Here's a couple screenies showing CPU usage along with AB graphs.
The problem with looking at the CPU graphs to determine a bottleneck is that the individual core-usage graphs are basically inaccurate. Although as you can see, the game appears to happily use 6 of the cores to a fairly significant degree, that unfortunately does not have to be true at all, because Windows is interjecting it's own thread management into the equation.
Note that when you switch over to the cumulative usage graphs, you can see that the total usage taps out at exactly 50% for at couple of very brief points.
This evidence unfortunately leaves one with two possible conclusions:
1) That this game is actually able to utilize up to half of my eight 'cores' (counting hyperthreading), hence the maxing out right at 50% of total CPU usage in the cumulative graph. Because the moment(s) at which this happens are so brief and infrequent, it implies that there should basically be next to no cpu-bottlenecking in this test on my rig, at the frequencies I'm running, at the settings I'm running this test at, OR,
2) This game actually could make use of > than 50% of my 'cores' if the load was high enough. In which case, the individual cpu core usage graphs actually are accurate, in which case in theory my total usage COULD go over 50%, it's just so happens to be maxing out at 50% on this test because of a GPU bottleneck at these settings. In which case, there's definitely not CPU-bottlenecking happening here on my rig.
My prior experience *suggests* to me that #1 is closer to the 'truth' of this situation. So I'm going to run another test here at a much lower resolution, and see if we can't push CPU usage up over 50% by dramatically increasing the framerate. Results to follow shortly.
Edit, okay, I've added two more shots to show usage at 1024x768 and 4xAF (trying to make the test easy on the GPU w/o changing relative CPU load).
Notice that our FPS is up considerably, but our GPU usage also fluctuates considerably more. Notice also that the individual CPU usage graph (shot 3) shows there's solid action on 6 cores, yet the overall usage graph (shot 4) very much stay's 'huddled' at 50%. Note also that the 'grouping' around 50% is much stronger than it was at the higher settings.
Also of interest is the fact that there are few brief moments where the usage does go over 50%, but it's hard to know whether this suggests that Metro is actually capable of using >4 of my cores (i.e. using the hyperthreading).
My guess is that it is NOT, that the bumps above 50% usage are actual just Windows doing something.
Thus, overall the evidence suggests to me that on *my* rig, there is only a *very tiny* bit of CPU bottlenecking happening at the settings chosen on this test.
However, it's entirely possible that someone who's rig isn't balanced in the way mine is (i.e. if they had a less powerful CPU + equal or more powerful gfx setup), that they *might* run into a CPU bottleneck that's actually somewhat significant on this test, at the chosen settings.
Oh, and the evidence very much also strongly suggests that this game can fully utilize up to four cores of one's CPU, but can't use the additional 'cores' added by hyperthreading. Thus, dual cores may well be at a disadvantage in this benchmark.
Thanks for reading, hope at least someone 'learned something' from these tests Edited by brettjv - 9/9/10 at 10:28am