GreenSock's BlitMask Interactive Example Performance

[RedGiant]

Why is the MB Ram in scrollRect and Normal Mask 1/3 of the BlitMask MB Ram?

 

While using the interactive example that GreenSock demonstrates BlitMask with, the MB Ram was clocking in at 34-36MB. The frame rate for BlitMask maintained a solid 60 fps. When I switched to the Regular Mask and scrollRect the MB Ram was at 12-14MB for both, they both ran 58-61 fps.

 

 

Thanks for superb plugin,

Rob

[RedGiant]

Ok. After delving a little deeper into the BlitMask Interactive example, the text that you used inside of the example was very helpful on giving me a general idea on Blitting and how it gets used in memory.

[GreenSock]

Good question. I'm kinda surprised by the numbers you're reporting actually - when I run it, the regular mask and scrollRect clock in around 60MB of RAM whereas BlitMask is around 80MB, so it's not a massive jump. BlitMask will of course use more because it has to capture a Bitmap version of the target from which it pulls its pixels during the scroll/animation. Also, the fps I see with the regular mask is much lower, like around 27fps whereas BlitMask runs at 60fps. I've got a Quad-core i7 at 3.6Ghz, so it's no slouch by any means. You're getting close to 60fps with the regular mask? Really? Something sure seems odd here.

[RedGiant]

I am running OS X with two 2.4GHz Quad-Core Intel Xeon and 12GB 1066MHz DDR3. I just did a retest.

 

BlitMask

58.4 Avg FPS

59.4 MB RAM

 

BlitMask(bitmapMode) This looked incredible!

60 Avg FPS

52.3 MB RAM

 

BlitMask(bitmapMode & smoothing)

60 Avg FPS

52.5 MB RAM

 

Regular Mask

57.9 Avg FPS

27.4 MB RAM

 

Regular Mask(cacheAsBitmap)

59.9 Avg FPS

28.2 MB RAM

 

scrollRect

59.6 Avg FPS

27.6 MB RAM

 

scrollRect(cacheAsBitmap)

60 Avg FPS

28.6 MB RAM

 

Here are the results taken at the end of each tween. I also attached a screenshot of the Regular mask example.

[GreenSock]

Strange question: what color does the FPS meter glow when the test is running for you? Red or green?

 

There's definitely something wrong with the numbers you're providing - it sounds like you might be getting the lighter-weight version for mobile devices which would explain why the performance seems so close.

[RedGiant]

The glow is green. Different results this time:

 

BlitMask

59.2 Avg FPS

78.9 MB RAM

 

BlitMask(bitmapMode)

59.9 Avg FPS

69.4 MB RAM

 

Regular Mask

58.9 Avg FPS

44.4 MB RAM

 

scrollRect

59.5 Avg FPS

44.5 MB RAM

 

Strange instance: when I initially loaded it, it was set to BlitMask w/ bitmapMode selected. The MB RAM was running at 117. I had never seen it at this level before.

[GreenSock]

you are clicking the "tween" button and letting it run for at least a few seconds right?

 

Can anyone else reproduce these results?

 

what browser?

 

Can you make sure that you see a "v0.23" in the upper right corner of the demo? http://www.greensock.com/blitmask/

[RedGiant]

Yeah I let it run until the tween is completely done.

 

I am running Chrome 13.0.782.218

 

At the top of the http://www.greensock.com/blitmask/ it reads:

Version 0.23, Updated 2011-8-30

 

 

Right now I am testing from my laptop at home and here are my results:

Windows 7 Quad Core i7 2.2Ghz 8GB RAM

 

BlitMask

7 Avg FPS

52.8 MB RAM

 

BlitMask(bitmapMode)

45 Avg FPS

46.2 MB RAM

 

Regular Mask

7.2 Avg FPS

20.9 MB RAM

 

scrollRect

6.7 Avg FPS

20.9 MB RAM

[GreenSock]

Ah yes, those numbers look much more accurate. 45fps average with BlitMask vs. 7.2fps average with a regular mask. These are what you were expecting too, right?

[RedGiant]

To be honest, I wasn't sure what to expect. I don't have much experience with Blitting. Those numbers that you said are more accurate (45fps vs 7.2fps) were run from my home PC laptop last night. I am back here on my work Mac, and Regular mask just averaged 55fps again.

[GreenSock]

What browser are you using on your Mac? Does it say "v0.23" in the upper right corner of the interactive swf itself (NOT the web page)? And what happens if you load the swf directly, like http://greensock.com/as/swf/BlitMask_Demo.swf?

[RedGiant]

Browser on the Mac is Chrome 13.0.782.218

 

 

Following the link: http://greensock.com/as/swf/BlitMask_Demo.swf I got these results:

First: At the top right corner the version was v0.23

 

Second:

BlitMask

13 Avg FPS

59.9 MB RAM

 

BlitMask(bitmapMode)

60 Avg FPS

50.4 MB RAM

 

Regular Mask

13.1 Avg FPS

25.7 MB RAM

 

scrollRect

13.2 Avg FPS

25.8 MB RAM

 

 

 

After that test, I tested this link again: http://www.greensock.com/blitmask/ I got these results:

First: The version was v0.23

 

Second:

BlitMask

40 Avg FPS

55.2 MB RAM

 

BlitMask(bitmapMode)

60 Avg FPS

48.3 MB RAM

 

Regular Mask

33.4 Avg FPS

25.7 MB RAM

 

scrollRect

42.5 Avg FPS

22.8 MB RAM

 

 

This is crazy.

[GreenSock]

Indeed, it is insane. Not sure exactly what is happening, especially because you get RADICALLY different results when you load the swf directly. This only seems to happen in Chrome on the Mac and even then it's not consistent.

 

Well, you'll have to either run your own tests or trust me when I say that BlitMask can run SIGNIFICANTLY faster than a standard mask when it's in bitmapMode and the target is large.

[Carl]

RedGiant,

 

Just so you know I have been seeing the same bizarre variance in the blitmask demo performance in Chrome when viewing the same swf in different places:

 

http://greensock.com/as/swf/BlitMask_Demo.swf

I get the expected results of BlitMask raging at 60fps

mask and scrollRect lagging at 15fps

 

These results are similar on the same system (iMac quad core i5) in FireFox and Safari when viewing the official blitmask page.

 

 

 

http://www.greensock.com/blitmask/

Viewing the official blitmask page, like you, in Chrome, BlitMask soars at 60fps and scrollRect and Mask can be anywhere from 40-60.

 

in FireFox and Safari on my iMac and a much older macbook, BlitMask runs at 60 and the others in the teens (as expected).

 

This is the strangest Flash Player performance variance I have ever seen.

I did a bunch of tests running the swf locally, changing the embed code, clearing cache and other totally fruitless experiments.

 

My latest test offers the most conclusive evidencethat there is just simply something different about the Chrome version of Flash Player.

 

Chrome uses its own version of Flash Player that is "baked into" the browser. Chrome keeps this version updated all the time and the user doesn't ever have to intervene.

The chrome version of Flash Player has its own version number (although currently it is the same as the other os variants)

 

Chrome does allow you to disable the integrated chrome version and use your SYSTEM version that the other browsers use.

 

Following the instructions on this page: http://kb2.adobe.com/cps/839/cpsid_83950.html

 

I was able to disable the integrated version and tell Chrome to use my system version of Flash Player and the demo on http://www.greensock.com/blitMask performed exactly as I have seen it in other browsers and on other machines with FP 10.

 

The only caveat here is that my system version on my iMac is Flash Player 11 Incubator that supports the Molehill / stage3D stuff.

Regardless, I'm near certain that this is the closest thing I can get to proof that the Chrome version is simply... different. Is it better optimized? don't know. Is it a glitch? don't know.

 

If you have the time to disable your integrated player, it would be great to see if the results offer conclusive proof that the integrated plugin is the culprit in Chrome.

 

Thanks to greensock for inspiring me to dig into more info on the Chrome version of Flash Player.

 

-Carl

[GreenSock]

Nice job, Carl - sounds like all indications point to a fancy Chrome-only flavor of the player that has some extra goodness in there. I wonder if it actually has to do with GPU acceleration (where it's offloading the CPU's work onto the graphics board on your computer). Just a guess.

[RedGiant]

That's an awesome find Carl. I disabled Chrome's Flash plugin to use the system's Flash, cleared my cache to the beginning of time, and restarted the Mac. It ran the almost identical results to my previously posted Chrome w/ baked Flash. So I ran Safari and FireFox:

 

Safari:

Official Blitmask page

18-22fps for scrollRect, Regular Mask, and BlitMask

Blitmask swf only page

10-12fps for scrollRect, Regular Mask, and BlitMask.

 

FireFox:

Official Blitmask page

40-55fps for scrollRect, Regular Mask, and BlitMask. It was very erratic.

Blitmask swf only page

13-15fps for scrollRect, Regular Mask, and BlitMask.

 

*Every test in every browser on every page, BlitMask w/ bitmapMode ran at 60fps.

 

I am unsure as to why the system's flash ran such close results to Chrome's baked flash player.

[Carl]

RedGiant,

 

Thanks for taking the time to test and post your recent results.

 

It's odd that the single-swf runs much worse, I suspect that it perhaps is filling the entire browser window and thus the re-draw regions are larger... possibly?

 

As for why the baked vs system results are very similar (unlike my results) a guess is that your system is so powerful that whatever optimizations may be present in the baked version are negligible given all the horsepower you have to run the system player.

 

I really have no idea.

 

Even though your results aren't exactly what I was expecting, they do provide further evidence that BlitMask w/ bitmapMode out-performs the other options by a considerable margin.