NVIDIA has unveiled an updated version of its DGX Spark mini-supercomputer, now featuring 64 GB of unified memory. This new model will be available beginning October 23 through several manufacturers, including Acer, ASUS, Dell, Gigabyte, HP, and MSI, with a starting price of $4,999. The device retains its core technology, which includes the GB10 Grace Blackwell Superchip, DGX OS, the CUDA environment, and NVIDIA’s software stack for AI development. The updated configuration allows users to run AI models with up to 80 billion parameters locally. According to NVIDIA, pairing two DGX Spark 64 GB units can create a combined system with 128 GB of memory, double the memory bandwidth, and the ability to support models as large as 200 billion parameters. In testing with the Qwen 3.8 27B model, NVIDIA observed up to 70% more performance when using two machines in a cluster. However, this improvement depends on the specific model and how effectively the workload can be divided between the two systems. NVIDIA also plans to release Sync Model Launcher by the end of the month. This new software will allow users to download models such as Qwen 3.8 27B and run them on a DGX Spark or across multiple machines through a graphical interface. NVIDIA Sync will automatically distribute the model across different nodes and aims to configure OpenCode to use the local model directly from a browser for development tasks. The DGX Spark 64 GB is designed for a specialized audience, such as those engaged in continuous operation programming, research involving AI agents, or tasks that require long context processing and multiple simultaneous agents. The machine is compatible with major local AI tools currently in use, including llama.cpp, Ollama, vLLM, and LM Studio.