The Next-Generation Championship in Branch Prediction

Pushing the boundaries of energy-efficient and high-performance branch prediction

CBP-NG is the first micro-architectural championship to directly consider the energy consumed and latency required to compute branch predictions, in addition to (of course!) prediction accuracy. By more closely aligning the championship's incentives with real-world design constraints, we hope to enable the emergence of a new generation of branch prediction algorithms to tackle tomorrow's micro-architectural challenges!

We're accomplishing this with a new model which allows measuring the energy and latency costs of hardware designs implemented in C++. Participants will implement their prediction algorithms in this model, where they will be able to measure the latency and power-efficiency of their algorithms using the same tools used to score them. A subset of the championship traces will be available during the development period, but some traces will be kept secret and used only for final scoring after the submission period ends.

If you're interested in trying your hand at this year's branch prediction championship, we recommend you:

1. Join the CBP-NG Google group to ensure you receive important communications as the championship progresses. The Google group also serves as a forum for participants to ask/answer general questions (individual communications can be directed to cbpngteam AT gmail DOT com).

2. Familiarize yourself with the championship's model and traces.

3. Review the championship rules and submission instructions for your code and papers, including how your submission will be scored, and the championship schedule, below.

4. Finally, get started developing and testing your winning algorithm!

• Aaron Lindsay (Ampere), Chair

• Pierre Michaud (Inria)

• Rami Sheikh (ARM)

• Mahesh Madhav (Ampere)

• Alaa Alameldeen (Simon Fraser University)