Official NVIDIA Blog & Developer Blog RSS (AI-filtered) + GitHub + Hugging Face.
50 updates · page 1 of 4
Turning a simulation idea into a working application means assembling assets, connecting physics and rendering, and checking that the scene behaves as intended.
Why it matters: Developer tooling release — update for the latest fixes, features, and compatibility.
Preparing CAD assets for robotics simulation requires more than converting geometry to OpenUSD: developers must configure and validate materials, collision...
Why it matters: A new model option to evaluate — compare quality, price, and context length against what you use today.
NVIDIA's KGMON team placed second in the KDD Cup 2026 Data Agents competition with a smaller, clearer harness that is easier to verify, built to answer natural-language questions.
Why it matters: Teams building data agents can study a competition system that won by shrinking the harness rather than adding to it.
NVIDIA and Microsoft said they are co-engineering hardware and software so AI agents can run on Windows PCs, including RTX Spark, as described by Jensen Huang and Satya Nadella.
Why it matters: People on Windows PCs can expect NVIDIA and Microsoft agent software tied to RTX Spark hardware.
How we taught robots to assemble GB300 tester trays and what it taught us about robot learning, mechanical intelligence, and good old-fashioned engineering The...
Why it matters: Developer tooling release — update for the latest fixes, features, and compatibility.
AI factories are some of the most complex operations in the world, combining GPUs, CPUs, switches, DPUs, and SuperNICs alongside schedulers, orchestration...
Why it matters: Developer tooling release — update for the latest fixes, features, and compatibility.
NVIDIA describes scaling decision optimization to 100 million variables and beyond with mPDLP in cuOpt for larger supply-chain and energy-grid models.
Why it matters: Operations teams can use mPDLP in cuOpt when a planning model grows past 100 million variables.
NVIDIA Megatron Core's bitwise determinism makes large-scale pretraining easier to debug, validate, and resume, including runs with trillions of parameters across thousands of GPUs.
Why it matters: Training teams can resume and compare pretraining runs more reliably when bitwise determinism is on.
NVIDIA Green Contexts let a single process share a GPU across independent work, such as a latency-sensitive operator next to a background kernel or preprocessing next to inference.
Why it matters: GPU developers can isolate latency-sensitive work from background kernels in the same process.
NVIDIA says telecom operators are building AI on open models for trust, control, and customization, and its telecom report finds 89% of respondents reporting open source models and software.
Why it matters: Telcos can read why peers are choosing open models for networks and customer care, with 89% of respondents citing open source.
Breast cancer is the most commonly diagnosed cancer among American women — yet the gaps in care are wide.
Why it matters: Developer tooling release — update for the latest fixes, features, and compatibility.
Local AI is becoming more useful by the token.
Why it matters: Developer tooling release — update for the latest fixes, features, and compatibility.
GPT-6 Astra Ultrafast now runs on NVIDIA Blackwell GPUs in the OpenAI API and for eligible ChatGPT Work and Codex users, with up to 8x faster token generation than Astra Standard.
Why it matters: Developers on Ultrafast can get up to 8x faster token generation, which can shorten coding-agent waits.
PixelUMM is an image-to-text model released under an other license.
Why it matters: Developers can try PixelUMM when they need an image-to-text model.
AI agents are becoming a standard part of development workflows, but general-purpose agents weren't built with specialized infrastructure software such as...
Why it matters: Developer tooling release — update for the latest fixes, features, and compatibility.