afaa77ec984f1580fc321aaa0ceacfa9def039c96782fbbbba3e21176c4b9926

1. **Parallelism:** GPUs have many cores optimized for parallel matrix operations, so they usually run large neural network inference much faster than CPUs.

2. **Cost/latency profile:** CPUs are often simpler and cheaper for small models or low request volume, while GPUs are more efficient for large models, batching, and high throughput.