edge ai News - Page 3 of 44 - CNX Software

July 20, 2026 by Debashis Das - No Comments

Globalscale Case8 – A MediaTek Genio 520/720 cyberdeck for gaming, home automation, and education (Crowdfunding)

Case8 Platfrom with gaming, home automation, and education variants

Globalscale Technologies has launched the Case8 series of open-source cyberdeck-style computing platforms based on MediaTek Genio 520 and Genio 720 processors. The platform is offered in hacker (PCBA only), gaming, home automation, and education variants. The Case8 is built around an OSM-L compute module with a MediaTek Genio 520 or Genio 720 SoC with up to eight cores, a Mali-G67 GPU, and a 10 TOPS NPU. Depending on the configuration, it comes with 4GB, 8GB, or 16GB LPDDR5 memory and 64GB or 128GB UFS storage, along with Gigabit Ethernet, Wi-Fi 6, Bluetooth 5.4, stereo speakers, microphones, and optional 4G/5G connectivity. Display options include a 5-inch panel for the gaming model and 7-inch touch or non-touch displays for the home automation and education variants, with camera configurations varying by model. Case8 cyberdeck platform specifications: OSM-L compute module (one or the other) MediaTek Genio 520 CPU – 2x Cortex-A78 up to 2.2GHz, 6x […]

July 20, 2026July 20, 2026 by Jean-Luc Aufranc (CNXSoft) - No Comments

Geniatech XPI-3576-CM5 – A Raspberry Pi CM5-compatible Rockchip RK3576 AI Compute Module with WiFi 6, Bluetooth 5.3

Geniatech XPI-3576-CM5 is a Rockchip RK3576 octa-core Arm Cortex-A72/A53 system-on-module based on the Raspberry Pi CM4/CM5 form factor with up to 16GB LPDDR5 and 128GB eMMC flash. The module also features a RealTek RTL8211F Gigabit Ethernet controller, an Ampak AP6275S wireless module for dual-band WiFi 6 and Bluetooth 5.3 connectivity, and a Rockchip RK806S-5 PMIC for power management. It features two 100-pin B2B connectors, allowing it to be plugged into carrier boards designed for the Raspberry Pi CM4/CM5, with some limitations as we’ll see further below. Geniatech XPI-3576-CM5 specifications SoC – Rockchip RK3576 Octa-core CPU – 4x Cortex-A72 cores at 2.2GHz, 4x Cortex-A53 cores at 1.8GHz GPU – Arm Mali-G52 MC3 GPU at 1 GHz with support for OpenGL ES 3.2, OpenCL 2.0, and Vulkan 1.1 NPU – 6 TOPS (INT8) AI accelerator with support for INT4/INT8/INT16/BF16/TF32 mixed operations. VPU Video Decoder – H.264, H.265, VP9, AV1, and AVS2 up to […]

July 20, 2026 by Debashis Das - No Comments

Efinix Titanium Edge FPGA family targets low-power edge AI and vision systems

Efinix has introduced the Titanium Edge FPGA family designed for low-power edge AI and machine vision applications. Based on the company’s Titanium FPGA architecture, the new family introduces lower static power consumption, configurable 2.5 Gbps MIPI I/O, built-in single-event upset (SEU) correction, and System-in-Package (SiP) variants that integrate SPI flash and HyperRAM for compact systems. The initial Titanium Edge FPGA family includes the Ti125, Ti40, Ti70, and Ti95 devices, with approximately 39K to 123K logic elements for embedded AI, machine vision, robotics, medical systems, and various other applications. Compared to the high-end Titanium Ti375 and the cost-efficient Topaz FPGA family, the new devices are designed to have about 50% lower static power consumption while retaining the company’s Quantum architecture. Efinix Titanium Edge Ti125 and Ti125 SiP specifications: FPGA Fabric Efinix Quantum architecture Up to 122,792 logic elements (109,440 XLR cells) 288 DSP blocks Embedded Memory – 5.9 Mbit SRAM Memory – […]

July 19, 2026 by Jean-Luc Aufranc (CNXSoft) - 9 Comments

Beelink EQi 304 review – Part 3: Ubuntu 26.04 on a Wildcat Lake mini PC

Beelink EQi Wildcat Lake mini PC Ubuntu 26.04 review

After checking out the hardware with an unboxing and a teardown of the Beelink EQi Wildcat Lake mini PC, we reviewed the Intel Core 3 304 computer with Windows 11 in detail. We’ve now had time to test the Beelink EQi mini PC running Ubuntu 26.04 with Linux 7.0/7.1 kernels. We’ll report our experience with feature testing, benchmark results, 2.5GbE, 10GbE, and WiFi 6 networking evaluation, storage and USB/Thunderbolt tests, and more, including power consumption measurements. Ubuntu 26.04 installation and system information I’d usually resize the Windows partition to install Ubuntu alongside it in dual boot configuration. However, since the Beelink EQi mini PC also comes with two PCIe Gen4 x1/x2 sockets, I installed Ubuntu 26.04 on a 256GB NVMe (PCIe Gen3 x4) SSD fitted into the PCIe x2 socket and left the Windows 11 partition on the UFS storage alone. Despite the Intel Core 3 304 chip being a […]

July 17, 2026 by Jean-Luc Aufranc (CNXSoft) - No Comments

reCamera Pro “Open AI Camera” supports computer vision, LLM, VLM, STT, and TTS workloads

Seeed Studio’s reCamera Pro is an AI camera powered by a Rockchip RV1126B quad-core Cortex-A53 SoC with a 3 TOPS NPU for on-device AI vision, but also large language models (LLMs), vision language models (VLMs), as well as speech-to-text (STT) and text-to-speech (TTS) with built-in audio capabilities. It offers a powerful update to the earlier reCamera modular AI camera based on an SG2002 RISC-V AI SoC, with up to 4GB RAM, 16GB eMMC flash, an 8MP camera sensor supporting up to 4K @ 30 FPS, an M12 lens mount, a built-in 1W speaker, two microphones, Gigabit Ethernet with PoE, dual-band WiFi 5 connectivity, a USB-C port for power and data, and and expansion through a GPIO and MIPI DSI connectors. reCamera Pro specifications: SoC – Rockchip RV1126B CPU – Quad-core Arm Cortex-A53 up to 1.2 GHz with 32KB L1 I-Cache and 32KB L1 D-Cache, unified 512KB L2 Cache MCU – […]

July 16, 2026July 16, 2026 by Jean-Luc Aufranc (CNXSoft) - No Comments

NVIDIA launches smaller, mainstream Jetson T2000 and T3000 modules for Edge AI and robotics applications

NVIDIA has just launched the Jetson T2000 and T3000 “mainstream” modules as smaller and lower cost options to the Jetson T4000/T5000 modules introduced last year for Edge AI and robotics applications. The Jetson Thor T3000 and the industrial version IGX Thor T3000, which adds functional safety, both deliver up to 865 FP4 TFLOPS of AI compute. They feature a 1536-core NVIDIA Blackwell GPU, an eight-core Neoverse Arm CPU, 32GB of LPDDR5X memory with 273GB/s bandwidth, and 25 GbE connectivity. The Jetson Thor T2000 delivers up to 400 TFLOPS of AI performance, features a 1024-core Blackwell GPU, 16GB LPDDR5 with 137GB/s bandwidth, and 10 GbE networking. Both are about half the size of the Jetson T4000/T5000 modules. Here’s a preliminary comparison between the Jetson Thor T2000, T3000, T4000, and T5000 modules. It’s preliminary because, as I write this article, NVIDIA has yet to provide the full specifications for the new T2000 and […]

July 15, 2026July 15, 2026 by Jean-Luc Aufranc (CNXSoft) - 7 Comments

Reverse-engineering brings RK3576 NPU support to open-source Rocket driver for mainline Linux

Back in 2024, we noted that the open-source driver for RK3588’s NPU by Tomeu Vizoso delivered decent performance for object detection. Since then, good progress has been made, and it’s now known as the Rocket driver in mainline Linux. However, the Rockchip RK3576 SoC is not supported by the Rocket driver in mainline yet, despite sharing a similar NPU. The good news is that Ga Hing Woo has been working on mainline Linux support for the Rockchip RK3576 NPU using the Rocket driver stack and has tested the updates on the Radxa ROCK 4D running Linux 7.1-rc5. This required some serious reverse-engineering work due to hidden registers not documented in the TRM. Ga Hing Woo (also written as Jiaxing Hu) sent us an email explaining some of the challenges: Single int8 convolution now comes out byte-exact against a CPU reference on the open stack. Getting there meant reverse-engineering the compute […]

July 13, 2026July 13, 2026 by Jean-Luc Aufranc (CNXSoft) - 2 Comments

Ambient Scientific GPX10 Pro MCU delivers years of always-on AI on a coin-cell battery

Ambient Scientific GPX10 Pro Cortex M4 MCU always on embedded AI

Ambient Scientific GPX10 Pro is a processor/microcontroller for always-on embedded AI applications on power-constrained edge devices. With a consumption of under <100μW, it can deliver years of always-on AI on one coin-cell battery. The SoC features ten software-programmable MX8 AI cores leveraging the company’s DigAn technology, ultra-low power ADC, an 8-bit DVP camera interface, 2048 KB SRAM, support for numerous peripherals, and an Arm Cortex-M4F core for non-AI workloads. GPX10 Pro key features and specifications: MCU core – Arm Cortex-M4F core clocked at 100 KHz to 100 MHz Memory 2048 KB system SRAM 64KB retention SRAM in “Always On” block Storage – External SPI or QSPI flash 10x MX8 programmable AI Cores for ultra-low power deep learning applications Based on the DigAn AI engine, which can change behavior between being a master and a slave. Up to 2,560 MAC operations per clock cycle (256 MAC per cycle per core) Peak […]