Nvidia hat auf der GTC 2026 KI-Systeme für den direkten Einsatz auf Satelliten vorgestellt.

Plattformen wie das physikalisch gehärtete Vera Rubin Space Module ermöglichen lokale Berechnungen und autonomes Ausweichen vor Weltraumschrott. Das Modul liefert das 25-fache an Rechenleistung eines H100, um Bandbreite beim Downlink einzusparen.

#Nvidia #EdgeAI #SpaceComputing #News
https://www.all-ai.de/news/news26/nvidia-ki-all

Space Computing: Nvidia bringt KI in den Weltraum

Der Chiphersteller präsentiert neue Hardware für orbitale Rechenzentren. Satelliten werten Daten künftig direkt im All aus.

All-AI.de

I’m considering making a video series that teaches #ReinforcementLearning using a 2-wheel balancing bot. Would you be interested in learning that? If you've done RL, what frameworks do you recommend?
👇
https://shawnhymel.com/3219/an-idea-im-exploring-reinforcement-learning-for-a-balance-robot/?utm_source=mastodon&utm_medium=social&utm_campaign=general_courses_blog

#edgeAI #AI #embedded #robotics #education

An Idea I’m Exploring: Reinforcement Learning for a Balance Robot - Shawn Hymel

One idea I’ve been thinking about exploring is a project that sits at the intersection of all three: robotics, machine learning, and microcontrollers. The

Shawn Hymel
Thin Raspberry Pi CM5 fanless mini PC offers HDMI video output, 2.5GbE + GbE, dual USB, M.2 Key-M socket

Waveshare "Gigabit / 2.5G Dual Ethernet Mini-Computer (B)" is a thin Raspberry Pi CM5 fanless mini PC with a 4K HDMI video output, 2.5GbE and Gigabit Ethernet RJ45 jacks, two USB ports, a microSD card slot, and USB-C power. Internally, it features an M.2 Key-M socket for a 2230/2242/2280 NVMe SSD or AI accelerator, two MIPI DSI/CSI connectors for cameras or displays, and a 40-pin GPIO header. The flat cables for the MIPI devices can be passed through an opening under the HDMI and USB 2.0 ports of the metal case, and the GPIO headers are also accessible externally through a small cover under the enclosure. Gigabit / 2.5G Dual Ethernet Mini-Computer (B) specifications: Main board - CM5-DUAL-ETH-BASE-B Supported SoMs - Raspberry Pi CM5 or CM5 Lite (and potentially other electrically compatible modules like the Banana Pi BPI-CM5 Pro, Radxa CM5, and Orange Pi CM5, but it's not been tested

CNX Software - Embedded Systems News

Introducing ClawBox — your own dedicated AI hardware.

NVIDIA Jetson Orin Nano 8GB + 512GB NVMe SSD. 67 TOPS. 15 watts. Runs OpenClaw (open-source).

No cloud. No subscriptions. Your data never leaves your device.

Just restocked 70 units — shipping worldwide to 108 countries.

https://openclawhardware.dev

#AI #EdgeAI #NVIDIA #OpenSource #SelfHosted #Privacy #ClawBox #Fediverse

ClawBox — OpenClaw Hardware | NVIDIA Jetson AI Assistant Box | 67 TOPS, €549

ClawBox is a plug-and-play NVIDIA Jetson AI assistant box by OpenClaw Hardware. 67 TOPS, 15 watts, runs 24/7. Self-hosted private AI with browser automation & voice control. €549, ships worldwide.

ClawBox by OpenClaw Hardware
NVIDIA's GPU VRAM limits hit? AllSafeUs explores augmenting with system RAM/NVMe. At EnergenAI, TIAMAT runs 120B models with 20M tokens/day on edge clusters. We optimize memory routing across 52 tools—no synthetic VRAM needed. Real efficiency > hacks. #AIInfra #LLMOps #EdgeAI tiamat.live

Paul Couvert (@itsPaulAi)

휴대폰 크기 장치로 최대 120B 규모의 로컬 AI 모델을 실행할 수 있다는 발표입니다. 완전 로컬·프라이빗 환경에서 오픈소스 모델을 호스팅하여 OpenClaw 같은 에이전트를 24/7 구동하거나 챗봇을 대체하는 등 다양한 용도로 활용할 수 있다고 설명합니다. 고성능 모델을 저비용으로 엣지에서 운영할 수 있다는 점이 핵심입니다.

https://x.com/itsPaulAi/status/2032819238528811188

#localai #edgeai #opensource #llm

Paul Couvert (@itsPaulAi) on X

You can run local AI models up to 120B without a $10,000 Mac Studio This phone-sized device is your own server for open-source models. 100% local and private. And you can use it to: - Power an agent like OpenClaw 24/7 - Completely replace a chatbot - Literally anything that

X (formerly Twitter)

MediaTek unveils 50 TOPS Genio Pro 5100 Cortex-X925/X4/A720 SoC, 7.2 TOPS Genio 420 Cortex-A78/A55 SoC for AIoT applications

https://fed.brid.gy/r/https://www.cnx-software.com/2026/03/12/mediatek-50-tops-genio-pro-5100-cortex-x925-x4-a720-soc-7-2-tops-genio-420-cortex-a78-a55-soc-for-aiot-applications/

MediaTek unveils 50 TOPS Genio Pro 5100 Cortex-X925/X4/A720 SoC, 7.2 TOPS Genio 420 Cortex-A78/A55 SoC for AIoT applications

After launching the Genio 360/360P hexa/octa-core SoCs last month, MediaTek has now expanded the lineup with the Genio Pro 5100 and Genio 420 AIoT SoCs at Embedded World 2026. The Genio Pro 5100 is a 3nm SoC with an "all big-core" architecture and a 50+ TOPS NPU for Edge AI applications. The Genio 420, on the other hand, is a cost-efficient 6nm platform designed for smart home, retail, and industrial IoT devices. MediaTek Genio Pro 5100 The Genio Pro 5100 integrates one Cortex-X925, three Cortex-X4, and four Cortex-A720 cores, as well as an Arm Immortalis-G925 GPU, and supports LPDDR5X memory up to 8533 Mbps. It can handle up to three 4K displays, up to 16 cameras via virtual channels, and 8K30 video encode/decode, and offers interfaces such as PCIe Gen4, USB 3.2 Gen2, USB 2.0, and dual 2.5GbE MAC. Genio Pro 5100 (MT8894) specifications:  CPU - 8-core Arm v9.2 processor

CNX Software - Embedded Systems News

Mark Gadala-Maria (@markgadala)

Perplexity가 OpenClaw의 대안으로 'Personal Computer'라는 항상 켜진 AI 에이전트 제품을 발표했습니다. Mac mini 기반으로 파일·앱·세션을 24/7로 탐색하며 작동하고 '닫힌 안전 환경'을 강조, 인접 솔루션에서 발견되는 백도어 문제를 피한다고 소개합니다.

https://x.com/markgadala/status/2031825096105497064

#perplexity #openclaw #aiagent #edgeai

Mark Gadala-Maria (@markgadala) on X

🚨Perplexity just launched an OpenClaw alternative that's actually safe to use. Personal Computer. always-on AI agent. Mac mini. yours: > runs 24/7 across your files, apps, and sessions > closed secure environment — no open backdoors researchers keep finding next door >

X (formerly Twitter)