Nvidia Brings Data-Center Silicon to the Desktop With the RTX Spark Superchip
TL;DR — At Computex, Nvidia unveiled the RTX Spark Superchip and said it wants to "reinvent the PC" with Microsoft — putting datacenterclass silicon into ordina
TL;DR — Nvidia announced the RTX Spark Superchip at Computex — a 20-core CPU fused to a 6,144-core Blackwell GPU over NVLink with shared memory — and a stated plan, with Microsoft, to "reinvent the PC." Ships this fall via Dell and Lenovo.
Nvidia is moving down the stack. At Computex, CEO Jensen Huang unveiled the RTX Spark Superchip and said the company intends, alongside Microsoft, to "reinvent the PC." Read it as an opening move against Intel and AMD in the one segment Nvidia doesn't yet control.
The silicon
- CPU: up to 20 cores
- GPU: Blackwell-generation, 6,144 cores
- Interconnect: NVLink — the data-center link, now on a PC die
- Memory: shared between CPU and GPU
The architecture, not the core count, is the signal. Unified memory + NVLink is data-center design language. It exists to keep a large model resident on-device, cutting the cloud round-trip.
Distribution
Not a prototype. Dell and Lenovo ship Spark-based laptops and desktops this fall, per Nvidia.
Read
Nvidia owns training and inference silicon. The PC was the gap. Spark closes it and reframes the "AI PC" category around on-device model execution — terrain Intel and AMD currently hold.
FAQ
What are the RTX Spark Superchip's specs?
Up to 20 CPU cores and a 6,144-core Blackwell GPU sharing memory over NVLink, in laptops and desktops shipping this fall.
Who does it threaten?
Intel and AMD — it's Nvidia's entry into the PC CPU/GPU segment they dominate.
Why does shared memory matter?
It lets the machine run large AI models locally rather than offloading to the cloud.
Sources: CNBC, The Spokesman-Review.
Image: Markus Spiske, CC0 / public domain, via Wikimedia Commons.
← Back to all posts