{"slug": "nvidia-s-tensorrt-llm-cuts-deepseek-v4-tuning-time-70-in-test", "title": "NVIDIA's TensorRT-LLM cuts DeepSeek V4 tuning time 70% in test", "summary": "NVIDIA merged pull request 21185 into TensorRT-LLM main on November 16th, 2026, adding an optional nvMatmulHeuristics policy that cut DeepSeek V4 Pro cold autotuning from 226.6 seconds to 68.2 seconds in contributor-reported tests. Across 48 FP8 BMM and MXFP8 GEMM shapes, the heuristic path profiled 838 tactics instead of 82,160 with the full sweep, a roughly 98x reduction, and server start-to-ready time fell from 514 seconds to 350 seconds on two nodes with four VR200 GPUs each. The feature stays off by default.", "body_md": "# NVIDIA's TensorRT-LLM cuts DeepSeek V4 tuning time 70% in test\n\n**The merged TensorRT-LLM policy profiled 838 tactics instead of 82,160 on DeepSeek V4 Pro, and the feature stays off by default.**\n\n        By [RuntimeWire Staff](https://runtimewire.com/author/runtimewire-staff)\n        · Published \n\nNVIDIA merged pull request 21185 into TensorRT-LLM main on November 16th, 2026, adding an optional nvMatmulHeuristics policy that cut DeepSeek V4 Pro cold autotuning from 226.6 seconds to 68.2 seconds in contributor-reported tests. Across 48 FP8 BMM and MXFP8 GEMM shapes, the heuristic path profiled 838 tactics, down from 82,160 with the full sweep, a reduction of about 98 times. Server start-to-ready time fell from 514 seconds to 350 seconds on two nodes with four VR200 GPUs each.…", "url": "https://wpnews.pro/news/nvidia-s-tensorrt-llm-cuts-deepseek-v4-tuning-time-70-in-test", "canonical_source": "https://runtimewire.com/article/tensorrt-llm-deepseek-v4-autotuning-nvmatmulheuristics", "published_at": "2026-11-16 14:36:36+00:00", "updated_at": "2026-11-16 14:52:13.031076+00:00", "lang": "en", "topics": ["ai-infrastructure", "large-language-models", "machine-learning", "ai-tools"], "entities": ["NVIDIA", "TensorRT-LLM", "DeepSeek V4 Pro", "nvMatmulHeuristics", "VR200"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/nvidia-s-tensorrt-llm-cuts-deepseek-v4-tuning-time-70-in-test", "markdown": "https://wpnews.pro/news/nvidia-s-tensorrt-llm-cuts-deepseek-v4-tuning-time-70-in-test.md", "text": "https://wpnews.pro/news/nvidia-s-tensorrt-llm-cuts-deepseek-v4-tuning-time-70-in-test.txt", "jsonld": "https://wpnews.pro/news/nvidia-s-tensorrt-llm-cuts-deepseek-v4-tuning-time-70-in-test.jsonld"}}