Myrtle.ai's VOLLO Achieves Record-Setting Performance in ML Inference

Myrtle.ai has announced groundbreaking results for its VOLLO inference accelerator at the STAC Summit in London. The device set new records in STAC-ML Markets benchmarks for gradient-boosted trees. VOLLO achieved below 2 microseconds latency and sustained 50 million inferences per second, significantly outperforming previous benchmarks.

These advancements, demonstrated on an AMD Alveo V80LL Compute Accelerator, are crucial for electronic trading. Here, rapid decision-making translates directly to financial returns, allowing firms to use larger models without sacrificing speed. Myrtle.ai CEO, Peter Baldwin, highlighted the ability to test models on VOLLO without FPGA expertise, showcasing its user-friendliness.

VOLLO's records extend to both decision trees and neural networks, solidifying its role in real-time trading. Its versatility also benefits sectors like telecoms, network security, and defense. Detailed results are accessible in the STAC Report, allowing developers to assess their models' performance on VOLLO by visiting myrtle.ai.