Hands-on: Nvidia's RTX Spark SoC packs AI, creative and gaming power
TechRadar's reviewer spent over a week testing Nvidia's ARM-based RTX Spark, which the report describes as the company's first System on a Chip: an 18-core CPU paired with a 5,120-CUDA-core Blackwell RTX GPU, or a 20-core CPU with a 6,144-core GPU in the higher tier. In the reviewer's own Geekbench 7.1 runs, the RTX Spark N1X scored 2,623 single-core, below Apple's M5 at 4,288, but its 21,899 multi-core score exceeded the M5's 17,926. Local agentic tests using Qwen 3 8B ran MLPerf Client data-analyst and SWE workflows in 63.7 and 50.8 seconds, processing up to 14,005 and 10,238 tokens; the reviewer also highlighted the chip's hybrid workflow, which keeps chosen local models on-device while offloading demanding tasks to cloud frontier models.
The reviewer says the hybrid setup lets users ring-fence local models for sensitive parts of an agentic workflow while outsourcing specific demanding tasks to cloud frontier models, rather than choosing between fully local and fully cloud.
Having been at Nvidia's RTX Spark event in San Francisco earlier this month, I was lucky enough to get a firsthand walk through of the many things the 'superchip' can do. Since then I've spent almost every waking moment putting the ARM-based RTX Spark through its paces and seeing what makes it so different from other chipsets that have come before.
The RTX Spark is Nvidia’s first System on a Chip (SoC). Combining an 18-core CPU and 5120 Cuda core Blackwell RTX GPU all in one, or a 20-core CPU and 6144-core GPU for its meatier version, it offers an integrated chipset in the vein of Apple’s M series chips. While Apple’s silicon is known for its oomph, that Blackwell RTX GPU means you’ll also get considerable gaming performance here to boot.
How does this translate in terms of raw numbers? It’s certainly got a lot going on under the hood, as the RTX Spark N1X chip I’ve been playing around with squeezes 20 cores into its CPU. While the 2,623 single-core performance I clocked on Geekbench 7.1 falls somewhat short of the 4,288 we got for Apple’s M5 chip, its 21,899 multi-core score massively outstripped the latter’s 17,926. That means that for tasks like content creation, software development and AI, you should be able to expect more oomph in your corner.
#Nvidia#RTX Spark#Apple#Qwen#Asus#Adobe#Blender#Unity#MLPerf