The auto-updated technology brief
Step 3.7 Flash from StepFun delivers enterprise-scale multimodal AI capabilities on NVIDIA GPUs, enabling real-time perception, search, and reasoning across images, documents, video, and language.
NVIDIA introduces Dynamo Snapshot to address the cold-start delay in GPU inference workloads on Kubernetes, enabling faster startup and improved scalability during traffic spikes.
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.
NVIDIA introduces CUDA Tile programming to enable developers to create highly optimized GPU kernels within large existing C++ GPU codebases using tile-based programming techniques.
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.
AWS Transform – continuous modernization (preview) automatically scans code repositories to detect, prioritize, and remediate technical debt at scale, enabling continuous autonomous modernization and integration with security workflows.
NVIDIA JetPack 7.2 enhances real-world AI agent deployment on Jetson devices with optimized memory and performance, supporting one-command deployment of NemoClaw for privacy and security.
NVIDIA Quantum InfiniBand introduces intent-based security profiles in Unified Fabric Manager enabling one-click multi-tenant fabric security, supporting General, Bare Metal Cloud, and Secured Bare Metal Cloud profiles to reduce deployment time from hours or days to minutes.
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 the MCG Toolkit to help software teams automate comprehensive and auditable AI model documentation amid increasing regulatory requirements such as California’s AB-2013 and the EU AI Act.
This article is the third in a series on model quantization, focusing on converting FP8 quantized checkpoints into high-performance inference engines using NVIDIA TensorRT, enabling faster inference and higher throughput.
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.