NVIDIA's Free PAIR Tool Turns Home PCs Into One AI Cluster
The company unveiled PAIR in beta at IFA 2026 and confirmed October launch dates for Lenovo and Acer RTX Spark machines.
- PAIR, free and open source, pools RTX 20-series GPUs, DGX Spark and Apple M4 chips.
- RTX Spark PCs launch in October with a 1-petaflop Blackwell GPU and 128GB unified memory.
- llama.cpp inference runs up to 1.9 times faster on the RTX 5090, NVIDIA's own benchmark shows.
Why it matters: Calling it local AI does not remove the toll: every petaflop still needs hardware only NVIDIA sells.