We are evaluating the performance of IPP mpeg2 decoder source code library on some of our target server. For this we are using IPP 7.1 sample source code for mpeg2 decoder compiled with IPP 7.1 core library. We are getting quiet good performance on one of our target hardware. But we are facing one issue that if we run the same source code on a target hardware with better configuration, the performance of the Mpeg2 decoder is degrading. The specification of the old and new hardware is as follows:
Old Hardware specification: Processor: 2 x Intel Xeon x5650 processor(2.66Ghz/6core/12 MB), RAM: 12GB DDR3, OS: Windows server 2003 standard with SP1
New Hardware Specification:Processor: Xeon E5-2690 Processor Kit[N8101-552F] x 2(2.90 Ghz/8core/20 MB), RAM: 32GB DDR3, OS: Windows server 2003 standard with SP1.
From above you can see that new hardware has very good configuration and even Xenon E5-2650 supports the AVX instruction set also, even though the performance is degrading on this new hardware.
IPP static libraries are used to build the binary. ippInit() is called befrore calling any other function of the IPP library(also checked with ippInitCpu and ippEnableCpu functions but behaviour does not change).
Kindly provide some direction to get the root cause of this issue.
Just for information (may be helpful for anyone):
Root cause of the above mentioned issue was identified as the different power option setting on two hardwares.
There is a setting for power option in [Control panel] → [Power option] of Windows. In new hardware, when it was set to “Balance” (Default setting), there is an impact on the operation speed. In “Balance” setting, the operating frequency of CPU corresponding to the operation status of hardware has changed in the range of 5%~100%, therefore the processing speed of program is also affected. The processing speed improved when the settings were changed to “High performance”. When comparing the processing speed of new hardware and old hardware at the time of “High performance”, processing speed is faster in case of new hardware in any of the conversion patterns.
Further processing speed was checked at the time of “Balance” & “High performance” in old hardware also, but there was not much difference. Though it is not clear whether it is due to OS or due to CPU, but in case of old hardware, it seems that there is not much impact of Power option.