RTX Spark's First Benchmarks Leak: Beating M5 by 54% — Real Deal or Smoke and Mirrors?

RTX Spark's Clang benchmark scores 43,149, beating Apple M5 by 54%. But single-core lags M5 Max by 30%. The real killer: running 120B-parameter LLMs at just 80W.

💡 What You Will Learn

RTX Spark's Clang benchmark scores 43,149, beating Apple M5 by 54%. But single-core lags M5 Max by 30%. The real killer: running 120B-parameter LLMs at just 80W.

Last week at Computex 2026, Jensen Huang dropped a bombshell, and the buzz hasn't died down yet. Today, the first benchmark scores for RTX Spark have finally surfaced — scoring 43,149 in the Clang compilation test, beating Apple's M5 by 54%, though still trailing the M5 Pro by a noticeable margin. So just how capable is this ARM-based chip? Let's break it down. Many people confuse RTX Spark with last year's DGX Spark desktop box, but they're very different animals. DGX Spark was a $4,699 desktop AI supercomputer aimed at developers, built like a Mac Mini-sized box. RTX Spark, on the other hand, is a new consumer-grade SoC unveiled at Computex this year — designed to be soldered directly into laptops and Mini PCs. It comes in two variants: the standard N1 and the high-end N1X, boasting a 20-core Grace CPU, 6,144 CUDA cores, and up to 128GB of unified memory. The entire chip consumes just 80W — less than a third of the desktop RTX 5070's 250W. Built on TSMC's 3nm process, the CPU was co-developed with MediaTek while the GPU uses the Blackwell architecture. NVLink-C2C connects CPU and GPU at 600 GB/s bandwidth, allowing direct memory access either way. Simply put, this is Jensen's take on Apple Silicon.