NVIDIA thông báo việc giao hàng thiết bị RTX Spark vào tháng Mười và ra mắt công cụ nguồn mở PAIR

The Huoxing Finance report states that NVIDIA has announced that RTX Spark devices equipped with the N1X chip will begin shipping in October this year. The N1X supports up to 128GB of unified memory, and the Blackwell GPU delivers up to petaflops of floating-point performance per second. The full-spec version features a Blackwell GPU with 6,144 CUDA cores, a 20-core Grace CPU, and supports 24GB to 128GB of unified memory; the other version has a GPU with 5,120 cores and an 18-core CPU, supporting only 24GB to 32GB of unified memory. NVIDIA has launched the open-source tool NVIDIA PAIR, which can connect multiple RTX devices, DGX Spark units, and even Apple devices within a home network to coordinate idle computing power for AI agent tasks. Compatible devices include computers with NVIDIA GPUs (RTX 20 series and later, RTX Pro GPUs, DGX Spark) and devices equipped with Apple M4 or newer chips. PAIR prioritizes the use of idle computing power and automatically adjusts as devices join or leave the network. During a media briefing, an example household was shown to have approximately 165 TFLOPS of underutilized computing power. A single Spark laptop completes an AI agent task with the Qwen 3.6 35B A3B model in an average of 18 minutes, while a three-device PAIR cluster completes it in an average of 8 minutes and 48 seconds. Three AI agent applications—Perplexity Portable Computer, Hermes Agent, and OpenClaw—will receive simplified local deployment methods.