Energy Efficient computing

Motivation:

1. Pack more transistors on a chip
   1. Too many transistors, core becomes too hot
2. Pack more cores on a chip
   1. The socket gets too hot
3. Around 1997, cooling was required for chip
4. How to dissipate power??

Energy efficiency required in

• Mobile computing (high powered screen, need high battery life)
• Enterprise computing

Heterogenous computing: big.LITTLE

• Fast processors
  • Run high performance tasks
• Slow processors
  • Run small, sequential background tasks

Nowadays hardware itself can migrate tasks to another core when running low performance tasks.

Modern HPC Infrastructure

Data Centers

A data center is a facility that centralizes an organization’s shared IT operations and equipment for the purposes of storing, processing, and disseminating data and applications

Achieves heterogenous computing with different server generations

Cloud Computing

Delivery of computing services (including servers, storage, databases, networks, software etc.) over the internet.

• Can be physically located on a data center
• Or owned and operated remotely by a 3rd party provider

How to reduce Energy Consumption

1. Move less data
   1. Reduce data transfer to/fro memory
   2. Exploit locality
   3. Compression
2. Specialized processing:
   1. Compute less (Avoid parallelizing)
   2. CPU-like cores + GPU-like cores