Des tests confirment que les LLM fonctionnent bien sur des ordinateurs portables sans GPU

Français

Une évaluation indépendante sur du matériel standard montre que les grands modèles de langage peuvent offrir des performances utilisables sans GPU. Les tests se sont concentrés sur la vitesse et l'aspect pratique pour les tâches quotidiennes.

Un ordinateur portable Intel i5 doté de 12 Go de RAM a servi de plateforme de test pour huit modèles utilisant Ollama et la quantification GGUF. Les performances ont varié de 34 jetons par seconde pour les plus petits modèles à environ 4 jetons par seconde pour les plus volumineux.

Articles connexes

Framework Laptop 13 Pro showcased on stage with 20-hour battery highlight, alongside Laptop 16 updates, OCuLink GPU kit, and wireless keyboard.
Image générée par IA

Framework launches Laptop 13 Pro with Panther Lake chips and 20-hour battery

Rapporté par l'IA Image générée par IA

Framework announced the Laptop 13 Pro, a redesigned 13-inch model featuring Intel's Core Ultra Series 3 Panther Lake processors and a battery delivering over 20 hours of life. The company also unveiled updates to its Laptop 16, an OCuLink external GPU kit, and a wireless keyboard during its Next Gen event on Tuesday. Preorders for the Laptop 13 Pro start at $1,199 for the DIY edition, with shipments beginning in June.

Linux user finds local LLMs easier than on Windows

A tech enthusiast shares their experience using Linux to run local large language models, claiming it is simpler than on Windows. They highlight the ability to access a ChatGPT-like interface directly in the terminal. The article was published on March 10, 2026.

Tutorial demonstrates running LLMs on Arduino UNO Q

Rapporté par l'IA

A new tutorial shows how to run large language models and vision-language models locally on the Arduino UNO Q microcontroller. Edge Impulse's Marc Pous has outlined steps using the yzma tool to enable offline AI inference on the board's Linux environment. This approach allows for privacy-focused applications in edge computing.

Linux

Study shows AI can deanonymize online users from posts

Technologie

OpenAI deploys GPT-5.5 Instant as ChatGPT's new default model

Technologie

Google announces Gemini 3.1 Pro AI model

Elon Musk praises efficiency of Tesla's AI4 computer

Elon Musk has stated that Tesla's AI4 computer, capable of handling the complexities of real-world driving, operates with only about a quarter of the power of an H100. This observation leads him to a specific conclusion about its performance. The comment was made in a post on X, formerly Twitter.

Intel launches mainstream Core Series 3 (Wildcat Lake) chips on 18A process

Rapporté par l'IA

Building on its Core Ultra Series 3 (Panther Lake) processors unveiled at CES 2026, Intel has announced Core Series 3 chips codenamed Wildcat Lake for mainstream and value laptops. These mark the first non-Ultra Core CPUs to use the advanced Intel 18A manufacturing process, delivering major performance gains, AI improvements, and all-day battery life. Laptop makers including HP, Acer, Lenovo, Dell, and Samsung plan to integrate them throughout 2026.

mercredi 22 avril 2026 11h22

Google unveils TPU 8t and TPU 8i chips for agentic AI

lundi 20 avril 2026 19h12

Quantum method promises AI boost from computers

jeudi 02 avril 2026 10h35

Red Hat and NVIDIA lead MLPerf Inference v6.0 benchmarks

mardi 10 mars 2026 10h22

Kali Linux launches local AI tools for penetration testing

mardi 03 mars 2026 12h23

Linux Foundation promotes LFWS307 SLM course amid rising AI trends

samedi 21 février 2026 09h38

Ollama releases version 0.17 with OpenClaw onboarding improvements

Ce site utilise des cookies

Nous utilisons des cookies pour l'analyse afin d'améliorer notre site. Lisez notre politique de confidentialité pour plus d'informations.
Refuser