Tuning Flash Attention for Peak Performance in NVIDIA CUDA Tile | NVIDIA Technical Blog
…Host-side code Now let’s look at the host-side code that launches the kernel: import torch from math import ceil def tile_fmha(q, k, v, sm_scale=None, is…
Search
Accelerating Long-Context Model Training in JAX and XLA | NVIDIA Technical Blog
…integration makes NVSHMEM accessible to high-level frameworks without requiring code changes Performance benefits scale with sequence length, with dramatic improvements for sequences ≥ 256K tokens Multinode deployments see the largest gains, making…
Nsight Graphics
…Debug shader code with the Vulkan Shader Debugger, which exposes shader source in your render pipeline in real-time so you can quickly make fixes directly to the code. Profile Ray-Tracing…
SDK Manager
…Other bug fixes and stability improvements Get Started With NVIDIA SDK Manager The SDK Manager empowers developers to work seamlessly across platforms—whether you're coding on Linux, Docker, or Windows. Linux…
How to Automate AI Model Documentation with the NVIDIA MCG Toolkit | NVIDIA Technical Blog
…creation of comprehensive AI model documentation in Model Card++ format, improving transparency and regulatory compliance by extracting information directly from source code and associated files. The MCG pipeline operates in three stagesIngestion…
cuTile.jl Brings NVIDIA CUDA Tile-Based Programming to Julia | NVIDIA Technical Blog
…Some Julia language features (notably iterator-based ‘for’ loops) aren’t supported in kernels or generate inefficient code The integration with CUDA.jl needs to improve to facilitate coexistence with SIMT kernels…
Building Telco Reasoning Models for Autonomous Networks with NVIDIA NeMo | NVIDIA Technical Blog
…Evaluation demonstrates significant accuracy gains (from ~20% to ~60%) for incident summary prediction and root-cause resolution, with ongoing robustness improvements via tool-calling benchmarks, LLM-as-a-judge safety checks, controlled…
Cut Checkpoint Costs with About 30 Lines of Python and NVIDIA nvCOMP | NVIDIA Technical Blog
…Asymmetric numeral systems (ANS) is a modern entropy coding technique that nvCOMP implements as a GPU-native codec (gANS) optimized for raw throughput. Both are lossless and exploit statistical patterns in 1B…
CUDA 13.2 Introduces Enhanced CUDA Tile Support and New Python Features | NVIDIA Technical Blog
…NVIDIA Nsight Copilot is a free AI-powered CUDA coding assistant that is now available to everyone with an NVIDIA Developer account. NVIDIA Nsight Systems 2026.1 includes: PyTorch profiling improvements to…
Automating and Optimizing Financial Signal Discovery with Multi-Agent Systems | NVIDIA Technical Blog
…In the NeMo Agent Toolkit ecosystem, this agent is specifically tuned for tool-calling and code generation. It takes the blueprint from the signal agent and produces Python code that calculates the…