NVIDA AI Talk
GPU Accelerated
Good for Machine Learning/Deep learning
Company Focus:
- Gaming
- Artificial Intelligence
- Autonomous driving
Hong Kong AI Center:
- Industry 4.0
- Smart City (IOT)
- Health Care
- FinTech
NVAITC Focus:
- Accelerated Computing
- High performance computing
- Advanced Visulization
- Large scale
- Deep Learning
- Computational Intelligente
HPC - High Performance Computing
- Internet Service
- Cloud Services
- Amazon P2
- Super Computing
Energy Efficiency - Energy cost money, 2020 50 G Flops per watts
How GPU Acceleration Works
GPU(5% code) + CPU
Tensor Core (For Deep Learning)
D = AB + C (Bias)
Two tensor core deal with one Full Warp 16*16
VOLTA: Independent Thread scheduling
