generative ai News - Page 3 of 12 - CNX Software

March 30, 2026March 30, 2026 by Jean-Luc Aufranc (CNXSoft) - 5 Comments

Select the right hardware for your local LLM deployment with this online guide

When it comes to deploying local LLMs, many people may think that spending more money will deliver more performance, but it’s far from reality. That’s why Sipeed created the “AI Agent Local LLM Inference Device Deployment Guide” hosted on the llmdev.guide website. The website lists common hardware with price, performance (tokens/s), power consumption, and more for various LLMs. If we take Qwen3.5 9B as an example, we can see that $4K+ hardware like NVIDIA DGX Spark or Apple Mac Studio M3 delivers about the same TPS as a machine equipped with a $260 Intel Arc B580 12GB GPU. If money is no object and you’d like the best performance, the NVIDIA GTX 5090 32GB makes the most sense. I reckon the price comparison is imperfect because some data points reflect the price of a complete system, while others only list the price of a graphics card. However, for Qwen 122B-A10B, […]

March 25, 2026March 28, 2026 by Jean-Luc Aufranc (CNXSoft) - 6 Comments

Alibaba XuanTie C950 – A powerful, RVA23-compliant 64-bit RISC-V core for Edge AI computing

Xuantie C950 high performance 64-bit RISC-V CPU

Alibaba has introduced the XuanTie C950 high-performance, 64-bit multi-core CPU IP with an out-of-order superscalar microarchitecture, RVA23 profile compliant, and support for “all optional extensions” such as Vector Crypto, Zacas, and Zama16. The company also says the XuanTie C950 supports the proprietary XuanTie AME (Attached Matrix Extension) ISA and supports integration with the company’s XuanTie TPE (Tensor Processing Engine) IP. The new 64-bit RISC-V core will be found in SoCs with up to eight cores per cluster, targeting high-performance applications, such as cloud computing, edge computing, and AI computing. XuanTie C950 specifications: Architecture – RVA23 Profile Up to 8x cores clocked at 3.2 GHz; 22+/GHz Specint2006 base, or a score of around 70 at 3.2 GHz Pipeline – Superscalar out-of-order microarchitecture with 8-wide decode Floating Point – RISC-V F/D Extension Vector – RISC-V Vector Extension v1.0 with Vector Crypto support Matrix – XuanTie TPE coprocessor integration (AME v0.5) Hypervisor – […]

March 18, 2026March 18, 2026 by Jean-Luc Aufranc (CNXSoft) - No Comments

PycoClaw – A MicroPython-based OpenClaw implementation for ESP32 and other microcontrollers

PycoClaw is a MicroPython-based platform for running AI agents on ESP32 and other microcontrollers that brings OpenClaw workspace-compatible intelligence to resource-constrained embedded devices. We had already covered the C-based Miniclaw for ESP32-S3 SoCs, the PycoClaw’s developer (Jonathan Peace) told CNX Software that it is a “full OpenClaw-compliant agent” that supports more LLM providers (OpenAI, Gemini, Ollama, etc.), interfaces with not only Telegram, but also ScriptO Studio and WebRTC, and offers features like OTA updates, extensions, and battery-optimized operation. The table below compares PycoClaw to OpenClaw, Nanobot, PicoClaw, NullClaw, and MimiClaw. MimiClaw still offers the lowest footprint and highest efficiency, but PycoClaw appears to offer many more features, including improved GPIO support. It works on ESP32-S3 with at least 8MB flash and PSRAM, ESP32-P4, and should soon support Raspberry Pi RP2350 boards with PSRAM as well. PycoClaw can be installed on supported hardware through a “one-click install” using a compatible web […]

March 17, 2026 by Debashis Das - No Comments

ADLINK DLAP-701 – An NVIDIA Jetson T5000/T4000 Edge AI platform for humanoid robots and vision sensing systems

Adlink DLAP 701 Edge AI Platform Powered by NVIDIA Jetson Thor

ADLINK has just launched the DLAP-701 Series, a NVIDIA Jetson T5000/T4000-based compact edge AI platform designed for humanoid robots, autonomous mobile robots (AMR), and vision sensing systems (VSS). It supports up to 128GB LPDDR5X memory and features various I/O options, including dual Gigabit Ethernet, a QSFP port supporting 4×25GbE LAN, multiple USB 3.2 ports, and HDMI output, along with M.2 slots for Wi-Fi 6, 5G, and NVMe storage, as well as an mPCIe slot. It also integrates CAN-FD interfaces for robotics and vehicle control and TPM 2.0 security. With an operating voltage range of 9-36V DC and an industrial temperature range of -20°C to 60°C, it is designed for demanding edge environments. ADLINK DLAP-701 specifications Supported system-on-module – NVIDIA Jetson Thor T5000 or Jetson Thor T4000 Memory – Up to 128GB (T5000 variant), 64 GB (T4000 variant) 256-bit LPDDR5X (273 GB/s bandwidth) Storage 128GB SSD for the OS (most probably ADLINK’s ASD+ industrial […]

March 11, 2026 by Debashis Das - No Comments

AMD Ryzen AI Embedded P100 series expands with up to 12 Zen 5 cores, 80 TOPS of AI performance

AMD Ryzen AI Embedded P100 series Embedded World 2026

At CES 2026, we saw AMD launch its new Ryzen AI Embedded P100 series of SoCs, with four- and six-core models designed for Edge AI. At the time of writing, the lineup had six SKUs, with higher-core variants to be announced later. Now at Embedded World 2026, AMD has expanded the Ryzen AI Embedded P100 series with six additional SKUs available in commercial and industrial temperature grades, while the automotive-grade SoCs remain unchanged. The new SOCs integrate 8-12 Zen 5 CPU cores (upgrade from 4-6 cores ), RDNA 3.5 graphics, and an XDNA 2 NPU on a single chip to deliver up to 80 TOPS (upgrade from 50 TOPS) of combined/system AI performance AMD claims up to 39% higher multithreaded CPU performance and 2.1× higher total system TOPS compared to Ryzen Embedded 8000 Series. They support unified CPU-GPU memory, enabling low-latency processing for tasks such as multi-camera machine vision, Visual […]

March 6, 2026March 8, 2026 by Jean-Luc Aufranc (CNXSoft) - 4 Comments

Echo Pyramid enables smart voice interaction applications on M5Stack Atom ESP32 IoT controllers

Designed for M5Stack Atom, AtomS3, and AtomS3R series IoT controllers based on ESP32 or ESP32-S3 wireless SoC, the Echo Pyramid base enables smart voice interaction applications such as far-field voice recognition, voice assistants, voice control, and more. The device features a built-in speaker, a MEMS microphone, an ES8311 HD audio codec for playback and capture, and an STM32 MCU for touch areas and RGB LED management. It’s powered via a USB Type-C port and can be expanded through a 4-pin connector for I2C modules. Echo Pyramid specifications: Supported IoT controllers – M5Stack Atom, AtomS3, and AtomS3R Microcontroller – STMicro STM32G030F6P6 32-bit Arm Cortex-M0+ CPU @ 64 MHz with 8KB SRAM, 64 KB flash Audio HD Codec – ES8311, handles playback and recording Microphone – LMA3729T381-0Y3S MEMS microphone ADC – ES7210 for microphone input Built-in speaker on the bottom of the pyramid Amplifier – AW87559 Class-D speaker driver for the speaker […]

February 22, 2026 by Jean-Luc Aufranc (CNXSoft) - 16 Comments

Taalas HC1 hardwired Llama-3.1 8B AI accelerator delivers up to 17,000 tokens/s

Taalas HC1 is an AI accelerator hardwired (i.e, implemented in hardware) with Llama-3.1 8B and delivering close to 17,000 tokens/s of AI performance with the model, outperforming datacenter accelerators such as NVIDIA B200 or Cerebras chips. The Taalas HC1 is about 10x faster than the Cerebras chip, costs 20x less to build, and consumes 10x less power. The main downside is that it only works with the model hardwired into the hardware, currently Llama-3.1 8B, although we’re told it “retains flexibility through configurable context window size and support for fine-tuning via low-rank adapters (LoRAs)”. Hardware accelerators usually come with memory on one side and compute on the other. Both operate at different speeds, and the memory bandwidth is usually the bottleneck for Large Language Models. Taalas technology unifies storage and compute on a single chip, at DRAM-level density, to massively increase the performance and reduce power consumption. Ultra-fast inference can […]

February 13, 2026February 13, 2026 by Jean-Luc Aufranc (CNXSoft) - No Comments

Mimiclaw is an OpenClaw-like AI assistant for ESP32-S3 boards

MimiClaw is an OpenClaw-inspired AI assistant designed for ESP32-S3 boards, which acts as a gateway between the Telegram messaging application and Claude online LLM to control the hardware by just chatting to it. We’ve just written about PicoClaw, an ultra-lightweight personal AI Assistant for cheap Linux boards that just needs 10MB of spare RAM. It was itself inspired by Nanobot, a lightweight assistant written in Python, that’s 99% smaller, in terms of lines of code, than the original OpenClaw project that started it all. Since most of the processing is done through messaging apps and online LLMs, it was only a matter of time until this type of solution was ported to microcontrollers. MimiClaw highlights: Written in C; relies on the ESP-IDF 5.5 framework System requirements – ESP32-S3 board with 16 MB flash and 8 MB PSRAM, such as the LILYGO T7-S3, FireBeetle 2 ESP32-S3, ESP32-S3-DevKitC-1-N16R8, Seeed Studio’s XIAO ESP32S3 […]