MetaVRain AI-powered 3D rendering chip is 1000x more powerful than current Nvidia GPUs
A South Korean team of researchers led by Professor Yoo Hoi-Jun working at the Korea Advanced Institute of Science & Technology has recently completed the development of an AI-powered chip that can transform 2D images into completely editable 3D models with very low power consumption and exponentially increased processing speed over existing solutions from Nvidia. DigiTimes reports that the new neural rendering processor is known as the MetaVRain and it can easily be integrated into a portable device the size of a smartphone. It essentially aims at simplifying the processes of turning real-world objects and persons into 3D avatars for the Metaverse.
As a result of this simplification, costs to produce 3D renders can be lowered substantially as the automated AI training does away with the cumbersome manual 3D editing through specialized suites like Blender. VRAM requirements can also be reduced by up to 180 times. The capturing and rendering processes thus becomes as easy as snapping a few pictures and uploading them into a smart application. After transforming the 2D material into an easily-editable 3D model, the smart app can then apply textures pre-trained from certain art styles like that of Picasso or Van Gogh.
In order to remedy the initial inefficiencies with the AI-assisted rendering processes, the team of researchers implemented an algorithm that emulates human visual cognition. Just like humans who first perceive the outlines of an object and then compose the whole shape, the MetaVRain chip first delineates shapes via low-resolution voxels and then applies previously rendered structures to increase calculation speeds.
Image capture is done through a very advanced CMOS sensor and the rendering process is done at least 911 times faster compared to current consumer GPUs. A demo video shows that MetaVRain can render a model at 32.8 fps, while an RTX 2080 GPU renders only less than 3% of a frame in one second. The superior speeds further reduce the energy requirements for a single rendered frame by 26,400 times (133 mW) compared to consumer GPUs. Such specs enable real-time rendering for VR/AR applications using just a smartphone.
While AI developments are now overshadowing everything related to the Metaverse, Professor Yoo Hoi-Jun and his team believe that AI-based chips like the MetaVRain can ultimately speed up the advent of the Metaverse.