Glossary What is Tera Operations Per Second (TOPS)?
By null / 2 Jan 2023 / Topics: Generative AI Data and AI
Tera Operations Per Second (TOPS) is a measurement for how many trillions, or tera, of operations a Neural Processing Unit (NPU), more commonly referred to as an AI chip or accelerator, can perform per second. That number is then multiplied by the maximum frequency it can achieve.
TOPS is a common metric used to analyze an AI chip’s performance, but there are other datasets that should be taken into consideration alongside it. In general, higher TOPS a device achieves results in increased performance.
Using Tera Operations Per Second (TOPS) to measure performance
TOPS was first commonly referenced by the company Movidius, which was acquired by Intel in 2016. Movidius specialized in developing low-power machine vision processors for edge devices. They used TOPS as a metric to measure the performance of their chips, such as Myriad X, which at the time boasted 4 TOPS.1 Since 2017, TOPS has become a commonly used metric to describe the performance of AI processors and accelerators.
For businesses that are considering a hardware upgrade, Tera Operations Per Second will be an important metric to learn. Knowing how many TOPS an AI-ready device can handle will be crucial to know during a new device refresh or selection process.
But not everyone has the time or expertise to really understand what types of hardware or devices their business needs to succeed. With our deep knowledge of AI, hardware and software, Insight can help businesses make the most of their investment and select the right hardware with the appropriate TOPS capacity. Our team can also provide guidance on how to optimize the use of AI-ready hardware to achieve maximum performance and minimize costs.
Learn more about Tera Operations Per Second
1 Silicon Republic. (2017). Intel-Movidius Breakthrough Chip Makes Machines Think and See Like Humans.
