Nvidia RTX Spark targets PC market with multi-generation product roadmap, demonstrating sustained commitment unlike prior Arm initiatives.
NVIDIA announced RTX Spark at GTC Taipei 2026, unveiled by CEO Jensen Huang as a brand-new PC platform designed in collaboration with Microsoft. Building on the DGX Spark project initiated three years ago with Microsoft, Arm, MediaTek, and other ecosystem partners, the RTX Spark represents NVIDIA's new chapter in the client platform segment. The RTX Spark builds upon the DGX Spark's GB10 Superchip—internally referred to as N1X—with a lighter configuration called N1, designed to bring AI-first computing to accessible form factors.
The RTX Spark features the GB10 Superchip, composed of two dielets: an S-Dielet containing the CPU and memory subsystem, and a G-Dielet containing the GPU core, connected via Advanced 2.5D packaging using TSMC's 3nm process technology. The CPU is based on ARM Architecture v9.2 with 20 cores custom-designed by MediaTek in two clusters of 10 cores each, with a private L2 cache per core and 16 MB L3 cache per cluster (32 MB total). The GPU is based on GB100 Blackwell architecture with 5th Gen Tensor Cores supporting DLSS, Reflex, ACE, G-Sync, CUDA, and NVFP4, delivering up to 31 TFLOPs of FP32 and 1000 TOPS of NVFP4 compute for AI workloads, plus 24 MB of L2 cache.
The memory subsystem supports 256b LPDDR5x (UMA) with speeds up to 9400 MT/s, enabling 301 GB/s raw bandwidth and up to 128 GB maximum capacity, with a high-performance coherent fabric supporting CHI-E Coherency Protocol. The GPU can access 600 GB/s aggregate bandwidth over the C2X interface, and users can dedicate up to 111 GB of a 128 GB system to the GPU directly through Windows without entering BIOS. The system includes 16 MB of System Level Cache serving as L4 for the CPU and 140W TDP. Connectivity features include PCIe, USB, Ethernet over PCIe, and support for four concurrent displays (3 DP + 1 HDMI) at up to 4K @ 120Hz with DP Alt-mode and up to 8K @ 120Hz with HDMI 2.1a.
Unlike Qualcomm's first attempt with Snapdragon X1 and X2 series, NVIDIA possesses an optimized Windows framework built on its established RTX platform, which has powered desktops and laptops with solid driver support and substantially improved Linux support. TensorRT and CUDA support remain strong on Windows, positioning NVIDIA better in the Windows-on-ARM segment. A key challenge remains the untested Grace CPU as a Windows chip; previous benchmarks were DGX-based through emulation or on Linux.
During NVIDIA's technical sessions, games and applications including Blender, Unreal Engine, LLM Studio, and OpenClaw ran at reasonably fast performance. Closed testing showed the chip achieving up to 10x faster AI performance and up to 2x faster GenAI performance, with gaming titles such as Alan Wake II, Fortnite, and Indiana Jones running well with DLSS Frame-Gen enabled. However, NVIDIA has not provided FPS figures or specified the settings used for these game tests.