Nvidia's RTX Spark Devices Set for October Launch with Open-Source Collaboration Tool
Nvidia has announced that its RTX Spark devices equipped with N1X chips will begin shipping in October. The N1X chip supports up to 128GB of unified memory and provides a peak single-precision floating-point performance of over 1 million billion operations per second.
The RTX Spark devices come in two versions: one with a Blackwell GPU featuring 6144 CUDA cores, 20 Grace CPU cores, and supporting 24GB to 128GB of unified memory; the other with a GPU sporting 5120 CUDA cores, 18 CPU cores, and limited to 24GB to 32GB of unified memory.
Nvidia has also released an open-source tool called PAIR (Parallel Acceleration for Idle Resources), which allows users to connect multiple RTX devices, DGX Spark, or even Apple devices over a home network. The tool can schedule idle processing power and collaborate on AI tasks.
PAIR prioritizes utilizing idle resources and automatically adjusts as devices join or leave the network. A demo showcased a household with approximately 165 TFLOPS of unused processing power. In one example, a single RTX Spark laptop averaged 18 minutes to complete an AI task, while three paired devices using PAIR reduced this time to 8 minutes and 48 seconds.
Three AI applications, Perplexity Portable Computer, Hermes Agent, and OpenClaw, will benefit from simplified local deployment with the new tool.