--- title: Using NVFP4 Low-Precision Model Training for Higher... description: NVFP4, MXFP8, and FP8-CS low-precision training formats are benchmarked against BF16 across multi-hundred-billion token pretraining runs on Llama 3 8B and an... canonical: https://daily.dev/posts/using-nvfp4-low-precision-model-training-for-higher-throughput-without-losing-accuracy-mburtmoay twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Using NVFP4 Low-Precision Model Training for Higher Throughput Without Losing Accuracy | daily.dev og:description: NVFP4, MXFP8, and FP8-CS low-precision training formats are benchmarked against BF16 across multi-hundred-billion token pretraining runs on Llama 3 8B and an... og:url: https://daily.dev/posts/using-nvfp4-low-precision-model-training-for-higher-throughput-without-losing-accuracy-mburtmoay og:image: https://api.daily.dev/og/posts/mburTmoay.png og:image:alt: Using NVFP4 Low-Precision Model Training for Higher Throughput Without Losing Accuracy og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Using NVFP4 Low-Precision Model Training for Higher Throughput Without Losing Accuracy **[NVIDIA Developer](https://daily.dev/sources/nvidiadev)** · 7 min read · 1 upvotes · 0 comments ## Summary NVFP4, MXFP8, and FP8-CS low-precision training formats are benchmarked against BF16 across multi-hundred-billion token pretraining runs on Llama 3 8B and an internal 8B model on NVIDIA B200/GB200 GPUs. Results show NVFP4 achieves up to 1.59x throughput over BF16 while preserving downstream accuracy on MMLU, HellaSwag, WinoGrande, and ARC-C benchmarks. Key findings include that keeping the final four transformer layers in BF16 is essential for NVFP4 stability, and MXFP8 slightly outperforms standard FP8 due to finer-grained block scaling. The NVIDIA NeMo Megatron Bridge library enables switching between precision formats via a single config flag, with production-ready recipes available for immediate use. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Similar posts on daily.dev - [Train Models Faster with JAX and MaxText Using NVFP4 on NVIDIA Blackwell](https://daily.dev/posts/train-models-faster-with-jax-and-maxtext-using-nvfp4-on-nvidia-blackwell-ouytd5oyn) · NVIDIA Developer · 0 upvotes · 0 comments - [Creating the NVIDIA Nemotron 3 Ultra NVFP4 Checkpoint with NVIDIA Model Optimizer](https://daily.dev/posts/creating-the-nvidia-nemotron-3-ultra-nvfp4-checkpoint-with-nvidia-model-optimizer-o6e6ls0pg) · NVIDIA Developer · 1 upvotes · 0 comments - [MXFP8 Training for MoEs: 1.3x training speedup vs BF16 for Llama4 Scout on GB200 cluster using TorchAO and TorchTitan – PyTorch](https://daily.dev/posts/mxfp8-training-for-moes-1-3x-training-speedup-vs-bf16-for-llama4-scout-on-gb200-cluster-using-torch-agasmbqcx) · PyTorch · 0 upvotes · 0 comments --- Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#llm](https://daily.dev/tags/llm), [#nvidia-nemo](https://daily.dev/tags/nvidia-nemo) [View this post on daily.dev](https://daily.dev/posts/using-nvfp4-low-precision-model-training-for-higher-throughput-without-losing-accuracy-mburtmoay) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Using NVFP4 Low-Precision Model Training for Higher Throughput Without Losing Accuracy","url":"https://daily.dev/posts/using-nvfp4-low-precision-model-training-for-higher-throughput-without-losing-accuracy-mburtmoay","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/using-nvfp4-low-precision-model-training-for-higher-throughput-without-losing-accuracy-mburtmoay"},"datePublished":"2026-02-23T18:04:24.222Z","dateModified":"2026-02-23T18:04:51.135Z","description":"NVFP4, MXFP8, and FP8-CS low-precision training formats are benchmarked against BF16 across multi-hundred-billion token pretraining runs on Llama 3 8B and an...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/9b4d201436a4bdfb98b97d11fe46225e?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/9b4d201436a4bdfb98b97d11fe46225e?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"NVIDIA Developer","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"NVIDIA Developer","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/86e45aab42ba48ce83103d01b1119910","url":"https://daily.dev/sources/nvidiadev"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/using-nvfp4-low-precision-model-training-for-higher-throughput-without-losing-accuracy-mburtmoay","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"machine-learning,llm,nvidia-nemo","timeRequired":"PT7M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"NVIDIA Developer","item":"https://daily.dev/sources/nvidiadev"},{"@type":"ListItem","position":3,"name":"Using NVFP4 Low-Precision Model Training for Higher Throughput Without Losing Accuracy"}]} ```