NVIDIA ACE now supports the open-source Qwen3-8B small language model (SLM), enabling developers to deploy dynamic, real-time NPC characters in on-device PC games. The open-source Qwen3-8B AI model is now available through NVIDIA ACE as an In-Game Inferencing (IGI) SDK plugin that simplifies pipeline integration and optimizes simultaneous AI inference and graphics rendering for accelerated game performance.

Qwen3 enables developers to create enhanced game characters capable of real-time reasoning through scriptless events, dynamic responses based on player input, and contextually accurate actions.

A number of new updates have been released for the NVIDIA RTX Branch (NvRTX) of Unreal Engine. These include Unreal Engine 5.6.1, RTX Mega Geometry, and further ReSTIR PT performance and image quality improvements. The Bonsai Diorama demo accompanies the NvRTX 5.6.1 release and showcases RTX Mega Geometry, as well as utilizing ReSTIR PT and the DLSS 4 Technology suite.

Various improvements to the core IGI SDK

Magpie Flow has been added for real-time, on-device text-to-speech with multilingual support for Spanish and German; MultiLoRA adapters have been introduced to support rapid fine-tuning of model weights with minimal computational overhead. CUDA support has also been introduced in Graphics for Vulkan backends to optimize graphics and AI workloads for accelerated performance. You can download the latest IGI SDK and plugins here.