wavecrafters blog

Our small contribution to the developer community

GTC 2014 Third Day

| Comments

PIXAR

Todays big keynote was presented by the people from PIXAR. It obviously wasnt full of big news like yesterdays keynote was, but it was anyway very impressive. They talked about how they use NVIDIAs GPUs to animate and light their movies. One of the most impressive examples they showed was their lightning techniques. They have a software that throws millions of rays to thousands of objects, and then they are capable of following those rays trajectories to form the final lighting. What is impressive is that its real time calculation (it actually takes them a couple of seconds until it stabilises a little bit, at the beginning its rather blurry, and you can almost see where the rays hit), and thats allows their lighting artists to completely change the lighting of the scene with just a few steps, a process that before took them a handful of hours.

About Pascal and Volta

Aparently Pascal doesnt substitute Volta, it is the transition from Maxwell to Volta. Given that the 3D stacked memory was a feature of Volta present in Pascal, as well as the hardware support (NVLink) for unified memory was initially planned for Maxwell, maybe we can adventure that Pascal will not feature the 16nm FinFet initially planned for Volta in 2016? In that scenario that would be the improvement Volta brings to the table. We are no expert in chip processes, but we’ve been reading articles about delays in the introduction of any chip going beyond 16nm, it is not even clear how economically viable they could be without a new important leap forward in chip engineering.

TALKS

Besides that, one of the talks that most impressed us was the one by the people of Shadertoy, Iñigo Quilez left everyone with their mouths open. He makes incredible 3D animations with just a fragment shader, using math to form the objects and shapes without having an array of vertices defining them. You can check out their webpage here: [Shadertoy](https://www.shadertoy.com/

COMPUTER VISION.

Another very interesting talk was the one by Ikuro Sato and Hideki Niihara. They use a deep neural network for pedestrian detection, using a tegra K1. They made a live demo of their work at the talk, and it was impressive to see how it could calculate distance, height and direction in real time. in this same line was the talk by NVIDIA representatives, that versed around the NVIDIA visionworks toolkit. This will be a library for computer vision and image processing that will allow programmers to develop new applications quickly. It will be modular and extensible, and will run on the Tegra K1. The possibilities for the automotive and mobile industries are enormous. We cant wait to have a K1 in our hands.

CUDA, GTC, conference

« second GTC day and Keynote Exclusive Prefix Sum Scan »

Comments