Computing

Nvidia RTX Spark Tested in Microsoft Surface Laptop Ultra

Tom's Hardware evaluates Nvidia's RTX Spark Superchip inside Microsoft's Surface Laptop Ultra, highlighting impressive local AI capabilities alongside performance challenges.

Radwan AbdulhadiEditor and technology writer

Published: October 15, 2026 · Updated: October 15, 2026 · 3 min read

Nvidia RTX Spark Tested in Microsoft Surface Laptop Ultra — Image: Tom's Hardware

Introduction to the RTX Spark Platform

Nvidia CEO Jensen Huang originally announced the RTX Spark platform with the grand ambition to reinvent the personal computer during an appearance at Computex. By pairing the GB10 Superchip with a tailored version of Windows on Arm designed for secure and contained execution of local AI agents, major hardware partners are working to establish the PC as an active and autonomous companion.

To evaluate this hardware vision, Tom's Hardware examined Microsoft's Surface Laptop Ultra in top-end configuration. The tested hardware features the full RTX Spark Superchip with 20 CPU cores, 6144 CUDA cores, and 128GB of LPDDR5X-9400 RAM.

Microsoft Surface Laptop Ultra — (Image credit: Future) · Source: Tom's Hardware

Testing Methodology and Comparisons

Initial evaluations by Tom's Hardware focused on broad platform experiences and competitive performance rather than system-specific metrics like battery life or display quality. For comparison, testers used a 14-inch MacBook Pro powered by Apple's high-end M5 Max system-on-chip, which includes a 40-core GPU, an 18-core CPU, and 128GB of LPDDR5X RAM.

When examining raw processing power, the MediaTek Arm CPU complex on the Spark serves as a notable performance bottleneck. While the platform feels responsive during daily use, Geekbench results show that both single-core and multi-core CPU performance significantly trail Apple's M5 Max silicon.

Local AI and Agentic Benchmarking

Evaluating local artificial intelligence workloads has evolved beyond simple prompt processing and token generation metrics. To better measure agentic workflows, Artificial Analysis developed a new benchmark called AgentPerf-local, which replays inference traces across multiple agentic AI tasks to track total time-to-completion.

Using compatible models such as the Qwen 3.8-27B and the Qwen 3.6 35B A3B mixture-of-experts model running on llama.cpp, the Microsoft Surface Laptop Ultra with RTX Spark demonstrated competitive speeds. In tests with the dense Qwen 3.8-27B model, the RTX Spark system finished the AgentPerf benchmark approximately 25% faster than the M5 Max MacBook Pro.

AMD's older Ryzen AI Max+ 395 system required roughly twice as long to complete the same demanding agentic test. Although AMD has emphasized the lower cost of entry for its systems, faster task completion times on higher-end hardware can translate to greater productivity for professional users over time.

Performance results for AgentPerf-Local on Qwen 3.8 27B — (Image credit: Future) · Source: Tom's Hardware

Gaming Performance Challenges

Despite carrying RTX branding and featuring a powerful GPU configuration with 6144 CUDA cores and 48 streaming multiprocessors, the RTX Spark platform encounters significant hurdles in gaming workloads. Initial findings indicate that gaming is not a strong suit for Nvidia's new Windows on Arm hardware architecture, leaving room for future software and hardware optimization.

Sources

Tom's HardwareWe test the RTX Spark Superchip in Microsoft's Surface Laptop Ultra — Nvidia's Windows on Arm platform impresses in local AI, but gaming isn't its strong suit