Edge AI Innovations #61

Hey, here are some of the note worthy news on Edge AI you won't want to miss:

1. VeriSilicon has launched a new class of GPGPU‑AI computing IPs tailored for automotive and edge‑server applications, combining high-performance general‑purpose GPU cores with dedicated AI accelerators at a fine-grained level. This hybrid architecture excels in sparsity-aware scheduling, supports mixed-precision formats (INT4/8, FP4/8, BF16, FP16, TF32), and delivers superior throughput for transformer-based models like LLMs—while maintaining energy efficiency in thermally constrained environments. Featuring high-bandwidth interfaces (3D-stacked memory, LPDDR5X, HBM) and support for PCIe Gen5/6 and CXL, plus multi-chip scale-out, this IP is optimized for edge inference and even incremental training. The offering natively supports popular AI frameworks (PyTorch, TensorFlow, ONNX, TVM) and a new GPCL programming layer, enabling developers to execute and scale large models like DeepSeek efficiently on next-gen edge servers and embedded platforms. This release signals a key shift: edge AI platforms now combine flexible parallel computing with specialized accelerators to manage complex multimodal workloads, real-time decision-making, and low-latency inference—all with the scalability to handle future AI model demands.

2. NEXCOM unveiled the EBC‑370/EBC‑370X: a compact 3.5″ single-board computer designed for advanced robotic and edge AI workloads. Powered by Intel's 13th Gen Core i7 CPUs (15 W/28 W TDP) and integrated Iris Xe graphics, it supports up to 32 GB DDR5 at 5200 MHz and offers an MXM slot for GPU or AI-accelerator upgrades. With multiple 2.5 GbE ports, USB 3.2 Gen2, RS‑232/422/485, M.2 interfaces for LTE/5G, and rich I/O options, the system excels at real-time image processing, LiDAR-enabled navigation, and autonomous machine control. Its rugged design operates from –20 °C to 70 °C, making it ideal for deployment in industrial robots, AMRs, smart factories, CNC machines, and inspection systems. The EBC‑370X stands out as a high-density, scalable edge AI controller perfect for space- and power-constrained applications, offering both flexibility and future-proofing for increasingly demanding vision and inference tasks.

3. OnLogic’s new Helix 500 series marks a significant leap forward in industrial edge AI platforms by combining desktop-class performance with rugged, fanless reliability. The Helix 500 chassis houses Intel’s 10th‑Gen Comet Lake CPUs (up to i7‑10700T, 8 cores/16 threads, 35 W TDP) alongside Intel UHD 630 graphics, all inside a compact 154 × 210 × 50 mm enclosure that can operate from 0 °C to 50 °C under full 24/7 load. It supports up to 64 GB DDR4 RAM, dual M.2 slots (NVMe/SATA) for high-speed storage, and optional VPU accelerators, making it ideal for real-time inferencing, machine vision, and robotics tasks. The unit also includes dual Gigabit Ethernet (with Intel vPro/AMT), triple DisplayPort outputs, rich USB connectivity, M.2 Wi-Fi/Cellular, and DIN-rail or VESA mounting—meeting the connectivity and form-factor needs of sensor-rich industrial environments. With its emphasis on passive cooling and modular expandability, the Helix 500 offers developers a reliable, high-density edge AI server for smart factories, AMRs, CNC machines, and embedded vision systems.

This newsletter focuses on the technological advancements and innovations that can affect Edge AI engineers, hands-on developers and decision makers. There are more exciting Edge AI news on the horizon, stay tuned for the weekly reports.