Artificial intelligence consumes energy on a scale that is becoming difficult to ignore, calling into question the very logic of infinite growth.
Google has released the Gemma 4 family of models – from compact versions for mobile devices to powerful systems capable of competing with solutions twice their size.
Google has unveiled TurboQuant, an algorithm that compresses AI's working memory sixfold, which could fundamentally change the approach to neural network infrastructure.
AI: Events
AMD at MLPerf Inference 6.0: A Million Tokens Per Second and a Debut in Video Generation
Technical context • Infrastructure
AMD has presented its MLPerf Inference 6.0 results, showcasing new performance records, the first video generation tests, and scaling up to the cluster level on the Instinct MI355X GPU.
ASUS has unveiled the UGen300 – a compact USB accelerator designed for running AI directly on-device, без cloud или подписок, and with support for language models.
AI: Events
AI Factories as Part of the Power Grid: NVIDIA and Partners Change Their Approach to Electricity Consumption
Technical context • Infrastructure
NVIDIA and Emerald AI have proposed treating large-scale AI infrastructures not as passive energy consumers, but as active participants in the energy system.
How the exaCB continuous benchmarking system helps monitor the performance of dozens of scientific applications on the exascale supercomputer JUPITER.
OpenAI has released GPT-5.4 – a model with a one-million-token context window, built-in computer control capabilities, and a reduced error rate. But what does this have to do with our power grid?
Alibaba has open-sourced the HiClaw and CoPaw bundle – a lightweight solution for AI agents that consumes significantly less memory and runs locally.