COMPUTEX 2023 NVIDIA Keynote

Written by Morris Fan @TaiNEX 2, Taipei, May. 31, 2023

Compare the old vision with the new, Computer rail tracing.

The five years.
Showing the RTX4060 Ti.
- Use AI to predict seven other pixels, so that we can save power consumption.
New launch for NVIDIA ACE FOR GAMES遊戲角色客製化生成模型，語音、聊天機器人、NVIDIA NeMo客製化、邊緣裝置到雲端皆可部屬。

Convai, the video game the user can design the plot by talking to the character.

One of the most important contributions is the importance of software, installed base. The thinking of the computer industry.

Two fundamental trends:

摩爾定律
the trend has ended.

Accelerated computing(domain sustained) and generated AI are the new trends. Since the CPU is flexible, everyone can write easy code, however, in the new 2 trends, we need to learn from the top down.

Each field needs its stack.

New computing model, 4M developers, 3000+ apps, 40M downloads, 15000+ start-ups, 40k companies. GPU servers and data centers are more important than servers. Dense computers not big computers.
Data centers are power-limited, so we use accelerated computing to achieve higher performance.

"If your goal is to get the work done, you will not care how. The more you buy, the more you save."

Without an algorithm, accelerated computing is nothing.

Software ecosystem

Rich of applications are needed for more utilizations.
life cycle optimization, it benefits everybody.
Reach the tipping point for Generated AI.
Every time you release the product, the demand keeps increasing. H100量產。

Reinvented everything, CUDNN, TENSOR CORE GPU, SXM, NVLINK(GPU TO GPU to extend the scale), HGX, DGX(Giant GPU Computer), MELLANOX, MAGNUM IO, NVIDIA AI.

Giant leaps every two years. Texts to proteins, images to captions. Introducing Grass Hopper super chip.
Grasshopper has made every data center have the ability. 3D video Facetime. The communications will not only be transferred but be regenerated. Redefine communication.

Announcing MGX: Used for Servers. Data centers, tightly-coupled processing,

Before generating AI, we need to separate data centers into 2 parts since the costs are too high to afford. However, by generating AI, we can connect all the data centers. NVIDIA AI ENTERPRISE, make sure the data is safely secured.

Digitized

Robot AI will be highly revolutionized.
The robot is grounded physically.
ChatGPT, Reinforcement Learn Highly Feedback. Build the product digitally before manufacturing it.
Use digital in design and corporate with commercial companies. From stationary robotics to mobile.
ISAAC AMR reduces cost and time. Onmimers and AI working together.
- Chip / Electric battery / Electric motors.