Use CPU(KL) to run FFT of Intel IPP takes longer than CPU(HL)

[Tsai__Richard]

Customer uses related Fast Fourier Transform(FFT) part of Intel Integrated Performance Primitives(IPP) to get calculated result (Please refer attached file).
Why use a new generation of CPU (Kaby Lake)) to run FFT of Intel IPP will takes longer than the previous generation of CPU (Haswell)? Could you expain this phenomenon? Thanks.

[Abhinav_S_Intel]

Thank you for reaching out to us. Can you please share some sample code to investigate this issue on our side?

[Tsai__Richard]

Could you tell in detail what kind of some sample code sharing with you?

[Abhinav_S_Intel]

Any sample code which is shareable and can reproduce Fast Fourier Transform(FFT) performance issue observed at your end. 

[Tsai__Richard]

Our AE keeps on asking customer's sharing sample code quickly. 

[Gennady_F_Intel]

Richard, are there any samples from your side?

[Tsai__Richard]

Keep on urging AE to ask for customer's providing sample code.

[Pavel_B_Intel1]

Richard,

which FFT do you mean: signal processing (1D) or image processing (2D)?

In any way I recommend to use one OS for perf measurement. win7 vs win10 - there are can be some patches for security vulnerability affect performance.

Pavel