RFC: Polishing Up The Result Graphs

[phoronix]

Phoronix: RFC: Polishing Up The Result Graphs

It's time for another call-out for requests for comments on how to better enhance the appearance/presentation of information on our benchmark result graphs seen on Phoronix as well as OpenBenchmarking.org and LinuxBenchmarking.com, etc...

http://www.phoronix.com/scan.php?pag...ning-Up-Graphs

[Xenopathic]

Please increase the font size on the graph number explanations, e.g. "Frames per second, More is better". At size 7 it's near impossible to read on some graphs.

Also, perhaps when calculating the maximum frame times, exclude the first second or so? Many games and benchmarks do a lot of loading in that time, so will naturally be slow. On a similar vein, it would also be really cool to show the minimum FPS (excluding first second) on the standard FPS graphs, even if it's without numbers, just to show if a certain driver has stutter problems.

[Zplay]

Hello,

When doing power usage benchmark, could you tell the total power usage ? It's only "average * total time" and you already have both informations.

[Michael]

Also, perhaps when calculating the maximum frame times, exclude the first second or so? Many games and benchmarks do a lot of loading in that time, so will naturally be slow. On a similar vein, it would also be really cool to show the minimum FPS (excluding first second) on the standard FPS graphs, even if it's without numbers, just to show if a certain driver has stutter problems.

IIRC in the code, I do have it shift/pop the start/end of the tests to avoid issues with the min/max. Only for tests that expose min/max information do I have something to report, depends on the game engine.

[GabrielYYZ]

I have two tips: 1. for the "bar" graphs (i.e. the ones showing fps in the example), try to use red for amd/green for nvidia and, if possible, different shades according to how well they perform. 2. for the line graphs (i.e. the ones showing power comsuption), avoid light colors (e.g. in the unigine example, drop the light blue, grey and "capuccino" brown), they're very hard to distinguish from the white background.

PS: I'll clone the git repo later and see if it can be done/i can do it. I just have to prepare for php first (all those $$$ )

[My8th]

+1 for increase in text size, I can make it out but just barely.

[Michael]

I have two tips: 1. for the "bar" graphs (i.e. the ones showing fps in the example), try to use red for amd/green for nvidia and, if possible, different shades according to how well they perform. 2. for the line graphs (i.e. the ones showing power comsuption), avoid light colors (e.g. in the unigine example, drop the light blue, grey and "capuccino" brown), they're very hard to distinguish from the white background.

PS: I'll clone the git repo later and see if it can be done/i can do it. I just have to prepare for php first (all those $$$ )

It would be possible to do coloring for companies quite easily.... when you get to the code, look at the get_special_color or get paint color function, inside pts_Graph/pts_Graph.php. If you were to check the first argument of what's being passed there against like Radeon, GeForce, Intel, etc, then pick a color from that if that color hasn't been picked, that should work too.

The line graph color picking is similarly derived from that function.

Have had a hell of a time with getting color picking right over the years after trying a variety of automated ways.

[W.Irrkopf]

When a card cannot be tested due to testfailure still include it in the results graph - just show it as failed

[JodiTheTigger]

1) Make the font for "less is better/more is better" much larger. It what I look for first to understand the bar graphs
2) Colour the best and worst bar graph result different colours (blue for best, red for worst, grey as normal. Avoid red/green colours due to colourblind issues)
3) Show frame period instead of FPS. Show min/max/average/sd on the same graph if possible.
4) Show frame period jitter and max period, as that's all I actually care about. I want to know the slowest my fast computer will go, and how much the frame period swings around.
5) Show obvious markers for 33.3/16.6/11.1ms (30/60/90fps) on graphs.

[Michael]

3) Show frame period instead of FPS. Show min/max/average/sd on the same graph if possible.
4) Show frame period jitter and max period, as that's all I actually care about. I want to know the slowest my fast computer will go, and how much the frame period swings around.

Like ~95% of the games don't expose this information (and libframetime doesn't work reliably) so the info is shown for tests/games that do support it like OpenArena, but sadly the others don't expose the per-frame time or even the min/max.