Nvidia Unveils RTX Spark: Blackwell-Based Superchip with 128GB Unified Memory

Nvidia has introduced RTX Spark, a new superchip platform combining a Blackwell-generation RTX GPU with what the company calls its most power-efficient CPU design to date. The headline specs include FP4 Tensor Cores, RT Cores, and up to 128 GB of unified memory in a thin chassis, with CUDA running natively to maintain compatibility with the broader AI software stack.

The positioning is deliberately broad: local AI prototyping and fine-tuning for developers, FP4 inference and 4:2:2 hardware encode/decode for creative workflows, and ray tracing plus the full DLSS suite for gaming. The unified memory pool is the most consequential detail for developers, since it puts model sizes that previously required workstation or datacenter hardware within reach of a portable machine.

Laptops and desktops built on the platform are not yet shipping — Nvidia is currently collecting signups for availability notifications, with no pricing or release date attached.