NVIDIA TITAN RTX Combines AI, Real-Time Ray-Tracing, VR and Data Research
NVIDIA TITAN RTX GPU brings enough performance to desktops for demanding PC applications including AI research, data science and creative applications.
Driven by the new Turing architecture, NVIDIA says TITAN RTX delivers 130 teraflops of deep learning performance and 11 GigaRays of ray-tracing performance.
Turing is built with new RT Cores to accelerate ray-tracing, plus new multiple-precision Tensor Cores for AI training and inferencing. These two engines, plus greater compute and rasterisation, are expected to extend the capabilities of developers, designers and artists working in many different industries.
RT Cores are accelerator units that carry out ray tracing operations with extreme efficiency, instead of earlier software emulation-based ray-tracing approaches. Combined with NVIDIA RTX software and filtering algorithms, Turing can render ray-traced imagery in real time, including photorealistic elements with physically accurate shadows, reflections and refractions.
Tensor Cores run deep learning neural services that produce impressive graphics effects for games and fast AI inferencing for cloud-based systems. A major advance on the NVIDIA deep learning platform, this new hardware accelerates computations that account for most of the work involved in training a neural network.
TITAN RTX is capable of combining AI, real-time ray-traced graphics, virtual reality and high performance computing. As well as the deep learning and real-time ray-tracing performance, the card delivers 24GB of high-speed GDDR6 memory with 672GB/s of bandwidth, and twice the memory of previous- generation TITAN GPUs to fit larger models and datasets.
Applications – Neural Networks, Data and Content Creation
A VirtualLink port with the performance and connectivity VR headsets need is included. This port is the work of the VirtualLink consortium (AMD, Microsoft, NVIDIA, Oculus and Valve) to overcome the fact that the multi-cable, multi-port configurations needed to connect VR headsets to PCs are inconvenient for users. VirtualLink aims to make only a single port necessary.
Also, 100GB/s NVIDIA NVLink can pair two TITAN RTX GPUs to scale memory and compute. NVLink addresses interconnection issues in multiple-GPU systems by supplying higher bandwidth, more links and greater scalability. The multi-precision Turing Tensor Cores in TITAN RTX can make a PC powerful enough to replace a supercomputer for AI researchers and developers, allowing faster training and inference of larger neural networks.
For data scientists, TITAN RTX accelerates data analytics with NVIDIA RAPIDS. RAPIDS open-source libraries integrate directly with commonly used data science workflows to speed up machine learning. Also, because real-time ray-tracing and AI enable faster iteration, and greater computation and memory bandwidth support real-time 8K video editing, TITAN RTX has advantages for creative applications as well.
TITAN RTX will be available in the US and Europe later in December 2018 for US$2,499. www.nvidia.com