FPGA Shows 160x Latency Reduction
A recent benchmark by Fortifai demonstrated that shifting an AI workload from 5,000 CPUs to a single FPGA resulted in a 160-fold reduction in latency, from 500ms to 3ms. The test of its Nol8 Agentic AI Data Plane also showed a 400x throughput increase, highlighting the performance advantages of FPGAs for latency-critical applications like flight control and sensor fusion.
- The Nol8 technology was acquired by Fortifai (ASX: FTI) through a deal with FastAI, which holds the license from the original developer, the Technion-Israel Institute of Technology. The platform is designed to process "data-in-motion," creating a high-speed bridge between AI model inference and real-world execution systems. - Fortifai's roadmap includes releasing a "Customer Benchmarking Engine" in July 2026 for enterprises to test real-world datasets, followed by a revenue-ready commercial platform by the end of calendar year 2026. - The key advantage of FPGAs in latency-critical systems is deterministic execution; they process tasks in predictable time windows without the software-induced jitter from operating systems or schedulers found in CPUs and GPUs. This is essential for the timing requirements of flight control and sensor fusion. - In aerospace, FPGAs are often favored over GPUs for certified systems because their logic is implemented directly in hardware, which is considered more verifiable than the complex firmware and driver stacks associated with GPUs. - Certifying complex airborne electronic hardware like FPGAs is governed by the RTCA DO-254 standard, a mandatory framework for receiving approval from airworthiness authorities such as the FAA and EASA. - For the highest Design Assurance Level (DAL A) under DO-254, a hardware failure's probability must be "extremely improbable," defined as less than 10⁻⁹ per flight hour. - In addition to latency reduction, the benchmark also demonstrated a 400-fold throughput increase from 5,000 to 2,000,000 events per second, while reducing the physical space and power requirements. - FPGAs' reconfigurability allows for post-deployment updates, enabling engineers to adapt to new algorithms or communication standards for assets with long lifecycles, such as satellites, without requiring a hardware replacement.
