FPS Graph - Performance Analyzer Tool

Wouldn't it be best to log (instead of display) on phones display a timeline afterward? (Whenever user pressed a key, pause game display previous 1000 frames result)

 

Wait... you are getting a drop in framerate on your Macbook Pro? I was only talking about low power devices, with specialty graphics cards, you should hardly get any framerate drop on a computer with a name brand graphics card... I also have a macbook pro and I don't have any visible drop in framerate.

Are you sure it's my script that is causing the drop in framerate? Maybe you can compare it with this generic fps counter, to see if there is a difference?

The new updates are not in the store yet (really this only covers android devices though). Hopefully the Asset Store accepts it soon!

 

The update to FPS Graph is in the Asset Store!

I hope this helps everyone with their Android woes!

 

Does this force VSync? I drop from 400 FPS to 60-90 FPS when this script is active on PC. Is this normal?

 

Unfortunately yes, it seems to be normal although the last update was supposed to fix it, however we're still seeing massive performance drops when it's turned on just like you

 

That's unfortunate. Suppose I'll just make my own simple script to calculate FPS using the below as example.

http://wiki.unity3d.com/index.php?title=FramesPerSecond

I'd also like to see an option to just render the FPS with no history. So it'd just render the digit.

 

I tried that script last year. If you're using a GUIText to display it, that alone can drop you from 30FPS to 18FPS on an iPhone 3GS. I suggest using NGUI or similar to display the FPS or it's not accurate.

 

I'm using CoherentUI and see no performance impact. I've never used Unitys GUI and NGUI is so silly clunky. I also don't mobile develop. Trying to have decent FPS tracking ingame so I can monitor performance outside my development environment lol. I wish Unity would just add a built in function to calculate accurate FPS like they should for all common features tbh. I really like this nice graph as it gives a good attractive history display. Be better if I could use it through API calls so I can implement my own display usages.

 

Unfortunately you cannot compare the FPS that you see in the Unity stats window with the FPS you see in FPS Graph. One is accurate (FPS Graph), the other is just an estimate (Unity's Stats window). Here is a lengthy forum post that details the problem with Unity Stats FPS: http://forum.unity3d.com/threads/150139-Is-unity-FPS-count-wrong-or-am-i-missing-something/page2#post_message_1074929

The script does not force VSync, so if indeed you were getting 400 fps (measured by something else than the Unity Stats window) and now are getting 60-90 than it could have something to do with FPS Graph although I have tried to keep it as lightweight as possible. Try turning down the history length for even better performance...

 

My own custom script reports around 300 FPS, Unity reports around 400 FPS, and FPS Graph reports around 60 FPS. Below is my own script as a PlayMaker Action.

Code (csharp):

1.  
2. using UnityEngine;
3.  
4. namespace HutongGames.PlayMaker.Actions {
5.     [ActionCategory(ActionCategory.Application)]
6.     public class ApplicationGetFPS : FsmStateAction {
7.         [RequiredField]
8.         public FsmFloat updateInterval;
9.         [RequiredField, UIHint(UIHint.Variable)]
10.         public FsmFloat storeFPS;
11.  
12.         private float accum = 0;
13.         private int frames = 0;
14.         private float timeleft;
15.  
16.         public override void Reset() {
17.             updateInterval = new FsmFloat { Value = 0.5f };
18.             storeFPS = null;
19.         }
20.  
21.         public override void OnUpdate() {
22.             DoAction();
23.         }
24.  
25.         void DoAction() {
26.             timeleft -= Time.deltaTime;
27.             accum += ( Time.timeScale / Time.deltaTime );
28.  
29.             ++frames;
30.  
31.             if ( timeleft <= 0.0 ) {
32.                 storeFPS.Value = ( accum / frames );
33.                 timeleft = updateInterval.Value;
34.                 accum = 0.0F;
35.                 frames = 0;
36.             }
37.         }
38.     }
39. }
40.  

Is it possible FPS Graph can have a 3rd script that's purely for API usage without the UI entirely? An array of history and a float for the current FPS would be all that's needed so our own UI solutions can be implemented. I don't know specifically what's causing FPS Graph to report FPS significantly lower than other solutions, but perhaps is the use and constant updating of Unity GUI (Unity GUI has horrible performance).

 

Yeah I will look into providing that for the next update, I am still a little shocked it slows down the scene that much. Would you mind trying commenting out the whole OnGUI function, to see if that is what is slowing it down? The OnGUI actually doesn't do that much besides help display the graph legend on the bottom (not really necessary), everything else is done with pure Open GL code, so it should be very fast...

 

I finally see what was causing the framerate to max out at 60fps. I am preparing a build, and it should be in the Asset Store shortly. Thanks for your patience everyone, and sorry I didn't really understand the problem until very recently (the script wasn't causing poor performance, it just wasn't reading the framerate in an optimal manner for levels above 60fps).

 

Great news FPSHighfive!

 

Haha, thanks! I also fixed some issues with publishing to Android and iOS, unfortunately it required me to update it with only Unity 4.x.

Quick poll, does anyone still use Unity 3.5???

The only reason I still use it, is to submit content to the Asset Store, but if other people are keeping it around for a good reason than I will continue to try and support it...

 

Quick poll answer: no. Me too I just keep Unity 3.5 for submitting to Asset Store (and for an old game I still didn't update, but which I will)

 

The new version of FPS Graph is in the Asset Store. This fixes the issue of the utility capping out at 60fps.

Thanks everyone for bringing this to my attention!

 

Great news, thank you for new release! I was really missing the graph, lol. While my script provides pretty decent accuracy it was near useless without that wonderful graph to show progress of FPS. Again, thank you!

 

I must second this request.

Also a question --- some (or most) Android devices have a hard framerate cap. One of my HTC is capped at 30 fps, another at 60 fps, and disabling vsync in Unity does not help there:

For continuous on-device performance monitoring during development (testing out the impact of game additions over time), it would be great to measure frame time even when currently the "FPS" would be still be at par or higher than the hard cap.

I suspect currently for a near-empty Unity scene (hence very high render performance) on a device with a hard vsync (that sadly cannot be disabled via Unity project settings or scripts), your graph would report just the vsync fps cap, so a constant 30 or 60 right? But that doesn't help analyze for example whether a newly added shader or script adds a heavy 5ms to frame time, and this will only become noticeable once the whole setup drops under the vsync --- by that time we have too much new stuff going on and would have to dive into the whole messiness of profiling, debug builds etc.

Of course it's really hard to measure whole frame time with vsync --- even if you measure CPU-side frame time, most of the time we might be GPU-bound. I guess if I want to avoid profiling/debugging I'll need to figure out a way to force vsync-disabled somehow on unrooted Androids. BUT since you're experienced with measuring frames, maybe you have some ideas on this that can be added into FPS Graph (or maybe already does), then I'd happily buy at 4x the normal price and I'm sure many others would, too. I mean most real-world, non-rooted mobile devices have vsync enforced from what I hear, so anyone with a solution to this would be highly appreciated I'm sure

 

Hi Metaleap, The vsync hard cap shouldn't effect the framerate measured by FPS Graph. iOS devices are also hard-capped at 60fps but FPS Graph is able to report if it is having better performance than that (it didn't always do this, but it was one of the things I fixed recently). You should be good to go, providing a API for the framerate could definitely be done, but it wouldn't give you a different value than what is already displayed, and my GUI is pretty optimized as to not caused much of a framerate drop itself... I hope that helps, let me know if you are running into framerate cap issues.

 

Thanks for getting back!

So this implies to me you're somehow able to measure both CPU and GPU time per frame --- without profiling APIs and despite the (assumed on my part) Unity-internal GL.SwapBuffers() call that in OpenGL (ES, too, I suspect) syncs the GPU+CPU frame processing (and which, with vsync on, simply takes that much longer, blocking until vsync has elapsed)?

I won't ask you how you accomplished that, just confirm that you did. If your counter not only measures CPU frame time but also GPU frame time (since they're parallel of course there's some overlap) regardless of vsync, I'm sold

 

It does, at least it attempts to . As you mentioned i can't really access any hidden Unity internals so it's really just based on some guesswork of what is going on behind the scenes, and measuring the difference between the OnPostRender, Update and LateUpdate. It seems to at least give a reasonable estimation of what is happening, but I can't guarantee it's 100% perfect in breaking down the GPU+CPU rate. Actually GPU might be a bit of a stretch it is really just measuring the "render" time, but that can be quite helpful at least for my debugging (but like you suggested it may not measure the overlapping periods accurately).

 

3.5 user here, would you please tell were to find me the last version still compatible with 3.5, not ready upgrade to 4 yet.

 

OK, interesting. I might check this out. But beware! OnPostRender (as I understand it and from my own experiments) is when the CPU is done with all render commands for the current frame. Typically the GPU driver buffers them and the GPU cores begin to process them in parallel as fast as they can, but typically the GPU isn't done by the time all graphics calls (including things like shader switching, texture switching etc.) have been issued from the CPU -- and so the next thing the CPU side is doing before the next frame is wait for the GPU to catch up, in desktop GL this is done via SwapBuffers but this is internals. But that's how it happens. And so with vsync on which delays SwapBuffers (or the wait-for-gpu-to-finish phase) by the vsync interval, GPU timing isn't doable from what I can tell so far (and that's why I was looking around for a solution).

As soon as you're GPU-bound, you'll notice your FPS estimate based on PostRender (PostRender minus last Update) gives a really fast estimate while your actual FPS (ie. counting how often PostRender happens per second) slows to a crawl. At least that's my current suspicion. But I'm not GPU-bound yet, so will only find out in the future.

 

I've got a discrepancy in what FPSGraph is showing me and what Unity's Stats are showing me.

Basically, Unity's Stats are showing me a framerate of about 10FPS (or less) while FPSGraph says I'm getting a fairly solid 120FPS. The scene/game is obviously running incredibly slow, so I'm inclined to believe that Unity Stats is more accurate.

I recorded a video of it in action so you can see what I'm talking about (FPS Graph shows a worse framerate (~80FPS at worst) in the video than it usually does, probably due to screen recording)

http://screencast.com/t/k9bWoC31

Any idea why the large discrepancy between what FPSGraph reports and what is actually happening? I've got to say it sounds a lot like what metaleap just described, except I can't imagine why my project would be using the GPU a lot. It seems to me like most of it should be CPU based on what's happening in the scene.

 

Wow that is some really strange behavior, at times they look in synch, but when it drops to 10-15fps FPSGraph seems to almost slow down it's graphing...

I honestly can't really fathom what could be causing this. Do you know of any special behavior that is taking place during that slow down period?

 

The framerate slowdown was due to a misconfiguration on the objects' layers, so it was detecting/calculating a lot of physics (kinematic rigidbody triggers) colliding with each other. I suppose the physics could have been something the GPU was calculating, in which case it was indeed what metaleap described.

 

Hey DentedPixel, have you run into this error before?:



Error text:

Code (csharp):

1. IndexOutOfRangeException: Array index is out of range.
2. FPSGraphC.addNumberAt (Int32 startX, Int32 startY, Int32 num, Boolean isLeading) (at Assets/FPSGraph/FPSGraphC.cs:225)
3. FPSGraphC.OnPostRender () (at Assets/FPSGraph/FPSGraphC.cs:540)
4.  

Code (csharp):

1. Matrix stack full depth reached
2. UnityEditor.DockArea:OnGUI()
3.  

It happens shortly after starting my game, every time. Using Unity 4.3.0 and the latest version of the script from the asset store.

 

Hi NewwaveNick,

I have not run into this error before, but based on my debugging I think it might be, because your garbage collection memory exceeds 100mb. Which I did not expect would ever happen. I am putting in a fix for this right away, but if you want to just comment out line 539-547 for the time being it should go away...

Thanks for alerting me to this!

 

I have submitted a new version that should avoid any problems with memory over 100mb in size (if that was your problem which hopefully it was).

It should be in the store within the next couple of days...

 

Hahaha, good grief. Sounds like I'm going to be spending some time tracking down that memory and making sure it's not doing that during gameplay.

Commenting out those lines did in fact stop the error. Thanks for the fixes, both temporary and long term!

 

Is there a version available which still just counts up frames rendered on iOS capped with v-sync at 60 fps? at least for a comparison? or perhaps there could be a toggle option for that?

I'm impressed with FPS Graph and how simple everything is. It's not often you import a pack and it just works. I also exposed a couple properties on the class which I toggle/use from my NGUI fps counter in the bottom left of the screen.


[HideInInspector]
[System.NonSerialized]
public bool showGraph = false;
[HideInInspector]
[System.NonSerialized]
public int currentFPS = 60;

showGraph toggles the graph on and off. currentFPS is set to what the actual frame rate is that you calculate (here I would like an actual frame count versus a supposed estimate of frame rate without v-sync). That way, when that dips below 60 i can pop up the graph and evaluate things.

Finally, i would recommend ditching OnGUI.. even including an OnGUI call in a script (even if it doesn't do anything), does GC allocations in my experience. You could do an OnGUI version of the script and an NGUI free version of the script. Though, for $5 I'm not complaining Great work.

 

Hey MrScarey,

Thanks for the compliment I am glad FPS Graph is working out well for you. Unfortunately I did not keep the version backed up that would max out at whatever framerate it was capped at

I have not delved into NGUI myself, so I am not sure exactly how to implement that, I wouldn't be against it, but since this isn't meant to be used on a production product (just for testing), slight performance issues aren't really a high priority...

I am not sure I caught what you meant about the currentFPS... so you were saying it would be helpful to have a "trigger" fps that would only pop-up the graph if the fps fell below the desired value? That's a neat idea. That might work well as a companion script instead of baked into the actual FPS Graph, but I could see that being helpful. Let me think about that, thanks for the idea...

 

I'm using this to help me measure performance, and I was doing tests with raycasts (figuring out how to use them as cheaply as possible). In my tests I found (using this)

(1000 physic rays being made per update at same position in same path. Same result with fixedupdate)
With no obstacles, correct framerate.
If I have a thousand boxes spread along a line, correct fps.
If I have a thousand boxes in the same point, incorrect fps (real fps very low even in standalone. Reported fps from this is regular high)
If I reduce the rays made per update (from 1000 to 0) the fps returns to norm.

Something is off with physics, or at least raycasts in how they report time.

 

Thanks for the heads up. Would you be able to send over an example project, so I can debug? Either with a direct message or email would work (dentedpixel at gmail.com).

Thanks!

 

Hi,

been a user of your LeanTween... love it...
and now i come across your FPSGraph which spiked my interest...

I do not own Unity Pro (yet) and is cringing for a performance test for Unity Basic... will this actually help measure what my asset is doing (like the Unity Profiler) ?

and is there a baseline score or a chart reference of what a good result is?

 

Hey Kurayami88,

Thanks! Glad you like LeanTween. I remember your handle from the LeanTween forum.

FPS Graph certainly does it's best to attempt at recreating a lot of the neat features of the Unity Profiler, of course there are some details that you can only get with the profiler. It also has an option to give a report after x-amount of time, which is good for performance testing (it's what I have been using in my latest videos when testing LeanTween performance).

In general I think it is pretty nifty asset, particularly for $5, but I guess I may have a biased opinion

 

bought the asset to give it a go... don't know how accurate it is but at least i know i'm getting somewhere

--------


not sure why the result page looks oversized... but the results are there...
game = matching gems game...
using 2d toolkit...
CPU bound game... ( i know this...) but never knew how much of a hit it took to play the game...
current config = android platform @ 800x1280 potrait mode...

it spikes when :-
1) game load/start...
2) it instantiates new objects into the scene ( i know this too...)

pooling is enabled so after instantiate, no longer spiking...
apart from that, even during intense activity, the results are promising
tq for giving me a baseline on my asset i'm so happy atm

 

Awesome, I am glad it's giving you some good feedback to work from. Oh man I never tested that portrait mode layout, it's looking a little messed up , but still legible... hehe, I'll have to work on that on the next update.

What are you using for pooling? I tried to roll my own solution to integrate within LeanTween, but I wasn't seeing much of a performance increase...

 

2nd run with pooling disabled...
performance strangely better? but it spikes a lot ... i mean SERIOUSLY ALOT during instantiate + destroy... can see the graph like having heart attack
lower min FPS too which is somewhat expected

using Pool Manager 5

 

Haha, "graph like having heart attack"

Good to know, maybe I will revisit my own pool manager and see if it at least helps with spiking...

 

think i'm understanding your stats a bit more... correct me if i'm wrong...

the score increases continually...
and the true measure if your game is performing well is to take the score output and divide it by the run time?

so like... my results above...
1st run = 16934 ms / 152.9 s = 110.75 ...
2nd run = 17466 ms / 146.1s = 119.54 ...
------
on my main menu page with no process checks scripts running...
915 ms / 13.1 s = 69.85

your LeanTween stress test results on the FPS graph screenshot...
6757 ms / 12.0s = 563.0

obviously lower = better?? ...
so then what is a good performance range?

 

That is correct, it's divided over the runtime and the lower score the better.

It's up to you really what is the cutoff for acceptable performance. I find looking at the lowest FPS to be my favorite indicator of "acceptable" performance, just because that can really turn off players, if you can get a <strong>steady</strong> FPS no matter if it is 25fps or 60fps the player won't be pulled out of the game with stuttering framerate...

 

tested on an actual device...

samsung galaxy s4 running android 4.3...
score = 79903 ms @ 162.5s
performance = 491.71 ....

lowest fps = 14fps
highest = 214 fps
average = 120 fps...

--------

in my case, i believe the lowest fps is not a good indicator, especially during game start init phase + the short burst of instantiate messes up the lowest fps value.... unless of course you reset the graph midway after all the initial stuff is done... hmm... good idea...brb...

------

midway test results :-
24540 ms @ 232.7s = 105.45

low fps = 27
high = 192fps
avg = 110 fps...

-----

would consider the avg a better score

 

oh ... side note... i would suggest you export your FPS graph using Unity 4.0 or such...
your current export to asset store using Unity 4.3 sets the baseline @ 4.3 ... Asset store would say that i need to upgrade my Unity to use the product...

but manually downloading it and manually importing the asset *in the asset library page, rather than the product itself page - of which you can choose to download/import manually* ( and i'm currently using Unity 4.2.2) ... it can run

------

just some inconvenience to some users who doesn't know better....

 

Oh good point, I thought I was getting the lowest FPS only after the first initial spike, but I will have to double check that, because otherwise it's kind of useless...

Argh, you are using Unity 4.2.2? I try to output a lot of my packages with Unity 3.5 when I can, but I guess I will have to think about downloading Unity 4.0 as well...

Can I ask why you haven't upgraded to the latest Unity? I can see benefits sometimes to staying with Unity 3.5, but not sure of the benefits of not upgrading to the latest 4.x.

 

perhaps maybe you already ignored the initial spike... *not sure as i don't know what you already did* but my own asset spikes during "new" objects thus early game instantiating (but not init phase) thus spiking of which is later pooled, of which i resetted the graph fps counter to check results again after all that

oh btw, your fps values resets when opening the graph and closing it... the score resets as well... but the runtime remains... skewed info...

Personally i would figure ppl still using 3.5.x due to pro license they don't want to pay for upgrading...

For myself, i reverted back to 4.2.2 for now, as initially 4.3 is unstable as crap during the official release.... only now, the current 4.3.2 seems stable... before this, 4.3 has serious performance issues. On the plus side, i was not using the new 2d features as well so upgrading was not necessary for me... lazy to upgrade my project to 4.3 (and be the guinea pig) until it fully matures on the new feature sets....

I'm not sure if you really need to get 4.0, but I at least suggest having a baseline support for 4.2.2 as there are also several producers who has downgraded back to 4.2.2 due to 4.3's instability.

 

Is it possible to show more memory data, e.g. System.GC.GetTotalMemory(false) , Profiler.GetTotalAllocatedMemory()

 

Hi Yuewahchan,

That's a good idea. I have tried to make the FPS Graph as "non-pro" friendly as possible, since the Profiler code is only available in the Pro version of Unity. But I could certainly turn this on if the user has pro. I'll add this for the next update, let me know if you want it earlier than that and I can direct message it to you.

Cheers,
Russ

 

Hey there!
Can someone tell something about the performance of this script?
How much CPU/Render does it use itself to show and calculate the informations?
I've been using some free solutions lately, but using Unitys profiler i noticed that some of these free scripts have a very poor performance (e.G. 5% of the CPU) which leads to wrong results.
Thanks

 

Just downloaded latest from the asset store and by default ie using FPSExampleScene the graph does not get rendered on Android. This is using Unity Pro 4.6.1p2.

Additional info. Seems to work fine on Android tablets both those with Atom/ARM ie HP 1800 and Nexus 10. Does not work on Galaxy S3, Droid Incredible. Trying others.

These does not seem to be directly your issue as just a basic app running GL functions will not render on those devices. So it seems to be a GL issue not sure if that is something you could help with ie why wont the GL calls render.

Thanks.