Overview
- Engineering-sample N1x recorded a 46,361 score in Geekbench OpenCL, outpacing current integrated GPUs from Intel, AMD and Qualcomm in early tests.
- The N1x’s 48 Compute Units deliver 6,144 CUDA cores—identical to the desktop RTX 5070—but its 1.048 GHz cap and 120 W envelope suggest lower raw throughput.
- The chip forgoes dedicated GDDR7 VRAM in favor of onboard LPDDR5X, trading memory bandwidth for integration and power efficiency.
- On the CPU side, the SoC combines two 10-core Arm ‘Grace’ clusters for 20 cores, mirroring NVIDIA’s GB10 superchip design in a consumer-oriented package.
- NVIDIA and MediaTek are targeting an early 2026 unveiling, with driver refinements and power-frequency tuning expected to unlock further performance.