Min / Max FPS Comes To Test Results

[Kivada]

most of the tests don't support constant FPS dumping by the upstream game/demo. Though PTS already supports frame showing, eg demo in vdrift-FPS-monitor

The question is why don't you do what the Windows game reviewers have been doing for a few years now in their completely closed source world? Nobody cares what the max FPS of any game is, the frame latency, the average FPS*, and minimum FPS* are what actually matter.

* you have to round out the very high and low anomalies as well, sometimes the game's loading registers FPS far beyond the capability of the GPU for a second or 2, sometimes they spit out 0 when the game plays at well over 60 FPS.

[smitty3268]

Using that current little white overlay that shows the deviation to show the min/max would be a smaller change from the current graphs. Then you'd just have to figure out a new way of showing the deviations - perhaps by adding dotted lines. Or you could extend the current bar graph with dotted lines to show the min/max i guess.

Obviously showing some kind of line graph to show the min/avg/max over time shows the most information. They're also a little harder to just glance at and require more thought to read them - especially if you add more than a couple options at once, like you sometimes do for Phoronix. The box and whiskers type is in between - most people are less familiar with them, so it takes longer to look at them, but they can at least show the same number of options that your current bar graphs do.

[GreatEmerald]

The posibility to make a Box whisker Plot with the data pts records would be nice and a way more accurate display the data. Min and Max frame say les than nothing about the Video output ( they can ocure once in a run with half the value or duble the value of the average reading but no one would recognise them, with median and the 25/75% percentile and may be whiskers and ausreisser would display the values better). if you can't implement such a calculation yourselve you can find such an implementation in oss statistical tools like R and others.

Yea, I was also thinking about a Box-and-whisker plot, but with the line being the average, the box being the average's deviation between tests and the whiskers being the FPS range size or standard deviation.

[Michael]

Michael, I think this can be worked around by means of dumping constant FPS graph!
This could work for every test that supports Min/Max.

Then, one could definitely see the anomalies.

Also, another approach improving current method could be to provide extra info in form of [minfps/percentage_of_time]...[average_fps]...[maxfps/percentage_of_time].

So, the graph would read:

[MIN:14/21%]...[AVG:30]...[MAX:105/1%]
instead of:
MIN:14/MAX:105

which would give much much more in-depth view on framerate distribution without bulky FPS graphs; although FPSgraph would also be awesome thing.

most of the tests don't support constant FPS dumping by the upstream game/demo. Though PTS already supports frame showing, eg demo in vdrift-FPS-monitor

[brosis]

Also: note that there could be anomalies. I would be happy to play a game that sticks around 50fps consistenly, and hits a couple frames at 3 fps at some point. But who would endure a game you can play at 50fps, with a stdev of 35? That game would consistently be in the vicinity of 50-35=15 FPS.

Michael, I think this can be worked around by means of dumping constant FPS graph!
This could work for every test that supports Min/Max.

Then, one could definitely see the anomalies.

Also, another approach improving current method could be to provide extra info in form of [minfps/percentage_of_time]...[average_fps]...[maxfps/percentage_of_time].

So, the graph would read:

[MIN:14/21%]...[AVG:30]...[MAX:105/1%]
instead of:
MIN:14/MAX:105

which would give much much more in-depth view on framerate distribution without bulky FPS graphs; although FPSgraph would also be awesome thing.

[mendieta]

Also: note that there could be anomalies. I would be happy to play a game that sticks around 50fps consistenly, and hits a couple frames at 3 fps at some point. But who would endure a game you can play at 50fps, with a stdev of 35? That game would consistently be in the vicinity of 50-35=15 FPS.

[mendieta]

So I guess it varies and some programs will just give you a sort of end result while others will give you a fps number every [whatever], right?

Yes, that makes the min arbitrary, to some extent, though useful to display. I think, more meaningful than min/max, would be to provide the standard deviation of frame rates, which gives an idea of the variability around the average. You'd like that number to be stable, meaning the average much larger than the standard deviation.

[YoungManKlaus]

The program's own output.

So I guess it varies and some programs will just give you a sort of end result while others will give you a fps number every [whatever], right?

[_ONH_]

The posibility to make a Box whisker Plot with the data pts records would be nice and a way more accurate display the data. Min and Max frame say les than nothing about the Video output ( they can ocure once in a run with half the value or duble the value of the average reading but no one would recognise them, with median and the 25/75% percentile and may be whiskers and ausreisser would display the values better). if you can't implement such a calculation yourselve you can find such an implementation in oss statistical tools like R and others.

[curaga]

One question regarding PTS ... are you measuring frame time and then converting it to FPS or are you relying on some output of the program (which may be only updated every second)?

The program's own output.