Differences

This shows you the differences between two versions of the page.

--- devwiki:nvidia [2022/09/20 03:22] – ying
+++ devwiki:nvidia [2022/09/21 09:00] (current) – [Nvidia for AI] ying
@@ Line 12: / Line 12: @@
     * in addition, digital twins also updates with the main physical copy, anythings happen to main one, the digital twin one update to the same.
   * nivida omniverse is the tool to simulate the real world, to build digital twins,
+    * its tutorials: https://docs.omniverse.nvidia.com/plat_omniverse/common/video-list.html
     * it bridge realtime collaboration between different users and different graphic softwares.
     * omniverse audio2face: to generate face animation from audio
     * nvidia OVX server provide hardware support to build larget scale digital twins
+    * omniverse system: digital twins+ robotics; design +content creation; integration; rendering; sensors; asset lib;
+      * AI: drive, ISAAC (for move+manipulate stuff), metropolis (auto infrastructure), holoscan (robotic medical)
+      * replicator: generate + train synthetic data for train+test AI model
+      * omnigraph, behavior, animation: run data center scale 3d application
+      * avatar (wip): build digital humans
+      * nvidia open source Material Definition Language (MDL): https://developer.nvidia.com/rendering-technologies/mdl-sdk
+        * https://developer.nvidia.com/rendering-technologies/mdl-sdk
+        * tut: https://www.nvidia.com/en-us/on-demand/session/gtcspring22-se2310/?playlistId=playList-5168fd54-82d2-4179-a612-491b68322489
+        * tut: https://www.nvidia.com/en-us/on-demand/session/gtcspring22-s41207/?playlistId=playList-5168fd54-82d2-4179-a612-491b68322489
+====== Nvidia MDL ======
+  * to define physically based material
+  * store specification for material exchange
+  * render-algorithm agnostic
+  * designed for high performance on GPU
+====== Nvidia Cuda programming ======
+  * video info: How CUDA Programming Works
+    * https://www.nvidia.com/en-us/on-demand/session/gtcspring22-s41487/?playlistId=playList-87118008-d10b-42f9-8c57-a50bbf006662
+    * the cube programming is designed based on how the GPU hardware works, and GPU also designed how GPU normally programms for best performance
+    * Cuda programming is:
+      * each thread has its thread ID, that its id determine which block of data it works on, and all threads finish the data together at the same data.
+      * optimize how code use memory can be important to fit more thing in the fixed size memory by better arrangement and swapping thing in memory blocks.
+====== CV-Cuda ======
+  * CV cuda: computer vision with cuda.
+====== Nivdia RTX stack ======
+  * 1st Gen: VkRay, DXR, DLSS1
+  * 2nd Gen:
+    * real-time denoise: spatial denoise
+    * caustics
+    * RTXDI: raytrace direct illumination, casting shadows from all lights, emissive surface
+    * RTXGI: real-time multiple bounce indirect lighting
+    * Reflex
+    * DLSS2: deep learning super resolution, AI generat pixel
+  * 3rd Gen:
+    * Displaced micro-meshes
+    * 2D SGM optical flow, shader execution reordering, real-time path tracing, opacity micro-maps,
+    * DLSS3: deep learning super resolution, AI frame generator
+====== Nvidia GPU architecture ======
+| core   ^ turing ^ ampere ^ ada |
+| shader | 16     | 40     | 90  |
+| RT     | 49     | 78     | 200 |
+| tensor | 130    | 320    | 1400 |
+| OFA    |        | 126    | 300  |
+====== Nvidia for AI ======
+  * large language model: enable single model to do various different task with one single model, context aware output. like text related, image related.
+  * NeMo LLM service, Prompt learning framework, to promp learn with pre-trained LLM for specific task.
+  * recommed system: like in shopping, social network