Discussion in 'FRAPS General Discussion, Guides, and Tutorials' started by raffriff, Sep 15, 2012.
How can the left and right images be interpreted?
Left one (capturing desktop) has v-sync on or some other form of framerate limiting. You get an occasional late frame, always followed by a seemingly "fast" but really on-time frame.
Right one is averaging 60 fps but has terrible microstutter for some reason.
Thank you. That's a great help. The left one is a version of an emulation application called Final Burn Alpha which has an Windows Aero Theme's microstutter fix. The right one is an older version that doesn't.
Another example here from a different build. Any interpretation would be great. Towards the end was where I was actually playing the game. I think the user inputs may account for the spikes.
This is the same as the one on the right above but with some longer events (maybe from user interaction, or from loading textures etc) throwing off the vertical scale. BTW, you can adjust your charts to get a consistent scale for comparison - click in the left margin and play around a bit.
Looking at it again, maybe "terrible" is too strong. Frames times are oscillating between about 25 ms (40 virtual "fps" : 1/0.025) and probably half that (80 "fps"). This is common, especially if the system is nearing its limit. So how does it look to the human eye? My guess is that 60 fps with this amount of microstutter is roughly equivalent to a smooth 30-40 fps. Not *bad* looking, really - just not *quite* as smooth as true 60 fps.
Thank you again. The differences between all three are subtle in practice however it's for an online fighting games which are highly sensitive for execution / consistency, etc so micro stutter is a real problem more for the "feel" where you need to be able to execute inputs within sometimes only one frame of gameplay. The emulation requirements are basic - i.e. low load on the GPU and system overall. Instead it's from some aggravation with aero though I don't know the specifics. Again it's subtle but the build in the first image feels silky and consistent in comparison.
Do you know, offhand, how to ask MS Excel to calculate the amount of time spent above a certain frame rate threshold?
The last graph in the sequence. "Time spent above 33.3ms." Obviously, starting with a FRAPS frame time file , you could do an =SUM(A2-B2), etc, to arrive at the time between any two updates -- but how do you tell Excel to report the number of frames above 33.3ms in relation to the total number of frames?
Step 1: Make a "diff" column (formula = current time-previous time)
Step 2: Make a "diff-66.67" column for calculation purposes (I'm using 66.67, you can use 33.33)
Step 3: Make a "clamped" column that shows the previous value if > zero, or zero otherwise
Step 4: Get total time over 66.67
Step 5: Get total of all times (EDIT ...or just use the time of last frame...)
Step 6: Get the result as a percentage
Here is the data in Fraps Bench Viewer. You can see that 97% is about right. (I don't recall benchmarking Premiere Elements; it's possible I hit a hotkey by accident. There were probably long pauses with zero fps as I was thinking about some edit):
Thank you very much. Very useful.
I've just tried to drop in a couple of fraps cvs files but it comes up with an error message ' insufficient data, invalid file or parse error'
What's going on?
In Fraps, be sure to enable "Frametimes" on the FPS tab. The other 2 options can be off. The proper benchmark file should be named "<exe name> <date> <time> frametimes.cvs"
If that's not the problem, please copy the entire error message, and attach a sample file (it must be renamed as .txt)
Ok you were right about the frametimes box, I had it unticked because I was only interested in the FPS, I only want the broad picture of what's going on, hence only FPS was ticked. Just did a quick run with it on, worked this time & got a nice detailed graph .
I was thinking that your program could graph the FPS.csv file but I see I was wrong, which means I can't graph my benchmarks , oh well nm.
I don't suppose it can be made to graph from the fps file can it? (I'm not after the micro detail you get from frames).
Btw I see the frametimes file is much bigger than the fps one! 216KB vs 1KB!
Thx for your help anyhow
Frafs Bench Viewer is getting featured on some YouTube hardware reviews, like this one from Newegg TV (starting @ 14:40). This video is a good Bench Viewer video tutorial, BTW.
EDIT this video also shows the recently improved microstutter performance in AMD drivers, but Crossfire (starting at around 20:30) is still a problem.
Good to hear that! Your programs for Fraps are really useful to a lot of people imo
Version 0.3.0.2. Minor bugfix: there was an error loading Excel-generated .csv files.
Thanks to user @DigiHound for reporting the problem by PM.
do you think you'd want to move your project to github? (tho i'm not sure if their binary/build hosting is free)
supposedly sf.net took over some "dead"/abandoned projects and injected adware into them, so executables on their site make me slightly notsureif.gif
@plonk420 Unfortunately, raffriff is no longer an active contributor to this forum and does not post. Your message is almost certainly not going to be read, although he seems to log into the forums semi-regularly and may notice your message.
Great tool, thanks for this!
I do have a question tho: for the very same set of data, I get Min FPS of 155 from Fraps, but 134 fps at 1% and 76 fps at 0,1% from Frafs.
What's going on? Shouldn't min fps be lower or equal to 76?
Just found out the answer.
Indeed, in a given second, there were at least 155 frames rendered. However, not every one of these 155 were the same: some took longer to render than others.
What the low 1% and 0,1% show are the individual frames that took longer to render (in miliseconds) than others. Since we don't really interpret well in ms, it's easier to translate that number to fps. So 13 ms becomes 76 "fps".
Separate names with a comma.