NVIDIA puts a personal AI supercomputer in a 14mm laptop: RTX Spark ships, with GPU sharing between agents and apps
Synopsis
NVIDIA's RTX Spark laptops are now on sale from ASUS, Dell, HP, Lenovo, Microsoft and MSI starting at $2,599, pairing a Blackwell RTX GPU with a Grace CPU on one superchip with up to 128GB of unified memory and up to 1 petaflop of FP4 AI performance. The accompanying updates include Nemotron 3.5 Transcribe, an open weight speech model that handles transcription, speaker labeling and translation in one promptable model across 40 languages, ranking No. 1 on Hugging Face Open ASR with about 4.3 billion active parameters. A new NVIDIA App resource control lets users split GPU resources between foreground apps and background agents.
Interpretation
Thin and light Windows laptops begin to carry work that previously called for a desktop workstation, with AI models and creative projects drawing on the same memory pool even when unplugged. Earlier thin machines separated CPU, GPU and memory, making it hard to hold both large models and large creative projects locally. RTX Spark places the GPU, CPU and up to 128GB of unified memory on one power-efficient superchip, in bodies as thin as 14mm and as light as three pounds, and NVIDIA states it can edit camera-native 4:2:2 video and render 3D scenes of up to 90GB.
Speech transcription moves from stitching several models together to a single model guided by plain-language prompts that transcribes, labels speakers, translates and answers questions about the audio. Transcription, speaker attribution and translation previously each required a separate model. Nemotron 3.5 Transcribe ranks No. 1 on Hugging Face Open ASR and No. 2 among open weight models on Artificial Analysis with about 4.3 billion active parameters, 80% fewer than the No. 1 open weight model, and transcribes 40 languages including seven Indic languages.
Long-running agents and a user's foreground application can divide the same GPU compute according to what matters at that moment. The two previously shared GPU compute without the user directly choosing the balance. NVIDIA App offers resource management under System > Advanced, letting users prioritize the foreground app for best performance or give a background agent more resources to finish faster, and requires hardware-accelerated GPU scheduling enabled in Windows Settings.
Perspective
The report covers the RTX Spark configurations, software updates and leaderboard rankings NVIDIA announced, for developers and creators considering running models and agents locally; it contains no third-party independent measurements of whole-machine performance.
The Nemotron 3.5 Transcribe rankings are leaderboard results on Hugging Face Open ASR and Artificial Analysis cited by the vendor, and quantized versions are described as coming soon, so local execution depends on those versions arriving.