Why do GPUs suit neural network training better than CPUs?
Why D is the answer
They run many parallel matrix operations
Training is mostly parallel matrix math, which is what GPU architecture does well.
In detail
- GPUs run many parallel matrix operations.
- Serial tasks can still suit CPUs.
Use case
An illustrative scenario, not a real incident.
Training multiplies huge matrices, and a GPU runs those operations in parallel across many cores.
See it in different ways
flowchart LR A["CPU: few fast cores"] --> B["Serial control work"] C["GPU: many parallel cores"] --> D["Matrix math"]
mindmap
root((AI))
Machine learning
Deep learning
Generative AI
Drivers
GPUs
Data
Algorithms
Stages
Training
Inferenceflowchart TD
Q{"How does the work divide?"}
Q -->|"Many parallel operations"| A0["GPU"]
Q -->|"Serial steps"| A1["CPU"]Key point: GPUs excel at parallel matrix math.
Why the other options are wrong
- A. They have a higher single-thread clock speed
- Their clock speed is not higher.
- B. They use less total power per server
- That is not the reason.
- C. They avoid the need for software
- That is not the reason.