Chapter
Predicting Branches in Computer Processing
10:18 - 16:16 (05:57)
A window of instructions needs to be predicted with 99% accuracy in order for it to be effective. 20 years ago, branches were predicted simply by recording the previous direction, but now it requires three to four orders of magnitude more bits.
Clips
The amount of parallelism found in modern programs written by humans is generally low due to the latency of operations, requiring extremely accurate branch prediction for effective execution.
10:18 - 12:27 (02:08)
Summary
The amount of parallelism found in modern programs written by humans is generally low due to the latency of operations, requiring extremely accurate branch prediction for effective execution.
ChapterPredicting Branches in Computer Processing
EpisodeJim Keller: Moore’s Law, Microprocessors, Abstractions, and First Principles
PodcastLex Fridman Podcast
In this episode, the host explains how modern computers can accurately predict branches during a computation, resulting in more efficient processing speeds and better performance.
12:27 - 13:36 (01:08)
Summary
In this episode, the host explains how modern computers can accurately predict branches during a computation, resulting in more efficient processing speeds and better performance.
ChapterPredicting Branches in Computer Processing
EpisodeJim Keller: Moore’s Law, Microprocessors, Abstractions, and First Principles
PodcastLex Fridman Podcast
The difference in processing 50 instructions versus 500 can take three to four orders of magnitude in bits, making bit accuracy crucial in executing the correct path and avoiding wasted processing time.
13:36 - 16:16 (02:39)
Summary
The difference in processing 50 instructions versus 500 can take three to four orders of magnitude in bits, making bit accuracy crucial in executing the correct path and avoiding wasted processing time.