Nvidia RTX Spark leads local AI tests but struggles with gaming

Nvidia’s RTX Spark platform delivered strong local AI performance in Microsoft’s Surface Laptop Ultra, outperforming competing Apple and AMD systems in agentic benchmarks. The tested laptop pairs the GB10 Superchip with 20 CPU cores and 6144 CUDA cores alongside 128GB of LPDDR5X-9400 RAM. The Windows on Arm platform targets local agents for work and creation.

AgentPerf-local testing measured task completion across inference sequences rather than isolated token-generation speeds. With a dense Qwen model, Apple’s M5 Max MacBook Pro took about 25% longer than the Surface Laptop Ultra, while AMD’s Ryzen AI Halo lagged substantially behind both. Nvidia also led with a lighter mixture-of-experts model. Different power and thermal configurations complicate comparisons, however, and battery-life results remain unavailable.

Gaming proved less convincing. Arm compatibility constraints, Prism emulation overhead, CPU limitations and shared memory bandwidth contributed to high input latency and compromises in demanding games. DLSS upscaling and frame generation helped, but dedicated gaming laptops remain the stronger option. Promised Windows features for secure, manageable local agents were not ready for testing. The tested configuration lists for $5899 with only 1TB of storage, making the platform a costly proposition primarily suited to users prioritizing privacy, offline access and experimentation with local models.