past records. Such latency is critical in high-frequency trading and real-time market analysis, allowing firms to leverage more complex modeling techniques more swiftly than ever before.

The Importance of Low Latency in Financial Markets

Latency refers to the delay before a transfer of data begins following an instruction. In the fast-paced environment of financial trading, even milliseconds can make a substantial difference. VOLLO’s deterministic latencies enable traders to make more informed decisions quickly, which is a vital competitive advantage in areas such as trading, risk assessment, and quoting.

With extensive operational usage, VOLLO has demonstrated its capability to create alpha generation for numerous leading trading firms around the globe. These institutions have utilized a broad range of models in standard machine learning workflows, which are then compiled into VOLLO and executed on their chosen FPGA hardware platform. The testing scenario employed the FBAP4@VP18-2L0S PCIe accelerator card from Silicom, featuring an adaptive AMD Versal™ Premium Series VP1802 SoC, installed on a Supermicro AS-2015CS-TNR server. The combination of these advanced technologies is optimized for low-latency inference applications.

Collaboration with Top Technology Providers

Peter Baldwin, CEO of Myrtle.ai, expressed enthusiasm about the reducing latencies associated with VOLLO, stating, “Since VOLLO first realized the full potential of FPGAs in these STAC tests in 2023, we have collaborated with our clients to further reduce latencies and expand the variety and size of models that VOLLO can execute.” This commitment reflects Myrtle.ai's dedication to enhancing the inferencing capabilities available for AI systems in quantitative trading.

In partnership with industry leaders such as AMD, Silicom, and Supermicro, Myrtle.ai is crafting a state-of-the-art infrastructure for intelligent trading that will redefine future financial market dynamics. Girish Malipeddi, Director of Data Center FPGA Business at AMD, noted, “The future of financial markets will be shaped by AI systems capable of interpreting data and acting in near-real-time.” His emphasis on the transformative capabilities of the AMD Versal™ Premium Series underscores the robust foundation on which VOLLO operates.

Accessing VOLLO’s Capabilities

Those involved in machine learning development can now test how their models perform on VOLLO without needing tools or FPGA expertise. For further insights or engagement, interested parties are encouraged to visit vollo.myrtle.ai or directly contact the company at [email protected]

The complete results of these benchmark tests are detailed in the STAC report (SUT ID MRTL260323), accessible through STAC Research.

About Myrtle.ai

Myrtle.ai is an AI/machine learning software company that delivers world-class inference accelerators on FPGA platforms across prominent FPGA suppliers. With deep expertise in neural networks, Myrtle.ai provides accelerators for applications, including fintech, wireless telecommunications, large language models, speech processing, and recommendation engines. The innovations generated by Myrtle.ai are poised to set new benchmarks in technological advancement across various sectors, pushing the envelope on what's possible in machine learning inference performance.