RTX Spark review: beats M5 Max in local AI, lags in gaming
Tom's Hardware tested Nvidia's RTX Spark platform in a top-trim Microsoft Surface Laptop Ultra carrying the full RTX Spark Superchip with 20 CPU cores, 6,144 CUDA cores and 128GB of LPDDR5X-9400 memory. Running Artificial Analysis's new end-to-end agentic benchmark AgentPerf-local with Qwen 3.8 27B and Qwen 3.6 35B A3B, the RTX Spark system finished about 25% faster than a 14-inch MacBook Pro with Apple's M5 Max, while AMD's Ryzen AI Max+ 395 (Strix Halo) needed roughly twice as long, about 40 minutes, on the heavier test. Gaming is a weaker point: roughly 300 GB/s of shared memory bandwidth, Arm CPU cores and the overhead of the Prism x86 emulator leave frame rates and input latency behind dedicated RTX 50-series gaming laptops, and the reviewer leans on DLSS Frame Generation to compensate. The outlet notes that launch embargoes limit this first round to platform-wide results, with battery life, build and display details to follow.
It is an early independent look that scores local AI systems by end-to-end agent task time rather than token speed, giving a comparable data point for the "run agents locally" pitch.