At IFA 2026, Nvidia revealed new software features and hardware details aimed at local AI processing. The announcements address the growing demand for local large language models (LLMs), with the company noting a significant increase in model downloads over the past year.
“One-Click” Local AI Setup and Performance Boosts
To simplify the setup process for local AI, Nvidia is rolling out a one-click installation feature for several prominent applications. Starting in September, users with RTX GPUs and DGX systems will be able to configure models in Hermes Agent and OpenClaw 2.0 without manually adjusting inference engines or quantization. Additionally, Perplexity is bringing its portable computer agent to RTX GPUs with 24 GB of VRAM or more, allowing users to process sensitive data locally while retaining the option to escalate complex queries to the cloud. Nvidia also confirmed performance improvements for local inference, citing up to 1.9x higher throughput in Llama.cpp and up to 1.4x in vLLM for its Blackwell and DGX Spark systems.
Nvidia Pair Pools Idle GPU Compute Across Home Networks
The company also introduced Nvidia Pair (Personal AI Router), a free, open-source utility launching on September 3 under an Apache 2.0 license. Pair is designed to distribute AI inference tasks across multiple idle PCs on a local network. Compatible with Windows, Linux, and macOS, the software proxies through Ollama and LM Studio. It dynamically routes tasks based on queue depth and GPU utilisation, allowing users to leverage unused compute power from gaming rigs or secondary laptops without interrupting active gaming or rendering tasks.
RTX Spark N1X Specifications and Availability
Finally, Nvidia shared specifications for its upcoming RTX Spark architecture. The first chip, the N1X, will launch in October in two configurations. The higher-end variant features a 6,144-core GPU, a 20-core Grace CPU, and up to 128 GB of unified memory for laptops and compact desktops. A secondary configuration will offer a 5,120-core GPU, an 18-core Grace CPU, and up to 32 GB of unified memory for laptops. Nvidia confirmed that major publishers are adding anti-cheat and native support for the Spark platform, meaning titles like Apex Legends, The Finals, and Anno 117: Pax Romana will run on the new hardware.
