NVIDIA has released the Nemotron 3 Ultra model, which leads open models in both accuracy and efficiency for agentic Register Transfer Level (RTL) coding tasks.
NVIDIA outlines how extreme co-design with its BlueField processors can address the infrastructure challenges posed by Agentic AI workloads. The company argues that as agentic systems trigger complex chains of model calls, tool interactions, and data lookups, traditional infrastructure patterns are insufficient for maintaining performance.
A video analytics AI agent capable of perceiving, reasoning, and acting on massive amounts of video footage must be integrated with existing workflows to be useful. These workflows include content management systems, messaging platforms, databases, ticket queues, and escalation paths.
NVIDIA has introduced a method using AI agents to accelerate the development of lightweight OpenUSD runtimes. This approach addresses the traditional difficulty of adapting large existing codebases for teams requiring specific memory footprints.
Developers building video analytics applications across large spaces must track the same object as it moves between camera views. Single-camera 2D tracking lacks reliable depth information and typically loses track of the object when it leaves the frame, limiting applications such as warehouse safety, retail analytics, and smart-building monitoring.
NVIDIA has introduced a new PTX instruction for carryless multiplication in CUDA 13.3, addressing a long-standing gap where GPUs lacked native support for this operation despite its presence on x86 CPUs for over fifteen years.
NVIDIA NeMo provides capabilities for coding AI agents to operate long-running machine learning workflows, specifically targeting reinforcement learning research.
NVIDIA demonstrates that using autonomous coding AI agents to post-train the Cosmos 3 vision reasoning model can push accuracy above 90% with minimal manual effort.
The NVIDIA Nemotron Model Reasoning Challenge invited the Kaggle community to explore techniques that improve reasoning accuracy under shared constraints of open models, benchmarks, and infrastructure. By the close of the competition, more than 5,000 active participants across 4,000 teams had generated thousands of submissions.
NVIDIA researchers have developed an Ising decoding method that reduces logical error rates for color code quantum error correction by more than 300 times.
Estimating the probability of low-likelihood, high-impact events in science, engineering, and finance is challenging because brute-force Monte Carlo sampling requires an excessive volume of model iterations.
NVIDIA addresses the challenge of rigorously evaluating robotics foundation models, which can now follow natural language instructions to manipulate objects. The company identifies evaluation as a critical unsolved problem and presents a new method to address these issues.
Large language model (LLM) training workloads increasingly hit GPU memory limits before compute is fully utilized. Model weights, gradients, optimizer states, communication buffers, and intermediate activations all compete for limited GPU high-bandwidth memory (HBM). As model size, sequence length, and batch size grow, HBM capacity often becomes the primary scaling bottleneck.
NVIDIA outlines how AI performance depends on three dimensions: accuracy, throughput, and interactivity. The company emphasizes that practical systems must optimize these factors together rather than focusing on a single metric.
This article from NVIDIA's developer blog explains how kernel fusion can improve memory bandwidth and reduce kernel launch overhead in GPU code. It addresses the common bottleneck where fast GPU compute outpaces device memory, preventing kernels from being fully utilized.
NVIDIA has introduced the BioNeMo Agent Toolkit to accelerate biomolecular structure prediction and co-folding workloads driven by AI agents. This toolkit addresses the need for speed and scalability across key pipeline steps, including Multiple Sequence Alignment (MSA) generation, co-folding inference, serving, and multi-GPU scale-out.