Nvidia launches Underdog app for local AI models on RTX Spark

Nvidia released Underdog, an app that runs local AI models on RTX Spark laptops using the Husky inference engine with no cloud connection. Speculative decoding with MTP speeds up agentic tasks by about 1.4x on RTX Pro 6000 Blackwell via llama.cpp.

• Underdog runs on the Husky engine and keeps all data on the laptop
• Up to 1.4x speedup on RTX Pro 6000 Blackwell via llama.cpp and MTP
• Day-zero support for Nemotron Bolt E4B, GLM, DeepSeek, Qwen, MiniMax and FLUX 3 Klein
• RTX Spark Arm Windows laptops launched Oct. 16 from ASUS, Dell, HP, Lenovo, Microsoft and MSI