F·18COMPUTER SCIENCE2025.02.051 MIN READ
CPU vs GPU: One Einstein vs 10,000 Elementary Students (Deep Dive)
CPU vs GPU: 아인슈타인 1명 vs 초등학생 10,000명 (완전정복)
Why did AI and deep learning abandon CPUs for GPUs? From ALU architecture to CUDA memory hierarchy and generative AI principles.
c
codemapo
INTERDISCIPLINARY DEV · SEOUL
Comparison Table
| Feature | CPU (Central Processing Unit) | GPU (Graphics Processing Unit) |
|---|
| Core Count | Few (4 ~ 64) | Massive (Thousands ~ Tens of thousands) |
| Per-Core Performance | Very High (Genius) | Low (Student) |
| Primary Goal | Low Latency (Fast response) | High Throughput (Mass processing) |
| Control Logic | Complex (Branch prediction, OOO execution) | Simple |
| Cache Memory | Large (L1/L2/L3) | Small (Shared memory) |
| Instruction Model | SISD (or limited SIMD) | SIMT (Single Instruction Multiple Threads) |
| Specialized Tasks | OS, web servers, complex logic | Graphics rendering, deep learning, crypto mining |
| Analogy | Ferrari, Einstein | Bus, Army of students |
| Price (Consumer) | $150 ~ $1,000 | $300 ~ $2,000 |
| Price (Server) | $1,000 ~ $10,000 | $3,000 ~ $40,000 (H100) |