myrtle.ai says its VOLLO inference accelerator set new STAC-ML Markets (Inference) records for gradient-boosted tree (GBT) inference, with STAC-audited results announced on 6 October 2026. All three tested models recorded 99th-percentile latency below 2 microseconds; the smallest sustained 50 million inferences per second at a p99 latency of 1.77 microseconds.
What VOLLO’s new benchmark results show
The results concern model inference under the STAC-ML Markets benchmark, not the full time required for a trading system to receive and prepare market data, run a model, and act on its output. The benchmark release describes GBTs but does not identify a specific library such as XGBoost or publish the tested models’ sizes or tree counts.
As an Amazon Associate I earn from qualifying purchases.
- Latency: p99 latency was below 2 microseconds on each of the three tested models.
- Smallest model: 1.77 microseconds p99 latency while sustaining 50 million inferences per second.
- Vendor comparison: myrtle.ai says p99 latency was more than 30% lower and throughput at least five times higher than the previous best results.
The company’s announcement is available in the PR Newswire release. The linked release says the results were unveiled at the STAC Summit in London on 6 October 2026.
How the results compare with another STAC-reported comparison
Runtimewire reports a separate STAC comparison of up to 42% lower p99 latency and up to 71% higher throughput when comparing systems at directly comparable model-instance counts. That is not the same numerical comparison as myrtle.ai’s headline claim of more than 30% lower p99 latency and at least five times the throughput. The available material does not reconcile the different figures, so they should be read as distinct comparisons rather than combined into one result.
Runtimewire says myrtle.ai’s release cites report SUT ID MRTL2026905, while the accessible STAC report and working-group listing identify the matching configuration as ML-20260925. Neither identifier should be treated here as definitively correct. The STAC report can be found at STACresearch.com/MRTL2026905.
Hardware and benchmark context
The tested system used an AMD Alveo V80LL Compute Accelerator installed in a Blackcore ICON 3132-SM+ server. STAC audited the results. myrtle.ai says that, following its STAC Tacana results announced in April 2026, VOLLO now holds deterministic-latency records for both decision trees and neural networks.
Rank #2
- [Local AI Inference & 70B Model Ready] Equipped with the AMD Ryzen 7 PRO 8845HS processor, NEXUS is engineered for heavy local AI workloads. With a full-size GPU bay, it runs 70B LLMs natively without an internet connection. Ideal for AI developers and tech enthusiasts who need private environment for coding and model testing.
- [132TB Mass Storage with ZFS Integrity] Features a hybrid storage architecture (3×NVMe + 4×3.5" HDD) supporting up to 132TB. Utilizing the enterprise-grade ZFS file system and ECC memory, it prevents data corruption and bit rot—a must-have for professional photographers and video editors safeguarding 4K/8K RAW footage.
- [OpenClaw-Driven Automation Workflow] The built-in OpenClaw execution layer allows complex automated tasks to be processed locally. Even when offline, your backup schedules and AI file organization continue seamlessly. Say goodbye to monthly cloud subscriptions and high latency.
- [Dual 10GbE & USB4 Ultra-Connectivity] Experience server-class speeds with dual 10GbE ports and a 40Gbps USB4 interface. It enables multi-user real-time collaboration on large project files directly from the NAS, ensuring zero-lag editing for creative studios and production teams.
- [Open-Source ZimaOS for Total Privacy] Running on the fully open-source ZimaOS, NEXUS ensures your data stays physically on-premise with no backdoors. It acts as a "Digital Fortress" for privacy-conscious families and small businesses who demand absolute data sovereignty.
These figures describe a defined inference benchmark and should not be read as end-to-end trading latency. Market-data receipt and preparation, downstream decisions, and system integration can all add time beyond model inference itself.
What developers can try
myrtle.ai CEO Peter Baldwin said developers can test their own models on VOLLO without FPGA expertise. That lowers one stated barrier to evaluating the accelerator, but the announcement does not specify test access, availability, deployment requirements, or costs. The benchmark claims also are not a hands-on evaluation of VOLLO on a reader’s own workload.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
