AMD Introduces Hyperloom for Autonomous Inference Optimization on GPUs
AMD's Hyperloom delivers a median 1.73× inference speedup on Instinct GPUs, with gains up to 7.31×, according to the company.
676 articles
The tools, frameworks, and platforms developers use to build with AI — coding assistants, agent frameworks, APIs, and developer tooling, with the updates that change how you ship.
AMD's Hyperloom delivers a median 1.73× inference speedup on Instinct GPUs, with gains up to 7.31×, according to the company.
AMD demonstrated a 43% to 54% increase in throughput for GLM-5.2-MXFP4 on four MI355X GPUs for prompts over 40,960 tokens.
Benchling processes over 600 code sessions daily across 250 tenants with zero security incidents using Amazon Bedrock AgentCore.
Positron, Posit’s IDE for data science, now runs on Amazon SageMaker AI, enabling seamless integration of data access and AI assistance within a single environment. A synthetic 50,000-loan portfolio was used to demonstrate the workflow.
OpenAI announced new learning paths for developers, leaders, educators, and students on September 21, 2026, to help build practical AI skills.
NVIDIA showcased five clean energy startups using AI to accelerate grid modernization and nuclear operations, with Southern California Edison cutting interconnection review times to two minutes.
Meta’s Muse AI app, downloaded over 900,000 times in its first week, prioritizes data collection over task automation, according to user testing.
ScrollEd, a Palo Alto startup, is launching an app that transforms text into scrollable video and quiz content, aiming to engage students with short-form educational material.
Unity Technologies announced plugins for Claude Code and OpenAI Codex on September 19, 2026, to prevent AI agents from relying on outdated tutorials.
HuggingFace announced Optimum Intel v2.2 and OpenVINO 2026.4 on September 18, 2026, expanding support for language, vision, and generative AI tasks.
Amazon SageMaker AI now deploys Hugging Face models using coding agents, reducing deployment time from days to hours. The tool supports real-time endpoints and multiple deployment modes.
The FAA plans to spend $875 million on an AI system called SMART to address air traffic control challenges, with a 12-year rollout beginning in the Washington, D.C., area.