18 August 2025

FPGAs, or Field-Programmable Gate Arrays, are like fully customizable mini-computers built on hardware. You can rewire them to do specific tasks extremely fast—like running neural networks—without needing a completely new chip each time. This flexibility makes them ideal for tasks where speed, power efficiency, and reliability matter.

However, traditional FPGAs are complex to design for, requiring deep hardware knowledge and specialized tooling. That’s where Myrtle.ai steps in. The Cambridge- based company is like a specialized translator—you speak in human-readable AI models, they translate to high-speed, energy-saving hardware action, bringing speed, affordability, and predictability to AI tasks in real-world settings—from financial trading to speech systems through to recommendations.

Myrtle.ai recently expanded the reach of its VOLLO® inference accelerator by enabling it on Napatech’s NT400D1x series of SmartNICs. In simpler terms, they've embedded their ultra-fast AI engine right into the network interface card—the piece of hardware that connects servers to the network—so the AI can run directly where data enters the system. Giles Peckham, Head of Marketing at Myrtle.ai, explained to Movetheneedle.news what this integration means.

Even Faster AI Decisions – Wider Range Of Applications

Previously, AI tasks ran inside the main computer—meaning data had to travel from the network to the CPU or GPU, be processed, then sent back. By running inference directly on the SmartNIC, Myrtle.ai eliminates that trip, achieving inference in under one microsecond (that’s one-millionth of a second!).

A SmartNIC is like a supercharged network card. But unlike a regular NIC, which just sends and receives data, a SmartNIC is also a mini-computer with built-in smarts. Napatech is the world leader in this space.

Peckham added that while Myrtle.ai has been focused on financial trading, the SmartNIC integration opens doors to other sectors: wireless networks, cybersecurity, and smart network management—any domain where microseconds—or even fractions thereof—matter for decisions.

Interesting in that regard is that Myrtle.ai’s VOLLO inference accelerator isn’t just for neural networks—it also handles decision trees, as well as more advanced versions like Random Forests and Gradient Boosting. A decision tree is a way for computers to make choices—like a flowchart. At each point, based on certain features, the tree asks a question ("Is value greater than X?") and then follows a branch. This continues until a final decision or prediction is reached.

Myrtle.ai’s Value Proposition

The beauty of their work, Peckham emphasised, is that rather than having deep technical know-how in hardware design, clients can rely on Myrtle.ai to convert their AI models into optimized FPGA configurations—for example, image recognition, speech, or recommendation systems—without hassle. Their offering includes not only the VOLLO platform, their FPGA-based solution for ultra-fast, low-latency AI inference, but also the Caiman-ASR tool, which converts spoken words into text instantly using FPGA acceleration. Myrtle.ai provides a full stack—from software (your trained model) to hardware (the FPGA setup)—with an approach that allows easy deployment in standard server racks. In other words, no need to learn hardware languages.

The SmartNIC integration is great news for Myrtle.ai, adding to quite an impressive list of achievements since the company was founded back in 2017. Peckham highlighted that their VOLLO solution achieved the lowest latency ever recorded in STAC-ML benchmarks—used by finance firms to assess real-time AI inference performance. It achieved 5.1 microseconds response times and handled over 800,000 inferences per second.  In another impressive achievement, in collaboration with Achronix, Myrtle.ai built a speech-to-text solution capable of recognizing spoken words across 1,000+ simultaneous audio streams. It delivered up to 20× better performance than conventional setups, replacing dozens of GPU or CPU servers with just one FPGA card.

Growth Potential and Market Outlook

Myrtle.ai is well-positioned for expansion. As industries—from finance to telecoms and cybersecurity—increasingly demand real-time AI decisions, technologies that deliver sub-microsecond, highly efficient inference are poised for growth. Meanwhile, the wider AI hardware market is shifting. Organisations are moving away from large, energy-intensive GPUs, especially for edge use cases. The rise of hardware-accelerated inference at the network edge—via SmartNICs and FPGAs—mirrors this trend. Solutions that can be deployed quickly, run lean, and stay future-proofed via software-configurability are winning – and it seems Myrtle.ai is on that winning path, too.