Overview
- Nvidia announced the RTX Spark at Computex on Monday and said devices using the ARM-based system-on-chip will begin shipping in the autumn.
- The SoC pairs an ARM Grace CPU with a Blackwell GPU and up to 128 GB of shared memory to support high-bandwidth local model inference and on-device AI agents.
- Microsoft has committed Windows-on-ARM optimizations and a Surface Ultra is expected, and several PC makers including Asus, Dell, HP, Lenovo and MSI are reported launch partners.
- Early analysis and developer feedback raise flags about the Spark’s memory bandwidth compared with Apple’s M-series and past DGX Spark inference complaints, leaving real-world performance and pricing unproven.
- If the chip meets claims it could shift more AI workloads to personal devices and squeeze Intel and AMD, but the market and regulators will watch autumn reviews and shipping devices for confirmation.