Nvidia adds controls to balance AI agents and everyday apps on RTX Spark laptops

Nvidia is giving RTX Spark laptop users a choice over which work gets priority: the app in front of them or an AI agent running in the background. In its October 28 update, the company detailed new GPU-sharing controls alongside local-model support for Windows laptops sold by six manufacturers, starting at $2,599.

Choose which work gets the GPU

On RTX Spark, long-running agents and ordinary applications share the graphics processor’s computing resources. Nvidia’s new setting lets users favor their foreground application or allocate more resources to a background agent. The choice is between keeping the active app performing well and giving the agent more capacity to finish its work.

The control sits under System > Advanced in the NVIDIA App. It requires hardware-accelerated GPU scheduling to be enabled in Windows Settings. That prerequisite is worth checking before looking for the resource-sharing option: installing the app alone is not the full setup Nvidia specifies.

Nvidia describes both options as useful at different moments, depending on which task matters most to the user. The setting concerns how applications and agents divide GPU compute on one laptop; it is a resource-allocation control, not a new model or a separate agent application.

$2,599Starting laptop price

Nvidia lists RTX Spark laptops from ASUS, Dell, HP, Lenovo, Microsoft and MSI at a starting price of $2,599.

128GBMaximum unified memory

Nvidia specifies up to 128GB of unified memory and says configurations with that capacity are on shelves.

One memory pool for models and creative work

RTX Spark combines a Blackwell RTX graphics processor and a Grace CPU in a single superchip. Nvidia rates the platform at up to one petaflop of FP4 AI performance. Those are the company’s stated hardware ceilings, alongside its maximum memory capacity, rather than specifications promised for every laptop configuration.

Creative projects and AI models draw from the same memory pool. Nvidia says the systems can edit camera-native 4:2:2 video and render 3D scenes of up to 90GB. Adobe Photoshop and Premiere have native RTX Spark support through beta releases, an important qualification for buyers considering the laptops for those applications.

The lineup includes different screen sizes, touchscreens, pen support and two-in-one convertible designs. Nvidia lists models as thin as 14mm and as light as three pounds. The hardware pitch spans development, creative work and gaming, rather than treating the machines solely as dedicated computers for running AI agents.

More local-model options

The software update also highlights support for Liquid AI’s LFM3 family. Nvidia says developers can run the models locally through llama.cpp and vLLM, software used to run AI models, and fine-tune them on RTX Spark. Fine-tuning lets developers adapt a model rather than only use it in its supplied form.

LFM3 spans text, vision, audio, omni and decision models built on a shared backbone, with base and post-trained versions available as open weights. Nvidia describes local workflows across several kinds of input, not just text chat. Its examples include:

• Agents in Hermes Agent, Pi and OpenClaw that triage inboxes and manage calendars.
• Reading documents and application screens, plus transcribing and translating speech.
• Answering spoken questions about handwritten math and routing support tickets or voice requests.