#USA #Cambridge #Myrtle.ai #VOLLO #STAC-ML

Myrtle.ai's VOLLO®: Revolutionizing AI Inference Performance

In a recent announcement, Myrtle.ai has made waves in the world of advanced analytics and financial trading by revealing its VOLLO® inference accelerator's extraordinary performance during the latest STAC-ML benchmarks. The company showcased its cutting-edge technology at the STAC Summit in London, where VOLLO® set new records in the category of gradient-boosted tree inference. With a p99 latency of under 2 microseconds and an astonishing throughput of up to 50 million inferences per second, these achievements highlight the potential for innovation in the fast-paced trading environment.

What Sets VOLLO® Apart?

Executed on the AMD Alveo™ V80LL accelerator housed in a Blackcore ICON 3132-SM+ server, VOLLO® outperformed its predecessors by significantly reducing latency by over 30% and boosting throughput by five times compared to previous best results. These impressive metrics were achieved while handling multiple model types concurrently, making VOLLO® a versatile tool for any trading firm.

In the world of electronic trading, speed is tantamount to success. The time between market data reception and decision-making can have a direct impact on returns. With VOLLO's low and deterministic latency, traders are now empowered to utilize larger and more sophisticated models that can exploit market opportunities without delay. This means improved decision-making capabilities without sacrificing model quality for speed.

Peter Baldwin, the CEO of Myrtle.ai, has expressed enthusiasm about VOLLO®'s capabilities, stating that trading firms are eager to leverage more powerful models without compromising on speed. This is now a reality with VOLLO®, which allows developers to test their own models seamlessly — requiring no special knowledge of FPGAs. Baldwin's insights emphasize the importance of flexibility in developing trading strategies and how technology can be the key to gaining an edge in the market.

Proven Success in the Real World

Following the impressive results of the STAC Tacana announced earlier this year, VOLLO® holds records for both decision trees and neural networks under deterministic latency conditions. The performance of VOLLO® has already been validated in real-world trading scenarios, tallying hundreds of thousands of hours of production trading that generated alpha for numerous leading trading firms worldwide.

Moreover, the architecture of VOLLO®'s system positions it as a preferred platform not only in financial trading but also in telecommunications, network security, and defense sectors. Its adaptability is critical for companies looking to enhance their computational capabilities across various applications.

Understanding STAC-ML and Its Importance

The STAC-ML Markets (Inference) benchmarks represent the industry's gold standard for evaluating real-time market data inference performance. Designed by leading quantitative analysts and tech experts from major financial institutions, STAC-ML benchmarks assess a technology stack's performance, resource efficiency, and the quality of execution of supplied models. The full report, detailing these findings, can be accessed at STAC Research.

Developers working in machine learning environments can now benchmark their own models using VOLLO®, simplifying the assessment process without requiring specialized FPGA tools. More details can be found at myrtle.ai/vollo-trees or by contacting their support team at [email protected].

About Myrtle.ai

Myrtle.ai is recognized for delivering high-performance AI and machine learning software solutions, specializing in low-latency inference accelerators built on FPGA platforms from leading suppliers. Their technologies extend across various fields, including financial trading, wireless telecommunications, large language models, speech processing, and recommendation systems. Notable trademarks include VOLLO, VOLLO Accelerator, and the VOLLO logo.

The STAC brand and its associated names are registered trademarks of the Strategic Technology Analysis Center, LLC, while AMD, the AMD Arrow logo, Alveo, and their combinations are trademarks of Advanced Micro Devices, Inc.