Adjusted Peak Performance
From Wikipedia, the free encyclopedia
Adjusted Peak Performance (APP) is a metric introduced by the U.S. Department of Commerce's Bureau of Industry and Security to more accurately predict the suitability of a computing system to complex computational problems such as simulating nuclear weapons.
The (simplified) algorithm used to calculate APP consists of the following steps:
- Determine how many 64 bit (or better) floating point operations every processor in the system can perform per clock cycle (best case). This is FPO(i).
- Determine the clock frequency of every processor. This is F(i).
- Choose the weighting factor for each processor: 0.9 for vector processors and 0.3 for non-vector processors. This is W(i).
- Calculate the APP for the system as follows: APP = FPO(1) * F(1) * W(1) + ... + FPO(n) * F(n) * W(n).
The unit of measurement is Weighted TeraFLOPS (WT).
Notes:
- Processors without 64 bit (or better) floating point support have an FPO of zero.
- The current APP limit is 0.75 WT.
- Use the document referenced below (or an updated version) if you need to calculate APP for export purposes. It holds more details and a number of clarifications and exceptions. Also consult a lawyer who specializes in these things.
