Klepsydra AI – Space Edition

Klepsydra AI is a cutting-edge software framework for Edge AI applications, delivering high-efficiency AI inference. It offers high performance, broad compatibility across edge computers and operating systems, and a lightweight, independent design with no third-party dependencies, making it a powerful and adaptable solution for next-generation edge computing.

The Space Edition of Klepsydra AI is built for space applications, ensuring high reliability and performance in extreme environments. ECSS-qualified by ESA, it supports radiation-hardened processors used in satellites and spacecraft. Additionally, it is compatible with the RTEMS, VxWorks and DDC-I DEOS operating systems, ensuring seamless integration for critical missions.

Klepsydra AI is a high-performance deep neural network engine for edge computers. It enables customers to deploy existing or newly trained models just like with standard edge AI solutions—only with a major advantage: up to 4x higher data throughput.

Compatibility:

• Among the networks already validated:
  • AlexNet, MobileNet V1 and V2, YOLOX, YOLOv4-v8 (S and N variants), ResNet50, VGG19, UNet, and LSTM.
• Hardware and OS Compatibility:
  • x86, ARM Cortex-A, RISC-V, and SPARC processors, and Linux, RTEMS6, VxWorks, and DDC-I operating systems.
• GPU Support:
  • NVIDIA GPU with CUDA kernels, CUDA Graphs, and TensorRT
    workloads.

A detailed support matrix is available upon request

Core Technology:

• Performance Improvement: 2× to 8× faster data processing and 75% power reduction.
• Key Techniques: Lock-free ring buffers, FPU vectorisation, and dedicated ring buffers per thread.
• Optimisation: Smart thread configuration for low latency, maximum throughput, or minimal CPU usage.

SDO (Streaming Distribution Optimiser)

• Tool Functionality: Optimises AI performance on target processors.
• Optimisation Features: Allows customisation for latency, throughput, or resource constraints.
• Optimisation Process: Cloud-based software collects data, runs an optimiser, and deploys the best configuration.