Adobe Rebuilds Photoshop and Premiere Pro for Nvidia's RTX Spark Superchip

A New Video Pipeline for Premiere Pro, No CPU Bottlenecks

Video editors often wrestle with lag when working on complex, multi-effect timelines. For Premiere Pro, Adobe has designed a completely new video pipeline built on RTX Spark's unified memory architecture, Blackwell GPU, and TensorRT software. This redesign allows for real-time editing and color grading, faster rendering of complex timelines, and GPU-accelerated AI inference without the bottleneck of moving data to and from a separate processor. The promise is a smoother experience when working with features like Firefly-powered Generative Extend .

Substance 3D Taps into Vast Unified Memory

The Substance 3D suite—including Stager, Modeler, and Sampler—is also being optimized for the RTX Spark's unified memory. The superchip is designed to handle 3D scenes exceeding 90 GB, a capacity well beyond what traditional VRAM-limited graphics cards can manage. This unlocks the potential to work with massive textures and highly detailed 3D assets without the constant performance swaps that can plague current workflows .

Agentic AI Comes to Creative Workflows

Beyond raw performance, the rebuilt apps will introduce agentic AI capabilities. Adobe is adding support for the Model Context Protocol (MCP) , allowing users to connect local AI agents directly to Photoshop and Premiere Pro. This enables the automation of multi-step, tedious creative processes entirely on the user's device, from batch processing specific effects to generating complex scene variations .

This is possible because the RTX Spark itself is designed from the ground up for personal AI agents. Nvidia claims the chip can run 120-billion-parameter language models with up to 1 million tokens of context locally, supported by the full Nvidia AI software stack and Microsoft's Windows Agent Framework . The idea is that a powerful local agent can understand and execute complex creative instructions without needing a constant cloud connection.

Under the Hood of the RTX Spark Superchip

The RTX Spark superchip (formerly known by the codenames N1/N1X) is Nvidia's bid to reinvent the Windows PC. Built in collaboration with MediaTek, the system-on-a-chip combines a 20-core Nvidia Grace Arm CPU with a Blackwell RTX GPU featuring 6,144 CUDA cores and fifth-generation Tensor Cores. The chip offers up to 128GB of unified LPDDR4X memory (though some sources note LPDDR5X configurations ) with 300 GB/s of bandwidth and delivers 1 petaflop of FP4 AI performance—roughly comparable to an RTX 5070-class desktop GPU in a laptop form factor .

Who Is Building Them and When You Can Get One

Nvidia is not making the laptops itself. Instead, a wide ecosystem of hardware partners will bring RTX Spark systems to market starting in fall 2026. The list of partners includes Acer, ASUS, Dell, HP, Lenovo, MSI, Gigabyte, and Samsung, with systems ranging from premium laptops as thin as 14 millimeters to small-form-factor desktop PCs .