NVIDIA AI Developer@NVIDIAAIDev🎉 NVIDIA has achieved the world’s fastest DeepSeek-R1 inference. 👀 A single system with eight Blackwell GPUs in an NVL8 configuration delivers 253 TPS/user or 30K TPS system throughput on the full DeepSeek-R1 671B parameter model. #GTC25 ➡️ t.coOpens with an observation
NVIDIA AI Developer@NVIDIAAIDev📣 Introducing NVIDIA Dynamo, a high-throughput, low-latency #opensource inference library for deploying AI reasoning models in large-scale distributed environments. Boost the number of requests served by up to 30x with DeepSeek-R1 on NVIDIA Blackwell ➡️ t.coOpens with an observation
NVIDIA AI Developer@NVIDIAAIDevNew open #PhysicalAI dataset just unveiled at #GTC25. 👏 This commercial-grade, pre-validated dataset can help researchers and developers kick-start physical AI projects. The dataset is openly available on @huggingface. Learn more ➡️ t.coOpens with a number
NVIDIA AI Developer@NVIDIAAIDevNew #NVIDIACosmos world foundation models, announced at #GTC25, address the challenges and constraints faced in #PhysicalAI development. 🤗 The new models are now available on @huggingface: t.co 🛠️ Learn how you can use them: t.coOpens with a number