#! https://zhuanlan.zhihu.com/p/521035797 TEST

[TOC]

Markdown Cheatsheet

Heading 1

Markup :  # Heading 1 #

-OR-

Markup :  ============= (below H1 text)

Heading 2

Markup :  ## Heading 2 ##

-OR-

Markup: --------------- (below H2 text)

Heading 3

Markup :  ### Heading 3 ###

Heading 4

Markup :  #### Heading 4 ####

Common text

Markup :  Common text

Emphasized text

Markup :  _Emphasized text_ or *Emphasized text*

~~Strikethrough text~~

Markup :  ~~Strikethrough text~~

Strong text

Markup :  __Strong text__ or **Strong text**

Strong emphasized text

Markup :  ___Strong emphasized text___ or ***Strong emphasized text***

Named Link and http://www.google.fr/ or http://example.com/

Markup :  [Named Link](http://www.google.fr/ "Named link title") and http://www.google.fr/ or <http://example.com/>

heading-1

Markup: [heading-1](#heading-1 "Goto heading-1")

Table, like this one :

First Header	Second Header
Content Cell	Content Cell
Content Cell	Content Cell

First Header  | Second Header
------------- | -------------
Content Cell  | Content Cell
Content Cell  | Content Cell

Adding a pipe | in a cell :

First Header	Second Header
Content Cell	Content Cell
Content Cell	\|

First Header  | Second Header
------------- | -------------
Content Cell  | Content Cell
Content Cell  |  \| 

Left, right and center aligned table

Left aligned Header	Right aligned Header	Center aligned Header
Content Cell	Content Cell	Content Cell
Content Cell	Content Cell	Content Cell

Left aligned Header | Right aligned Header | Center aligned Header
| :--- | ---: | :---:
Content Cell  | Content Cell | Content Cell
Content Cell  | Content Cell | Content Cell

code()

Markup :  `code()`

    var specificLanguage_code = 
    {
        "data": {
            "lookedUpPlatform": 1,
            "query": "Kasabian+Test+Transmission",
            "lookedUpItem": {
                "name": "Test Transmission",
                "artist": "Kasabian",
                "album": "Kasabian",
                "picture": null,
                "link": "http://open.spotify.com/track/5jhJur5n4fasblLSCOcrTp"
            }
        }
    }

Markup : ```javascript
         ```

Bullet list
- Nested bullet
  - Sub-nested bullet etc
Bullet list item 2

 Markup : * Bullet list
              * Nested bullet
                  * Sub-nested bullet etc
          * Bullet list item 2

-OR-

 Markup : - Bullet list
              - Nested bullet
                  - Sub-nested bullet etc
          - Bullet list item 2 

A numbered list
1. A nested numbered list
2. Which is numbered
Which is numbered

 Markup : 1. A numbered list
              1. A nested numbered list
              2. Which is numbered
          2. Which is numbered

An uncompleted task
A completed task

 Markup : - [ ] An uncompleted task
          - [x] A completed task

An uncompleted task
- A subtask

 Markup : - [ ] An uncompleted task
              - [ ] A subtask

Blockquote

Nested blockquote

Markup :  > Blockquote
          >> Nested Blockquote

Horizontal line :

Markup :  - - - -

Image with alt :

Foldable text:

Title 1

Content 1 Content 1 Content 1 Content 1 Content 1

Title 2

Content 2 Content 2 Content 2 Content 2 Content 2

Markup : <details>
           <summary>Title 1</summary>
           <p>Content 1 Content 1 Content 1 Content 1 Content 1</p>
         </details>

<h3>HTML</h3>
<p> Some HTML code here </p>

Link to a specific part of the page:

Go To TOP

Markup : [text goes here](#section_name)
          section_title<a name="section_name"></a>    

Hotkey:

⌘F

⇧⌘F

Markup : <kbd>⌘F</kbd>

Hotkey list:

Key	Symbol
Option	⌥
Control	⌃
Command	⌘
Shift	⇧
Caps Lock	⇪
Tab	⇥
Esc	⎋
Power	⌽
Return	↩
Delete	⌫
Up	↑
Down	↓
Left	←
Right	→

Emoji:

:exclamation: Use emoji icons to enhance text. :+1: Look up emoji codes at emoji-cheat-sheet.com

Markup : Code appears between colons :EMOJICODE:

Videos

视频介绍

大气河是天空中巨大的水汽河流，每条河流的水量都比亚马逊河的还要多。它们一方面为美国西部提供了关键的降水来源，但另一方面，这些巨大的强风暴也会导致灾难性的洪灾和暴雪。NVIDIA 创建了 Physics-ML 模型，该模型可以模拟全球天气模式的动态变化，以超乎想象的速度和准确性预测大气河等极端天气事件。此 GPU 加速的 AI 数字孪生模型名为 FourCastNet，由傅里叶神经算子提供动力支持，基于 10 TB 的地球系统数据进行训练。依托这些数据，以及 NVIDIA Modulus 和 Omniverse，我们能够提前一周预测灾难性大气河的精确路线。在一个 NVIDIA GPU 的助力下，FourCastNet 只需几分之一秒即可完成预测

Images

H100FLOPS

H100 采用风冷和液冷设计，是首个实现性能扩展至 700 瓦的 GPU。在过去六年里，通过 Pascal、Volta、Ampere 和现在的 Hopper 架构，我们相继开发了使用 FP32、FP16和现在的 FP8 进行训练的技术。在 AI 处理方面，Hopper H100 FP8 的 4 PetaFLOPS 性能是 Ampere A100 FP16 的 6 倍，这是一次巨大的代际飞跃。

H100TransformerEngine

Transformer 无疑是最重要的深度学习模型。Hopper 引入了Transformer 引擎，Hopper Transformer 引擎将新的 Tensor Core 与能使用 FP8 和 FP16 数字格式的软件结合，动态处理 Transformer 网络的各个层，Transformer 模型训练时间可从数周缩短至数天。

H100MultiInstance

在云计算方面，多租户基础架构能够直接转化为收益和服务成本。一项服务可将 H100 划分为多达 7 个实例，Ampere 也可实现此操作。但是，Hopper 新增了完整的每实例隔离和每实例 IO 虚拟化，便于支持云端的多租户。H100 能够托管七个云租户，而 A100 仅能托管一个。每个 H100 实例的性能相当于两个完整的 T4 GPU（我们非常热门的云推理 GPU）。

H100ConfidentialComputing

Real World	Digital Twin(Domain Randomization)

GIF:

NvidiaDriving3

MathJax

Inline

Equation aa: $-b \pm \sqrt{b^2 - 4ac} \over 2a$, b: $x = a_0 + \frac{1}{a_1 + \frac{1}{a_2 + \frac{1}{a_3 + a_4}}}$, c: $\forall x \in X, \quad \exists y \leq \epsilon$

Block equation:

\[-b \pm \sqrt{b^2 - 4ac} \over 2a x = a_0 + \frac{1}{a_1 + \frac{1}{a_2 + \frac{1}{a_3 + a_4}}} \forall x \in X, \quad \exists y \leq \epsilon\]