NVIDIA has introduced RTX Spark, a new AI computing platform for Windows laptops and compact desktop PCs that combines a Grace CPU and Blackwell GPU on a single chip.
The company says the platform can deliver up to 1 petaflop of FP4 AI performance and support up to 128GB of unified memory, allowing larger AI models to run locally instead of relying on cloud services. If those claims hold up in independent testing, RTX Spark could become one of the most powerful AI platforms available for Windows PCs.
The flagship RTX Spark configuration features a 20-core Grace CPU, a Blackwell GPU with 6,144 CUDA cores, and up to 128GB of LPDDR5X unified memory. NVIDIA says this combination is designed for AI developers, researchers, content creators, and professionals working with increasingly large AI workloads.
Most Windows laptops keep system RAM and graphics memory separate. RTX Spark takes a different approach by letting the CPU and GPU share the same memory pool. NVIDIA says this reduces data transfers between the two processors, helping AI models and other demanding workloads run more efficiently on the device.
NVIDIA also says the flagship version can run AI models with up to 120 billion parameters on a local machine. That claim has not yet been verified by independent reviewers, and real-world performance will depend on software optimization and individual workloads.
RTX Spark is manufactured using TSMC’s 3nm process and uses NVIDIA‘s NVLink-C2C technology to connect the CPU and GPU. According to the company, the chip contains approximately 70 billion transistors.
A petaflop represents one quadrillion floating-point operations per second and is commonly used to measure AI computing performance. NVIDIA’s quoted FP4 performance figure is intended for AI inference workloads rather than traditional graphics or gaming performance.
Multiple Configurations Planned
NVIDIA plans to release RTX Spark in several configurations aimed at different price segments and performance levels.
| Configuration | CPU | CUDA Cores | Unified Memory |
|---|---|---|---|
| RTX Spark | 20-core Grace | 6,144 | Up to 128GB |
| Tier 2 | 18-core | 5,120 | 64GB |
| Tier 3 | 12-core | 2,560 | 64GB |
| Tier 4 | 10-core | 2,048 | 64GB |
These versions will target premium AI laptops, mainstream AI PCs, and entry-level systems, giving manufacturers more flexibility across different product categories.
RTX Spark systems will be available from ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI, while Acer and Gigabyte are also expected to introduce devices later. NVIDIA says the first products powered by RTX Spark will launch in fall 2026.
NVIDIA has not announced pricing. RTX Spark will compete with AI PC platforms from Qualcomm, AMD, and Intel. While those companies have largely focused on improving NPU performance for AI features, NVIDIA is taking a different approach by pairing its Blackwell GPU with a large unified memory pool to support larger AI models running directly on local hardware.
The real test for RTX Spark will come once pricing is announced and independent benchmarks become available. With the first systems expected in fall 2026, reviewers will finally be able to measure NVIDIA’s claimed performance in everyday workloads.
Source: NVIDIA
