Accelerator
A specialized piece of hardware (GPU, TPU, IPU, etc.) optimized for specific computations like neural networks.
Think of it like
A specialized tool for a specific job — like a pasta maker, not a general-purpose knife.
Example
AWS Trainium for training, Inferentia for inference, Nvidia H100 for both.
How it actually works
Accelerators are faster and more power-efficient than CPUs for their specific task. But they're expensive, hard to program, and inflexible — if you need a different computation, a CPU might be simpler. The trend is toward domain-specific accelerators (for LLMs, vision, etc.) instead of general-purpose GPUs.
For product teams
Accelerator choice affects cost, speed, and lock-in. Choose wisely.
For engineers
Profile your workload (compute pattern, memory bandwidth). Choose accelerator that matches it.
Related
- Hardware trends.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome