NVIDIA RTX Technology

Fifth-generation Tensor Cores accelerate deep learning workloads, up to 3X higher throughput using fine-grained scaling techniques, optimizing sparse neural networks. They support a wide range of precision formats, including the newly added FP4 support for massive throughput increases, along with TF32, BF16, FP16, FP8, and FP6 data types—enabling flexible AI model development. Support for the new Second-Generation FP8 Transformer Engine used in our data center-class Blackwell GPUs is also included. Developers gain tools for rapid LLM prototyping, scalable inference deployment, and backward-compatible framework support (e.g., TensorFlow, PyTorch).

Fourth-generation RT Cores enable true-to-life neural graphics by delivering up to 2X faster ray tracing performance over the prior generation, achieving photorealistic scene rendering with physically accurate shadows, reflections, and refractions. Create photoreal, physically accurate scenes and immersive 3D designs with neural graphics-based technologies, such as RTX Mega Geometry. These cores are engineered for industrial-grade precision—ideal for prototyping, immersive 3D environments, and content creation for media and entertainment—while dramatically reducing iteration cycles for unprecedented productivity gains. Working with applications leveraging APIs such as NVIDIA OptiX, Microsoft DXR, and Vulkan ray tracing, systems based on RTX PRO GPUs will power interactive design workflows to provide immediate feedback for unprecedented levels of productivity.

Blackwell Streaming Multiprocessor (SM) integrates next-gen CUDA cores and neural shaders to redefine AI-augmented graphics and compute performance. Featuring up to 1.4X higher FP32 throughput versus the previous generation, the Blackwell SM architecture scales to 24,064 CUDA cores per GPU, delivering up to 125 TFLOPS of single-precision performance for industrial-grade simulations, 3D modeling, and ray-traced rendering. Enhanced neural shaders embed AI directly into programmable pipelines, enabling hybrid workflows like RTX Mega Geometry and DLSS 4.0 multi-frame generation. With support for FP64, FP32, FP16, and BF16 precision modes, a larger L2 cache, and higher clock speeds, Blackwell SMs accelerate tensor operations, sparse neural networks, and photorealistic rendering for AECO, manufacturing, and media workflows. This unified design ensures seamless concurrency for ray tracing, AI inferencing, and compute tasks—empowering professionals to tackle AI-augmented workflows with unprecedented efficiency.

Deep Learning Super Sampling (DLSS) is a revolutionary suite of neural rendering technologies that uses AI to boost FPS, reduce latency, and improve image quality. ‌The latest breakthrough, DLSS 4, brings enhanced Ray Reconstruction and Super Resolution, powered by NVIDIA RTX PRO™ Blackwell Series GPUs and fifth-generation Tensor Cores. The NVIDIA RTX Blackwell architecture features AI multi-frame generation that boosts DLSS 4’s frame rates up to 2X over the previous DLSS 3/3.5, while maintaining or exceeding native image quality and providing low system latency.

Bring small neural networks into programmable shaders, enabling a new era of graphics innovation. The applications of neural shading are vast, including radiance caching, texture compression, materials, radiance fields, and more. Train and deploy neural networks within shaders to unlock new compression and approximation techniques for next-generation asset generation. The RTX Neural Shaders SDK enables developers to train their game data and shader code on an RTX PRO AI Workstation and accelerate their neural representations and model weights with NVIDIA Tensor Cores at runtime.

A new RTX technology aimed at dramatically increasing the geometric detail that is possible in ray-traced applications. Accelerate BVH building for cluster-based geometry systems, enabling up to 100x more ray-traced triangles and better performance in heavily ray-traced scenes. RTX Mega Geometry is available today through NVIDIA RTX Kit, a suite of rendering technologies to ray trace scenes with AI, render scenes with immense geometry, and create photorealistic visuals.