Researchers at the MIT Lincoln Laboratory Supercomputing Center published findings from an ongoing survey tracking the evolution of commercial AI accelerators. The survey, known as the Lincoln AI Computing Survey, has evaluated over 120 systems to date.
Tracking AI Accelerators
Since 2018, a team from the Lincoln Laboratory Supercomputing Center (LLSC) has conducted the Lincoln AI Computing Survey to summarize commercial AI accelerators. Albert Reuther, a staff member at the LLSC, stated that the survey was initiated after receiving questions from government sponsors regarding a sharp rise in research and commercial AI accelerators.
The survey evaluates technologies currently on the market to help identify the best accelerators for specific needs. AI accelerator types include central processing units, graphics processing units, application-specific integrated circuits, field-programmable gate arrays, and dataflow accelerators.
Expanding Scope and Metrics
While the first paper in the series studied 57 accelerators, the latest paper examined more than 120 accelerators. The research team primarily compares accelerators using peak performance and peak power metrics, sorting them by whether they reside on a chip, card, or system.
Data for the papers are drawn from public sources, supplemented by daily news and citation searches run by Reuther. In addition to performance and power comparisons, each paper explores a new aspect of the field, such as transistor designs, lower numerical precision, and architectural choices.
Key facts and where they come from
- The Lincoln AI Computing Survey has been conducted by an LLSC team since 2018.
Since 2018, team from the Lincoln Laboratory Supercomputing Center (LLSC) has been conducting the Lincoln AI Computing Survey
- The first survey paper studied 57 accelerators, while the latest examined over 120.
The first paper in their series studied 57 accelerators, while the latest one looked at more than 120 accelerators.