NVIDIA’s new research
NVIDIA (NVDA) is moving quickly with its traditional research focus of developing advanced GPUs (graphics processing units) for AI (artificial intelligence), gaming, and automotive. It is also putting research efforts toward leveraging its technology into other AI applications.
At the SIGGRAPH computer graphics event, the company unveiled some of its research projects in which it brings AI in graphics.
NVIDIA combines AI and virtual reality
NVIDIA’s Project Holodeck renders real-world objects in a photorealistic virtual world and also simulates real-world physics. This enables virtual objects to react to other objects, people, and the environment as they would in the real world. NVIDIA’s ISAAC Lab robot simulator uses VR (virtual reality) to train robots that interact with people.
NVIDIA has combined both tools to help researchers train robots interacting with humans in practical environments created on the Holodeck platform. This helps reduce the risk and cost of training.
AI in graphics
NVIDIA is bringing AI to computer graphics to create realistic human facial animations at a lower cost and quicker turnaround. The company initiated collaborative research with game maker Remedy Entertainment. NVIDIA used Remedy’s animation data to build video- and audio-driven neural networks. NVIDIA combined this data with its AI technology to train machines on this network to produce facial animations directly from an actor’s videos and voices.
NVIDIA’s solution generated facial animation for an entire game using just five minutes of training data. The solution can help game developers add more supporting game characters, create live animated avatars, and produce games in multiple languages.
NVIDIA is also looking to bring AI support in creating the background for a game. It is developing an AI solution for an anti-aliasing problem. The problem occurs when ray tracing and rasterization renders blur images in the virtual world. The company trained a neural network to identify blurred artifacts and replace those pixels with smooth anti-aliased pixels, making the images sharper.
NVIDIA is looking for an AI solution to trace virtual light rays. It is also supporting immersive VR with its VRWorks 360 Video SDK (software development kit), which enables production houses to do 360-degree stereo stitching live. VR camera firm Z Cam would integrate VRWorks SDK in its V1 Pro 360-degree VR camera.
Moore’s law is slowing as the performance improvement and cost reduction with every shrink in the die size is decreasing. In such an environment, NVIDIA is looking to enhance its GPU performance by moving from single-chip module to MCM (multi-chip module). It is developing an MCM design wherein multiple smaller GPU modules are linked through a high-speed interconnect.
Intel (INTC) and Advanced Micro Devices (AMD) have developed similar multi-chip designs. Intel’s embedded multi-die interconnect bridge connects multiple chips from different generations and AMD’s Infinity Fabric connects multiple GPUs at minimal latency.
Next, we will look at NVIDIA’s financial performance.