Welcome Stranger to OCC!Login | Register

Serious Statistics Review

   -   
» Discuss this article (1)

DirectX 11 – Thebes – Karnak:

Finally at the end of these API tests! Then the v-sync tests start, but those will be less repetitious so I will not have to worry about running out of ideas to write. First up, the course graphs.

 

It is kind of neat to look at the graphs and see confirmation that I was successful in repeating the course between multiple runs.

Anyway, here are the averages:

Average Frame Time (ms) Framerate (FPS)
Run 1 6.401436 156.2134
Run 2 6.467237 154.6257
Run 3 6.379428 156.7525

Just barely behind Vulkan for these runs. Time to see where the percentiles fall.

Percentiles 0.1% 1% 99% 99.9%
Run 1 4.380 ms (228.32 FPS) 4.652 ms (214.94 FPS) 11.215 ms (89.17 FPS) 14.137 ms (70.74 FPS)
Run 2 4.284 ms (233.42 FPS) 4.616 ms (216.65 FPS) 11.876 ms (84.21 FPS) 17.028 ms (58.73 FPS)
Run 3 4.357 ms (229.51 FPS) 4.664 ms (214.41 FPS) 11.541 ms (86.65 FPS) 14.268 ms (70.09 FPS)

It is close, but Vulkan is still ahead here, at least when looking at the 99% data. Based on this data, it is not surprising how many cells are going to be empty in the next one.

  16.667 ms / 60 FPS 20.000 ms / 50 FPS 33.333 ms / 30 FPS
Run 1 0.0170776% -- --
Run 2 0.10997542% 0.05175314% --
Run 3 0.01913754% -- --

Time for the QQ plots.

 

The difference data is up next.

Diff Percentile 0.1% 1% 99% 99.9%
Run 1 -4.681659 ms -1.867570 ms 1.936570 ms 4.852611 ms
Run 2 -7.558144 ms -2.191840 ms 2.287280 ms 7.638020 ms
Run 3 -4.769896 ms -1.962740 ms 2.012740 ms 4.969636 ms

That sporadic behavior we can see in the second run's course graph shows up here in the extremes, but the 1% and 99% values do not seem too far off from the others.

By the way, something I have been neglecting to mention is that I have also been looking at the ratio of frame times increasing and decreasing in case there is an imbalance, but none of the data has exhibited this. It has all been at 50/50 or 51/49, which is not very interesting and thus why I have not mentioned it.

 

 

 

 

All that is left are the display time graphs, which still exhibit the effect of the double buffering, and then we can get to the v-sync tests.




Related Products
Random Pic
© 2001-2017 Overclockers Club ® Privacy Policy
Elapsed: 0.0339879990   (xlweb1)