Nvidia has unveiled the RTX Spark superchip, a new 'superchip' designed for Windows laptops and compact desktops, featuring 128GB of unified memory and 1 petaflop of AI compute. The chip will be available in devices from Asus, Dell, HP, Lenovo, Microsoft Surface, and MSI this fall.
TL;DR
- Nvidia's RTX Spark superchip combines a 20-core Grace CPU and a Blackwell GPU with 6,144 CUDA cores, targeting Windows laptops and compact desktops.
- The chip's 128GB unified memory and 1 petaflop AI compute capability position it for local AI workloads, competing with Qualcomm, AMD, and Intel.
- RTX Spark will be available in laptops from major manufacturers this fall, marking Nvidia's direct entry into the Windows PC chip market.
What happened
Nvidia introduced the RTX Spark superchip at Computex 2026, a new class of chip designed to power Windows laptops and compact desktops directly. The chip pairs a 20-core Grace CPU with a Blackwell GPU, featuring 6,144 CUDA cores and up to 128GB of unified memory, delivering 1 petaflop of FP4 AI performance. Nvidia CEO Jensen Huang unveiled the chip, highlighting its role in enabling personal AI agents to run locally on Windows PCs. Six laptop makers—Asus, Dell, HP, Lenovo, Microsoft Surface, and MSI—are lined up to ship RTX Spark machines this fall, with Acer and Gigabyte expected to follow. The chip is part of a family that includes the DGX Spark desktop and the DGX Station for Windows workstation, all built around the Grace-plus-Blackwell architecture.
Why it matters
RTX Spark's 128GB unified memory and 1 petaflop AI compute capability make it a strong contender in the AI-PC market, enabling local AI workloads that previously required cloud-based solutions. This positions Nvidia directly against Qualcomm, AMD, and Intel, who have already established a presence in the Windows AI-PC market with their respective chips. The chip's design, which fuses a 20-core Grace CPU and a Blackwell GPU on a single die, offers a significant advantage in terms of memory bandwidth and power efficiency. However, the actual performance of RTX Spark in running large language models locally remains to be seen, as no independent benchmarks have been published yet. The chip's launch this fall will be a critical test of Nvidia's ability to compete in the Windows PC chip market.
Key facts
- RTX Spark features a 20-core Grace CPU and a Blackwell GPU with 6,144 CUDA cores.
- The chip supports up to 128GB of LPDDR5X unified memory with roughly 300GB/s of bandwidth.
- Nvidia rates RTX Spark at 1 petaflop of FP4 AI performance.
- The chip will be available in laptops from Asus, Dell, HP, Lenovo, Microsoft Surface, and MSI this fall.
- RTX Spark is built on TSMC's 3-nanometer process and packs 70 billion transistors.
- The chip's power envelope ranges from 45W to 80W, with transient boosts up to roughly 245W.
- RTX Spark is positioned as infrastructure for AI that runs locally instead of in the cloud.
- The chip is part of a family that includes the DGX Spark desktop and the DGX Station for Windows workstation.
Context
Nvidia's entry into the Windows PC chip market with RTX Spark marks a significant shift in the company's strategy, moving from selling graphics chips to other companies to offering a complete Windows PC platform. This shift is driven by the growing demand for AI capabilities in consumer devices, which Nvidia is well-positioned to meet with its expertise in AI and GPU technology. The launch of RTX Spark also highlights the increasing importance of local AI workloads, as consumers seek more privacy and control over their data. As the AI-PC market continues to evolve, Nvidia's RTX Spark superchip is poised to play a crucial role in shaping the future of personal AI agents and local AI workloads.