NVIDIA Unveils RTX Spark Superchip With 128GB Unified Memory to Challenge Apple Silicon

NVIDIA and Microsoft are teaming up to challenge Apple Silicon with RTX Spark, a new Windows PC platform designed to natively power local artificial intelligence agents.



The RTX Spark superchip is an Arm-based design created in collaboration with MediaTek. It combines a 20-core NVIDIA Grace CPU with a Blackwell RTX GPU featuring 6,144 CUDA cores and fifth-generation Tensor Cores. The processing components are linked by a high-speed NVLink-C2C interconnect. MediaTek contributed to the custom CPU design, helping optimize power efficiency, performance, and connectivity. NVIDIA says the platform delivers up to one petaflop of AI compute and supports up to 128GB of unified memory.


The architecture bears similarities to Apple's silicon strategy, pairing Arm CPU cores with a powerful GPU and a large pool of unified memory accessible to both processors. By scaling up to 128GB of unified memory, NVIDIA is targeting the same class of AI and creative workloads served by Apple's high-end hardware, including the MacBook Pro with M5 Max. NVIDIA says RTX Spark systems can run AI models with up to 120 billion parameters locally while supporting context windows of up to one million tokens.



On the software side, NVIDIA and Microsoft are introducing new Windows security primitives designed to help AI agents operate safely on users' primary PCs. NVIDIA's OpenShell runtime adds policy controls that let users determine what agents can access, route requests to local models based on privacy preferences, and mask personal information before queries are sent to cloud services. Open-source projects including Hermes Agent and OpenClaw have already announced support for the platform.

Apple users have seen similar efforts to improve local AI performance on consumer hardware, with applications such as Ollama taking advantage of Apple's MLX framework for faster local AI processing on Mac. RTX Spark represents NVIDIA's attempt to bring comparable local AI capabilities to Windows devices while pairing them with the company's CUDA and RTX software ecosystem.

NVIDIA claims RTX Spark can render 90GB 3D scenes, edit 12K 4:2:2 video, generate 4K AI video, and play AAA games at 1440p resolution at more than 100 frames per second using technologies such as DLSS and Reflex. The company also highlighted support for new features including DLSS 4.5 Ray Reconstruction.


Creative professionals are a major focus for the platform. NVIDIA says Adobe is rearchitecting Photoshop and Premiere from the ground up for RTX Spark, with the goal of delivering up to twice the AI and graphics performance across creative workflows. Adobe's Substance 3D Painter and Stager will also run natively on the platform.



NVIDIA is also bringing the architecture to enterprise developers through DGX Station for Windows, a Blackwell-powered system designed for developers building and deploying AI agents.

RTX Spark will power a new generation of slim Windows laptops and compact desktop PCs arriving this fall from partners including Microsoft Surface, Dell, HP, Lenovo, ASUS, and MSI. NVIDIA says laptops will be available in 14- to 16-inch sizes, with some designs measuring as little as 14mm thick and weighing about three pounds.