NVIDIA announced updates to its RTX Spark ecosystem, bringing personal AI supercomputing capabilities powered by Blackwell RTX GPUs and Grace CPUs to Windows laptops from manufacturers including ASUS, Dell, HP, Lenovo, Microsoft, and MSI. The systems deliver up to 128GB unified memory and 1 petaflop of FP4 AI performance for local execution.
Alongside hardware updates, NVIDIA introduced Nemotron 3.5 Transcribe, a 4.3-billion parameter open-weight speech model. The promptable single model handles transcription, speaker diarization, translation across 40 languages, and audio Q&A simultaneously. Available on Hugging Face, quantized versions will soon support local processing on RTX Spark and Jetson devices or cloud deployment via Blackwell GPUs.
NVIDIA also introduced a new resource management control within the NVIDIA App for RTX Spark systems. This feature allows users to dynamically allocate system GPU compute between foreground applications and background-running AI agents.
Why it matters
Nemotron 3.5 Transcribe consolidates speech processing, speaker labeling, and translation into one open 4.3B parameter model.
New GPU resource management features enable continuous background AI agent execution alongside primary desktop applications.
Unified memory laptops allow developers to run compact open-weight models locally without relying on cloud infrastructure.
Source: blogs.nvidia.com



