See this ancient bug from 2005 http://sourceforge.net/p/iometer/bugs/48/.
While running iometer on my i7-4770k system it started spewing the following errors:
*** Performance counter percentage is less than zero: -0.00202604
....
Browswing the source code shows that IOMeter calculates timing using an interval based on QueryPerformanceCounter() with a fixed CPU frequency as the denominator.
This is inaccurate for CPUs that run at multiple frequencies over time. On my system the frequency can rapidly switch between 800Mhz to over 4Ghz. This will naturally make the reported results highly inaccurate, especially for things like latency measurements.
This is very surprising considering how ubiquitous and relied upon IOMeter seems to be across the different review sites.
While running iometer on my i7-4770k system it started spewing the following errors:
*** Performance counter percentage is less than zero: -0.00202604
....
Browswing the source code shows that IOMeter calculates timing using an interval based on QueryPerformanceCounter() with a fixed CPU frequency as the denominator.
This is inaccurate for CPUs that run at multiple frequencies over time. On my system the frequency can rapidly switch between 800Mhz to over 4Ghz. This will naturally make the reported results highly inaccurate, especially for things like latency measurements.
This is very surprising considering how ubiquitous and relied upon IOMeter seems to be across the different review sites.
Last edited:
