This article is available in: French
At SIGGRAPH 2023, NVIDIA unveiled new GPUs, partnerships as well as improvements to their Omniverse platform. Here are the major announcements.
Three new GPUs
These new graphics cards come with respective 32GB, 24GB, and 20GB GDDR6 memory respectively. They support DLSS3. According to NVIDIA, they deliver effective performance improvements of up to 2x for rendering and 1.6x for 3D display compared to the previous generation.
NVIDIA RTX 5000 is available now ($4,000). NVIDIA RTX 4500 and 4000 will be available this fall ($1,250 and $2,250 respectively). Both RTX 4500 and 5000 are double slot GPUs.
Furthermore, NVIDIA announces the arrival of new workstations from HP, Lenovo, BOXX, Dell, incorporating up to 4 NVIDIA RTX 6000 Ada GPUs, each with 48GB of memory (192GB in total). 5.8 TFLOPS of power at your fingertips for AI applications.
NVIDIA explains that these workstations are designed with generative AI and intensive graphical tasks in mind (including working on large digital worlds on Omniverse)
Picasso: AI creates backgrounds for your 3D scenes
NVIDIA also discussed a partnership with Shutterstock regarding NVIDIA Picasso, a cloud-based generative AI system.
Artists will be able to input prompts (text or image) to generate HDRI maps that can be used as backdrops for their 3D scenes. These 360° images will be licensed and available in 8K.
Autodesk is integrating Picasso’s generative AI services into Maya. We can expect other software developers to follow suit with their own tools.
An update for Omniverse
Regarding Omniverse, NVIDIA’s real-time collaborative 3D project platform, a major update is announced (in practice, some of the announcements have already been available for several weeks).
Notably, Adobe Firefly will soon arrive as an API, allowing the development of applications using generative AI.
Other APIs are also on the menu:
- RunUSD, which validates OpenUSD file compatibility with different standard versions and generates path-traced images via Omniverse Cloud.
- DeepSearch, an agent capable of semantic search within massive databases of untagged assets.
- ChatUSD, which is capable of answering questions about USD and generating Python scripts involving USD.
- USD-GDN Publisher, designed for publishing experiences based on OpenUSD in the Omniverse Cloud GDN. This includes the ability to stream these applications on the web and to mobile devices.
NVIDIA also seizes this opportunity to remind us of its involvement in OpenUSD, as Omniverse is built on this technology. The company highlights its recent announcement about the Alliance for OpenUSD and underscores the adoption of OpenUSD by companies like Wonder Dynamics with its AI tools, Luma AI for NeRF assets creation, Move AI with motion capture technology using a single camera.
Watch the full NVIDIA Keynote
The keynote also contains other announcements, including when it comes to AI in areas that don’t involve computer graphics. You can watch the full keynote below.