Generative AI: Revolutionizing PC Software with NVIDIA RTX AI PCs
Generative AI is rapidly reshaping how we experience and interact with PC software. From crafting digital humans to enhancing writing tools, intelligent agents, and creative applications, this transformative technology is opening new frontiers. The power behind these innovations? NVIDIA RTX AI PCs, which simplify the exploration of generative AI and enhance performance on Windows 11 systems.
NVIDIA TensorRT: A Leap Forward in AI Deployment
NVIDIA has reimagined its TensorRT technology for RTX AI PCs, combining top-tier performance with a more compact package size. This results in seamless AI deployment across more than 100 million RTX AI PCs globally. Unveiled at Microsoft Build, TensorRT for RTX enjoys native support from the Windows ML inference stack, offering app developers broad hardware compatibility and cutting-edge performance.
Developers eager to integrate AI features can use NVIDIA’s software development kits (SDKs). These SDKs offer diverse options, from NVIDIA DLSS for superior graphics to multimedia enhancements like NVIDIA RTX Video. This month, prominent software applications from companies like Autodesk and Topaz Labs are rolling out updates to leverage RTX AI features and acceleration.
For those new to AI, NVIDIA provides prepackaged, optimized AI models through NVIDIA NIM. These models are compatible with popular applications such as Microsoft VS Code and ComfyUI. Notably, the FLUX.1-schnell image generation model will soon be available as a NIM microservice, with updates to the FLUX.1-dev microservice enhancing support for more RTX GPUs.
Accelerated AI Inference with TensorRT for RTX
Today’s AI PC software stack often forces developers to choose between performance and custom hardware optimizations. Windows ML addresses these challenges by leveraging ONNX Runtime and seamlessly connecting to an optimized AI execution layer maintained by each hardware manufacturer.
For GeForce RTX GPUs, Windows ML automatically utilizes the TensorRT for RTX inference library, delivering over 50% faster performance for AI workloads compared to DirectML. This efficiency ensures that developers can focus on innovation without being bogged down by performance concerns.
TensorRT, initially designed for data centers, has been adapted for RTX AI PCs. It now uses just-in-time, on-device engine building to optimize AI model execution for specific RTX GPUs, reducing the library’s file size by eightfold. Developers can access TensorRT for RTX through the Windows ML preview and as a standalone SDK on the NVIDIA Developer platform.
Expanding the AI Ecosystem on Windows 11 PCs
NVIDIA provides a wide array of SDKs for developers wanting to incorporate AI features or boost app performance. These include NVIDIA CUDA and TensorRT for GPU acceleration, NVIDIA DLSS for enhanced graphics, and NVIDIA Riva for generative AI.
Leading applications are embracing these SDKs to introduce unique features. For instance, LM Studio has upgraded to the latest CUDA version, resulting in a 30% performance increase. Topaz Labs is releasing a generative AI video model to enhance video quality, while Chaos Enscape and Autodesk VRED are integrating DLSS 4 for better performance and image quality. Bilibili is incorporating NVIDIA Broadcast features like Virtual Background to improve livestream quality.
By collaborating with Microsoft and top AI app developers, NVIDIA is driving AI innovation on RTX-powered machines through the integration of Windows ML and TensorRT.
Simplifying Local AI with NIM Microservices and AI Blueprints
Developing AI on PCs can be daunting, with over 1.2 million AI models available on platforms like Hugging Face. NVIDIA NIM simplifies this process by offering curated AI models packaged with all necessary files, optimized for full performance on RTX GPUs. These containerized NIM microservices can be effortlessly run on PCs or in the cloud.
NVIDIA NIM microservices are accessible through build.nvidia.com or popular AI apps such as Anything LLM and ComfyUI. During COMPUTEX, NVIDIA will launch the FLUX.1-schnell NIM microservice, designed for fast image generation, and update the FLUX.1-dev microservice to support a broader range of GeForce RTX GPUs.
Developers can also explore NVIDIA AI Blueprints, which provide sample workflows and projects using NIM microservices. These blueprints serve as a starting point for developers to customize and extend AI functionalities.
Project G-Assist: Enhancing User Experience with AI Assistants
NVIDIA’s Project G-Assist, an experimental AI assistant integrated into the NVIDIA app, allows users to control their GeForce RTX systems using voice and text commands. This offers a more intuitive interface compared to traditional control panels. Developers can utilize Project G-Assist to create plug-ins and publish them on platforms like Discord and GitHub.
The Project G-Assist Plug-in Builder, a ChatGPT-based app, facilitates no-code or low-code development with natural language commands. New open-source plug-in samples are available on GitHub, demonstrating how on-device AI can enhance PC and gaming workflows, such as sharing game highlights on Discord or automating IoT routines with IFTTT.
SignalRGB is developing a G-Assist plug-in for unified lighting control across multiple manufacturers, which will be available through the SignalRGB app. Furthermore, the AI community can leverage G-Assist as a custom component in Langflow for integrating function-calling capabilities in low-code or no-code workflows.
Conclusion
Generative AI, powered by NVIDIA RTX AI PCs, is set to redefine the PC software landscape. With advancements in TensorRT and the expansion of the AI ecosystem on Windows 11, developers have unprecedented tools at their disposal. Whether through NIM microservices or Project G-Assist, NVIDIA is making AI accessible and practical for developers and enthusiasts alike. As AI continues to evolve, staying informed and engaged with these developments is crucial for anyone interested in the future of technology.
For more Information, Refer to this article.
