Open App →
Back to News Feed
Nvidia Developer Blog November 17, 2026 positive

Co-Designing Sparse Attention for Efficient Long-Context Inference

AI / LLM
<img width="768" height="432" src="https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-768x432.png" class="webfeedsFeaturedVisual wp-post-image" alt="" style="display: block; margin-bottom: 5px; clear:both;max-width: 100%;" link_thumbnail="" decoding="async" srcset="https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-768x432.png 768w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-179x101.png 179w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-300x169.png 300w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-625x352.png 625w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-1536x864.png 1536w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-645x363.png 645w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-660x370.png 660w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-500x281.png 500w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-160x90.png 160w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-362x204.png 362w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-196x110.png 196w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-1024x576.png 1024w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-960x540.png 960w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy.webp 1877w" sizes="(max-width: 768px) 100vw, 768px" title="llm-optimize-deploy" />Agentic workloads, long documents, and large codebases are pushing context lengths from tens of thousands of tokens to hundreds of thousands. At those lengths,...<img width="768" height="432" src="https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-768x432.png" class="webfeedsFeaturedVisual wp-post-image" alt="" style="display: block; margin-bottom: 5px; clear:both;max-width: 100%;" link_thumbnail="" decoding="async" loading="lazy" srcset="https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-768x432.png 768w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-179x101.png 179w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-300x169.png 300w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-625x352.png 625w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-1536x864.png 1536w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-645x363.png 645w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-660x370.png 660w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-500x281.png 500w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-160x90.png 160w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-362x204.png 362w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-196x110.png 196w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-1024x576.png 1024w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy-960x540.png 960w, https://developer-blogs.nvidia.com/wp-content/uploads/2026/11/llm-optimize-deploy.webp 1877w" sizes="auto, (max-width: 768px) 100vw, 768px" title="llm-optimize-deploy" /><p>Agentic workloads, long documents, and large codebases are pushing context lengths from tens of thousands of tokens to hundreds of thousands. At those lengths, attention becomes the most expensive part of a transformer layer. Dense attention scores every query against every key, so its cost per query grows with the context length N, and at 128K tokens it can take most of the prefill time in a…</p> <p><a href="https://developer.nvidia.com/blog/co-designing-sparse-attention-for-efficient-long-context-inference/" rel="nofollow" data-wpel-link="internal" target="_self">Source</a></p>
Read original article ↗

Related Articles

Pegatron Showcases High-Density Servers and NVMe Storage for AI and HPC at SC26

CHICAGO, Nov. 17, 2026 — PEGATRON, a globally recognized provider of comprehensive AI infrastructure and server solution

HPC Wire · November 18, 2026

PCSS and PIONIER Showcase Polish AI and Quantum Projects at SC26

Nov. 17, 2026 — The Poznań Supercomputing and Networking Center (PCSS), together with the PIONIER Consortium, is partici

HPC Wire · November 18, 2026

Linux Foundation Announces the Open Financial Trust Alliance to Strengthen Coordinated Protection Against Fraud, Scams and Financial Crime

Global, open, cross-sector alliance connects existing fraud prevention frameworks to enable interoperability Summary On

PR Newswire · November 18, 2026