DailyPost 2465
NVIDIA’S NEW AI RESEARCH
For the uninitiated the research in AI and products thereof today may possibly start and end with ChatGPT, not even be bothered about OpenAI, the AI stable from where it comes from. Some news from Google and Microsoft on similar LLM lines and that is the end of the story. That itself is so much and so difficult to technically digest that there is no space and hunger for the research coming out of NVIDIA, pathbreaking indeed, which would define the full exploitation of the multimedia universe and insane utilities that would come out of it. At this nascent stage understanding the direction, and having some idea of the mammoth change would certainly suffice.
NIVIDA is about to present 20 AI focused research papers in a conference, and some of the results and simply mind blowing. These papers were released in May’23 and the passionate have painstakingly gone through it all, to unravel our AI for themselves and make in known to the world. As they say, the investment is the that in you. The world does not believe in it and so we are not ready to make AI knowledge / acumen investments in ourselves. The areas being comprehensively covered are gaming, animation and computation. The speeding up the processes can make a difference in from 10x, going up to even 100x in many instances.
The papers can be grouped for better understanding. One group of papers is to make graphics better or cost fewer resources to render. Complex scenes have to be rendered offline for long hours. A single frame of a Marvel movie takes 7 hours to render, and there are 24 frames in a second. PIXAR movies take 20 hours a frame to render. You can well understand the nature of research happening to transform it all. Then we have a paper on Real Time Neural Appearance Model. It is all about complete system of neural material shaders. This research paper states that neural models are trained on high-fidelity layered materials. The output seems being churned out is magical.
Another pathbreaking paper is titled Random-Access Neural Compression of Material Textures. Compression and uncompressing without losing any data is achieved. This method used 30% less memory than current state of art compressions. It can work on images at 4 times higher resolutions. This compression method is 13 times faster, compared to the state of the art tech for the same size images. It is just getting started. Anybody who thinks NVIDIA makes nothing but GPUs for gaming and data centres is really not paying attention. They innovate at every layer of AI from hardware to industry specific applications. It’s just a peep into the AI world which NVIDIA is determined to unravel as a part of its pathbreaking research and development.
NIVIDIA IS A COMPANY TO WATCH IN THE UNFOLDING MULTIMEDIA AI WORLD.
Sanjay Sahay