NVIDIA Blackwell GeForce RTX 50 Series: New AI SDKs and Tools Unveiled

NVIDIA has recently unveiled its next-generation PC GPUs, the GeForce RTX 50 Series, along with a suite of new AI-powered SDKs and tools designed for developers. Powered by the NVIDIA Blackwell architecture, 5th Gen Tensor Cores, and 4th Gen RT Cores, the GeForce RTX 50 Series represents a significant leap forward in AI-driven rendering technologies, including neural shaders, digital human technology, geometry, and lighting. This article delves into the new and updated SDKs that empower developers to harness the full potential of NVIDIA Blackwell GeForce RTX 50 Series GPUs.

To ensure seamless compatibility with the GeForce RTX 50 Series, NVIDIA recommends that developers update to the latest versions of its AI frameworks. CUDA Toolkit 12.8 and NVIDIA TensorRT 10.8 are now available, optimized to maximize AI performance on the RTX 50 Series GPUs. PyTorch updates for native Windows support on NVIDIA Blackwell RTX GPUs have been uploaded to the PyTorch GitHub repository, with PiPy binaries and packages for Windows to follow soon. PyTorch for Linux x86_64 on NVIDIA Blackwell RTX GPUs is now accessible through the nightly builds. For detailed instructions on updating your applications, refer to the Software Migration Guide for NVIDIA Blackwell RTX GPUs: CUDA 12.8, PyTorch, TensorRT, and Llama.cpp Guide.

The GeForce RTX 50 Series GPUs, combined with the latest SDK updates, enable developers to create revolutionary gaming experiences. NVIDIA DLSS (Deep Learning Super Sampling) is a suite of neural rendering technologies that leverage AI to boost FPS, reduce latency, and enhance image quality. DLSS 4, powered by GeForce RTX 50 Series GPUs and 5th Gen Tensor Cores, introduces DLSS Multi-Frame Generation, capable of generating up to three additional frames and working in tandem with the full suite of DLSS technologies to deliver up to an 8x increase in frame rates compared to traditional brute-force rendering. Furthermore, DLSS Ray Reconstruction, DLSS Super Resolution, and DLAA technologies are now powered by Transformer-based models, improving image and lighting details and stability across all GeForce RTX GPUs.

NVIDIA DLSS is a suite of neural rendering technologies that leverage AI to boost FPS, reduce latency, and enhance image quality. DLSS 4, powered by GeForce RTX 50 Series GPUs and 5th Gen Tensor Cores, introduces DLSS Multi-Frame Generation, capable of generating up to three additional frames and working in tandem with the full suite of DLSS technologies to deliver up to an 8x increase in frame rates compared to traditional brute-force rendering. Furthermore, DLSS Ray Reconstruction, DLSS Super Resolution, and DLAA technologies are now powered by Transformer-based models, improving image and lighting details and stability across all GeForce RTX GPUs.

NVIDIA ACE (Avatar Cloud Engine) is a suite of digital human technologies that use generative AI to bring game characters and digital assistants to life. ACE now allows you to easily add agent capabilities to digital humans in your games or applications. It includes: Early access to new multimodal SLMs for advanced and autonomous agent workflows, supporting longer contexts and complex reasoning tasks. Audio2Face 3D NIM uses real-time audio to provide advanced lip sync and facial animation.

New and updated SDKs supporting content creation on Blackwell RTX GPUs include the following.

NVIDIA Maxine is a collection of high-performance, easy-to-use NVIDIA NIM microservices and SDKs for deploying AI features to enhance audio, video, and augmented reality effects for video conferencing and remote presence. New features include: Studio Voice can make any microphone sound professional. “Virtual Key Light” can reshape faces to use the effect of a virtual key light (coming soon).

NVIDIA Iray SDK is an intuitive, physically based rendering technology that generates photorealistic images for interactive and batch rendering workflows. Updates include: Improved diffuse and glossy BRDFs using the new NVIDIA MDL SDK 1.10. Improved tessellation and displacement of geometry. Accurate and reliable rendering of caustics. A new mode that automatically enables and disables caustic sampling, improving rendering quality or performance. Support for faster cluster or network rendering.

The NVIDIA Video Codec SDK is a set of APIs for performing hardware-accelerated video encoding and decoding on Windows and Linux. Updates include: Support for 4:2:2 H.264, HEVC encoding and decoding to leverage 9th generation NVENC encoding in Blackwell. Introduction of MV-HEVC and UHQ AV1 to improve encoding quality. 2x memory footprint decode optimizations and 2xH.264 decode throughput per NVDEC compared to previous generations. These updates are coming soon through FFMPEG, Microsoft DXVA, and MFT frameworks.

Ready to experiment, develop, and optimize new AI features on over 100 million RTX PCs worldwide? Get started with AI on NVIDIA RTX PCs. For more information on adding support for NVIDIA Blackwell RTX GPUs in your AI applications for higher performance, check out the Software Migration Guide for NVIDIA Blackwell RTX GPUs: CUDA 12.8, PyTorch, TensorRT, and Llama.cpp Guide.