Abstract
Benefits
Optimised for Parallel Computing
- We can calculate pixels in batches or parallel
- Allow training Deep Learning Models that perform tons Matrix Multiplication on dataset
Cons
Hard to program
- That is why Nvidia’s CUDA Toolkit is good news. It allows us to program GPU, giving that flexibility into GPU
Comparison with CPU
- CPU one core is way more power, and able to handle complication logics like Branching
- However, a lot of the real world application needs to run Instruction in a sequential manner
- GPU shines when we need Parallelism (并行性)
