As it could not be otherwise, NVIDIA attended SIGGRAPH 2024, one of the most important international events if we talk about graphics and AI technologies. During the event, the company’s CEO, Jensen Huang, shared some very interesting news, which we will review below.

ChatRTX was one of the main topics of the event and received an important update that brought notable improvements, among which we can highlight new user interface more attractive and functional.

The company offered a technical demo with a large language model (LLM) using user data to solve quick queries, which is also optimized for PCs and workstations with NVIDIA GeForce RTX graphics solutions. This new version can be downloaded from the ChatRTX website and it is possible access the source code on GitHub.

Another of the great protagonists were NVIDIA ACE and Maxine. Jensen Huang took the SIGGRAPH stage to introduce James, a digital human powered by NVIDIA NIM microservices and, as our most advanced readers might have guessed, powered by NVIDIA ACE. This served to demonstrate the possibilities of interactive digital experiences powered by artificial intelligence.

As it could not be otherwise, NVIDIA Studio was also present. The green giant highlighted the arrival of important updates that make life easier for content creators and are integrated into the most popular applications today. Among the most important changes we can highlight:

• RTX Video HDRwhich is now available through Wonder8 Filmora and DaVinci Resolve, two very well known and used professional applications.
• NVIDIA Audio2Face, NVIDIA DLSS and other generative AI integrations in various applications and platforms such as Replikant, Adobe, Topaz and Getty Images.

Finally, NVIDIA has confirmed improvements in Stable Diffusion 3.0, which received specific optimizations for TensorRT that improve performance by up to 60%. The 3.1 and 8 billion Meta Llama models have also been optimized to work with NVIDIA RTX and GeForce RTX graphics cards. Both are natively supported by TensorRT-LLM, an open source library created by NVIDIA that leverages the power of tensor kernels in AI and deep learning related tasks.