Myrtle.ai is dialing machine learning latency down to nearly nothing.
The UK-based ML acceleration specialist has announced that its VOLLO® inference engine now supports Napatech’s NT400D1x SmartNICs, enabling machine learning inference at sub-microsecond latencies. Yes, that’s faster than a typical blink—and a serious play for environments where every nanosecond counts.
This move marks a significant milestone in edge AI computing, particularly for ultra-low latency applications like financial trading, telco operations, and cybersecurity—sectors that can’t afford to wait on bloated inference pipelines.
VOLLO + Napatech: High-Speed Matchmaking
VOLLO has already made waves as a benchmark leader, especially under the STAC® ML tests used widely in finance to measure speed and performance. The addition of Napatech’s SmartNICs—well-known for their deep penetration in the network acceleration market—means ML models can now run directly adjacent to network data, slashing processing time and unlocking real-time intelligence.
The supported model suite is impressively broad for such a latency-optimized setup: LSTM, CNN, MLP, Random Forests, and Gradient Boosting Trees are all compatible. That’s good news for developers juggling both classical and deep learning pipelines.
Edge ML for Real-Time Stakes
Why all the fuss over latency? In automated trading, a few microseconds can be the difference between profit and loss. In telecom or cybersecurity, faster inference can mean fewer dropped calls or earlier breach detection. Myrtle.ai and Napatech are offering something rare in this space: general-purpose ML inference that doesn’t sacrifice performance for flexibility.
“The demand for ever-lower latencies is unrelenting,” said Peter Baldwin, CEO of Myrtle.ai. “This new release brings VOLLO’s performance directly to Napatech’s hardware, empowering our users to push boundaries even further.”
Napatech’s Chief Product & Marketing Officer, Jarrod J.S. Siket, added, “The VOLLO compiler makes SmartNIC-based ML development incredibly approachable. It complements our hardware and deepens our value in key sectors like finance.”
ML Next to the Wire
Running inference “next to the network” isn’t just a clever phrase—it’s a growing architectural trend. With data volumes exploding and real-time needs rising, organizations are embracing edge inference to avoid the drag of centralized processing.
Myrtle.ai’s VOLLO offering aligns neatly with this evolution. By leveraging SmartNICs as ML accelerators, they’re sidestepping the bandwidth and latency challenges of shuttling data back and forth between CPUs, GPUs, and centralized datacenters.
Developers curious to try it can download the VOLLO compiler now from vollo.myrtle.ai and test their models on the NT400D1x platform.
For firms chasing the bleeding edge of low-latency ML, this collaboration looks like a smart bet.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI