The auto-updated technology brief
DynoSim addresses the complexity of tuning modern LLM serving deployments by simulating the Pareto frontier across interacting system choices such as model backend, tensor-parallel shape, and autoscaling thresholds.
DiffusionGemma, developed by Google DeepMind and optimized for NVIDIA platforms, offers a new approach to accelerate token-by-token text generation for real-time AI applications like chat assistants and copilots, improving responsiveness and reducing serving costs.
NVIDIA's Blackwell GPU sets a new STAC-AI benchmark record for large language model inference in financial trading, enabling advanced analysis of unstructured data to improve trading insights.
NVIDIA Cosmos 3 is a frontier foundation model enabling physical AI systems like robots and autonomous vehicles to understand, predict, and act in the real world.
Training speech AI models to accurately recognize clinical terminology is challenging. NVIDIA's Nemotron Speech and agent skills help accelerate evaluation of clinical ASR models, improving recognition of complex drug names and medical terms.
NVIDIA introduces AI agents and FLARE Auto-FL to accelerate federated learning research by automating experiment design and evaluation, addressing challenges in metric improvement assessment.
NVIDIA introduces NVFP4 precision support in JAX and MaxText to accelerate large language model pretraining on Blackwell GPUs, improving throughput and reducing training time and costs.
NVIDIA explores building foundation models for financial intelligence using transaction data, highlighting the value of sequential patterns in customer histories over traditional hand-engineered features.
NVIDIA achieved a clean sweep in MLPerf Training v6.0, delivering the fastest training times at scale and highest per-accelerator performance across all benchmarks, being the only platform to submit on every test.
NVIDIA introduces the Vera CPU, designed to advance agentic AI workloads in AI factories by building on new scaling laws beyond traditional GPU-based AI training and inference.
NVIDIA introduces DFlash speculative decoding to boost inference performance by up to 15x on Blackwell GPUs, improving low-latency AI workflows by optimizing autoregressive LLM token generation.
NVIDIA demonstrates an open source example combining Hermes Agent with NVIDIA NemoClaw to accelerate and secure product research across Outlook, Slack, and GitHub, leveraging AI agents for data synthesis and decision-making.