NVIDIA ACE Adds Open Source Qwen3 SLM for On-Device Deployment in PC Games

To help create real-time, dynamic NPC game characters, NVIDIA ACE now supports the open source Qwen3-8B small language model (SLM) for on-device deployment.

In addition, the RTX Branch of Unreal Engine 5 includes a new rendering tech demo with source code for developers to use as a reference when implementing the latest NVIDIA RTX path tracing technologies.

Integrating AI speech and real-time reasoning into NPC game characters

NVIDIA ACE is a suite of digital human technologies for middleware and game developers that provides ready-to-integrate cloud and on-device AI models for the most important aspects of digital humans—speech, animation, and intelligence. The open source Qwen3-8B AI model is now available through NVIDIA ACE as an In-game Inferencing (IGI) SDK plugin, which simplifies the integration within your gaming pipeline and optimizes simultaneous AI inference and graphics processing for accelerated game performance. Qwen3 enables developers to build enhanced game characters capable of real-time reasoning through non-scripted events, dynamic responses based on player inputs and contextually accurate actions.

Alongside the release of Qwen3-8B, NVIDIA is also updating the IGI plugin for Magpie TTS Flow, an experimental text-to-speech (TTS) model for testing dynamic voice capabilities on-device in real-time. The plugin now adds multilingual support, allowing developers to test Spanish and German within their end-to-end character pipelines.

In addition, several improvements have been made to the core IGI SDK including:

MultiLORA adapters to support fine-tuning model weights quickly with minimal computational overhead.
CUDA in Graphics support for Vulkan backends, which optimizes graphics and AI workloads for accelerated performance when both are running simultaneously.
A new sample that teaches developers how to use function calling in their language model of choice.

Download the latest IGI SDK and plugins today.

Latest NVIDIA RTX Unreal Engine 5.6.1 update features Bonsai Diorama rendering demo

Video 1. New Bonsai Diorama demo showcasing the latest NVIDIA rendering technologies for Unreal Engine developers

NVIDIA RTX Branch of Unreal Engine is an optimized and feature-rich branch that contains all the latest NVIDIA rendering technologies for Unreal Engine developers. Today, we’re releasing a set of new updates including the Unreal Engine 5.6.1 release, 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 along with utilizing ReSTIR PT and the DLSS 4 Technology suite. Access the Project Source, documentation, and developer guide here. Be sure to join our live Level Up with NVIDIA webinar next Tuesday, October 28, 2025 at 10AM PT to get more details about the Bonsai scene and have the chance to ask us questions directly during a live Q&A. Register here.

Join us at Epic MegaJam this week

demo image from the Zorah demo in unreal engine — *Figure 1. A still of the Zorah demo which uses the NVIDIA RTX Branch of Unreal Engine 5*

For Unreal Engine developers, be sure to participate in our ‘RTX On: Push the Limits with RTX Neural Rendering in Unreal Engine’ modifier category for a chance to win a GeForce RTX 50 Series GPU.

The ‘RTX On’ category invites developers to harness the full range of NVIDIA’s latest graphics technologies using the NVIDIA RTX Branch of Unreal Engine (NvRTX). Submissions should showcase innovative use of at least one NVIDIA RTX feature from NvRTX—such as RTX Mega Geometry, RTX Direct Illumination, RTX Indirect Illumination (ReSTIR GI), RTX Path Tracing (ReSTIR PT), in addition to one of the technologies from the DLSS 4 suite—to create immersive visuals or interactive experiences that surpass traditional rendering techniques.

Projects must clearly demonstrate tangible benefits or novel effects made possible by RTX neural rendering. Enter here by Friday, October 24, 2025 at 8:59 PM PT.