Nvidia RTX Spark laptops arrive Friday with 74-billion-parameter AI on 24GB
What happened
NVIDIA RTX Spark laptops go on sale Friday with a local AI model called Nemotron 3.5 Bolt 74B. It fits in 24GB of laptop memory by activating just 9 billion of its 74 billion parameters during each response. The platform runs on ARM architecture with Windows on Arm and includes PAIR, a free home-network router that auto-discovers compatible GPUs.
Why it matters
A frontier-class AI model that runs on a laptop instead of a server farm changes who can use powerful AI. Small business owners and freelancers could run private AI tools without sending data to the cloud. The partial activation trick means you get large-model quality at a size your computer can handle.
Who's doing it
NVIDIA built the platform in partnership with MediaTek. Two hardware configurations are planned: a 20-core CPU with 6,144 CUDA cores for laptops and compact desktops, and an 18-core with 5,120 CUDA cores for laptops only. NVIDIA rates the platform at up to 1 petaFLOP FP4 of AI compute.
Try it
- Wait for RTX Spark laptops to go on sale Friday and check NVIDIA's website or major retailers for the two listed configurations.
- If you buy one, set up the included PAIR router on your home network so it can find any compatible GPUs you already have.
- Download the Nemotron 3.5 Bolt 74B model through the platform's software and run a test prompt to confirm it loads within the 24GB memory limit.
Read the original at techtimes.com
Comments
The panel is AI Daylee's cast of fictional characters, written by AI. They react to what's on this page and haven't used anything themselves. Reader comments aren't open yet.
The AI Daylee Digest, every Monday
A week of prompts to try, each tested in ChatGPT, Claude and Gemini, plus the AI news worth your time. Free, one email a week.
The Boss hype translator
This is a paradigm-shifting democratization of edge AI compute. Translation: Nvidia wants you to buy a whole laptop because the server farm was too expensive to rent you.
Karen what's the catch
So this PAIR router auto-discovers my GPUs? GREAT. Who approved it scanning my home network and what data does Nvidia collect when I download that 74B model?
The Professor fact check
The article conflates parameters with quality. Activating 9 billion of 74 billion parameters means most of the model is dormant per token, which is not the same as large-model performance.
The Yinzer
Yinz think a laptop running 9 billion parameters at a time is frontier-class? That's like saying Primanti's is fine dining 'cause they got a big menu, n'at.