Simon is a Computer Science BSc graduate who has been writing about technology since 2014, and using Windows machines since 3.1. After working for an indie game studio and acting as the family's go-to technician for all computer issues, he found his passion for writing and decided to use his skill set to write about all things tech.
Since beginning his writing career, he has written for many different publications such as WorldStart, Listverse, and MakeTechEasier. However, after finding his home at MakeUseOf in February 2019, he would eventually move on to its sister site, XDA, to bring the latest and greatest in Windows, Linux, and DIY electronics.
Sign in to your XDA account
Summary
- RTX Spark Arm Windows laptops launched on October 16th across major OEMs and offer powerful local AI processing.
- The Underdog app runs local models with the Husky inference engine, with no cloud required; the data stays on your laptop.
- Speculative decoding (MTP) tasks speed up by ~1.4x on RTX Pro 6000 Blackwell via llama.cpp, and some models get day-zero support.
Back in mid-October, Nvidia's RTX Spark devices hit the shelves. These powerful Arm-based System-on-Chip devices run Windows and are designed to be processing powerhouses, especially for local LLM models.
Now, Nvidia has revealed what to expect from these shiny new RTX Spark devices, including a new app called Underdog.
Running models on an RTX Spark gets even easier with Underdog
It's a match made in heaven
In a recent post on Nvidia's blog, the company introduced its new range of RTX Spark-powered devices, including laptops from Asus, Dell, HP, Lenovo, Microsoft, and MSI, with Acer in the works.
A key reason someone might want to drop money on a new RTX Spark is its onboard AI capabilities. To support that, Nvidia announced its new Underdog app, which is ready to download now. This app uses a Husky model-specific inference engine to run local AI models on your system, and because it never needs to speak to the cloud, all of your data stays safe on your laptop.
There's also some good news for people who love llama.cpp. Nvidia announced "new speculative decoding support with MTP for Qwen 3.8 Flash Next," which speeds up end-to-end agentic tasks by 1.4x on an RTX Pro 6000 Blackwell.
The company also took the time to announce that these models are getting day-zero support: Nemotron Bolt E4B 4B, GLM,
DeepSeek, Qwen, MiniMax, and FLUX 3 Klein diffusion model. This means they should work with your setup on release day, so you don't need to wait for Nvidia to publish support.