Passer au contenu de la page
Passer au menu arborescent

Accueil - MovieStar, rendering up to twice the speed with a dual core

Crée le :
Dernière révision :
[english] [français] [2007] [2006] [2005] [2004] [2003] [2002]

MovieStar, rendering up to twice the speed with a dual core

English post

MovieStar (Flash Player 9.0.115.0) is officially the last Flash Player release that will be downloaded when public users upgrade to the latest Flash Player version. In the summary of the new features that were added to this release, Adobe announce that developers of the Flash Player have worked to Speed the rendering of vectors, bitmaps, filters and video, including true 1080p video, with new multi-core support that takes advantage of up to 4 CPUs.

I was really interested with this announce. I have so made a really simple Flash animation, all timeline-based with 36 steps that tests rendering with : vector, bitmap, filters, rotation, scale, translation.

You can download the sources of the benchmark or give it a try by your own here, but if you think to just read the results, what you need to know is that I have used a 640x480 animation published at 120 fps with a background that uses a lot of hairy strokes. Shape tests were made with a sort of square+circle+oval+rectangle vector pattern. Bitmap tests were made using a 24bits alpha PNG with smoothing on. Actually, benchmarks were only done with the standalone non-debug version of the two players.

Intel Dual Core

The most interesting tests were made on the same Intel Dual Core based computer that authorizes to activate or deactivate the second core support. With the same computer, I was so able to test rendering speed of the player with Flash Player 9.0.45.0 and Flash Player 9.0.115.0 in so called "Single" or "Dual" Core mode and compare them (units are Frame Per Second) :

You can first see that the old Flash Player 9.0.45.0 did not benefits Dual Core mode at all. But Flash Player 9.0.115.0 with Dual Core mode activated will run with near twice the speed of the Flash Player 9.0.45.0 in some tests and an overall 156% its speed. It is a pretty good result.

Here follow the two tests results tables used to build the previous comparison chart, first in "Single Core" mode, second in "Dual Core" mode :

Intel Dual Core - Pentium D 830@3Ghz - single core mode Flash Player
9.0.45.0
640x480		 Flash Player
9.0.115.0
640x480
shape move quality Duration FPS Duration FPS
vector rotate animation quality low 00:04:367 54.73 00:04:410 54.20
medium 00:09:115 26.33 00:08:823 27.20
high 00:16:053 14.95 00:15:607 15.38
best 00:16:073 14.93 00:15:650 15.34
rotate dropshadow quality low 00:41:016 05.85 00:39:969 06.00
medium 00:47:651 05.04 00:47:354 05.07
high 00:53:406 04.49 00:52:750 04.55
rotate blur quality low 00:38:031 06.31 00:38:631 06.21
medium 00:51:506 04.66 00:52:321 04.59
high 01:05:247 03.68 01:07:114 03.58
scale animation quality low 00:05:703 42.26 00:05:940 40.57
medium 00:10:504 22.75 00:10:525 22.71
high 00:17:430 13.77 00:17:229 13.93
best 00:17:476 13.73 00:17:295 13.88
translate animation quality low 00:16:923 14.18 00:16:886 14.21
medium 00:10:244 23.43 00:10:233 23.45
high 00:16:916 14.19 00:16:842 14.25
best 00:16:991 14.13 00:16:881 14.22
bitmap rotate animation quality low 00:07:035 34.12 00:07:031 34.13
medium 00:18:485 12.98 00:18:061 13.29
high 00:33:381 07.19 00:32:132 07.47
best 00:33:297 07.21 00:32:346 07.42
rotate dropshadow quality low 00:44:809 05.36 00:42:763 05.61
medium 00:51:762 04.64 00:50:039 04.80
high 00:57:867 04.15 00:56:181 04.27
rotate blur quality low 00:43:379 05.53 00:43:077 05.57
medium 00:56:594 04.24 00:56:884 04.22
high 01:10:422 03.41 01:10:410 03.41
scale animation quality low 00:06:603 36.35 00:06:532 36.74
medium 00:18:111 13.25 00:17:345 13.84
high 00:32:324 07.42 00:30:971 07.75
best 00:37:745 06.36 00:34:878 06.88
translate animation quality low 00:06:829 35.14 00:06:745 35.58
medium 00:19:949 12.03 00:19:137 12.54
high 00:35:122 06.83 00:33:650 07.13
best 00:52:996 04.53 00:51:757 04.64
Total 18:01:362 07.99 17:44:399 08.12


Intel Dual Core - Pentium D 830@3Ghz - dual core mode Flash Player
9.0.45.0
640x480		 Flash Player
9.0.115.0
640x480
shape move quality Duration FPS Duration FPS
vector rotate animation quality low 00:04:160 57.45 00:03:180 75.16
medium 00:08:866 27.07 00:05:531 43.39
high 00:15:808 15.18 00:09:184 26.13
best 00:15:826 15.16 00:09:190 26.12
rotate dropshadow quality low 00:40:516 05.92 00:23:027 10.42
medium 00:47:242 05.08 00:26:301 09.13
high 00:52:905 04.54 00:29:543 08.12
rotate blur quality low 00:37:803 06.35 00:28:359 08.46
medium 00:51:170 04.69 00:42:143 05.69
high 01:04:240 03.74 00:56:626 04.24
scale animation quality low 00:05:684 42.40 00:04:633 52.02
medium 00:10:482 22.80 00:07:007 34.11
high 00:17:358 13.83 00:10:689 22.45
best 00:17:449 13.75 00:10:710 22.41
translate animation quality low 00:16:867 14.23 00:10:346 23.20
medium 00:10:186 23.56 00:06:855 35.01
high 00:16:872 14.22 00:10:380 23.12
best 00:16:890 14.21 00:10:411 23.05
bitmap rotate animation quality low 00:06:935 34.61 00:04:220 56.87
medium 00:18:326 13.10 00:10:124 23.71
high 00:33:041 07.26 00:17:803 13.48
best 00:33:030 07.27 00:17:707 13.55
rotate dropshadow quality low 00:44:157 05.44 00:24:207 09.91
medium 00:51:068 04.70 00:27:904 08.60
high 00:57:078 04.20 00:31:375 07.65
rotate blur quality low 00:42:734 05.62 00:30:520 07.86
medium 00:56:237 04.27 00:44:508 05.39
high 01:09:566 03.45 00:58:864 04.08
scale animation quality low 00:06:519 36.82 00:04:107 58.44
medium 00:17:994 13.34 00:09:986 24.03
high 00:32:101 07.48 00:17:433 13.77
best 00:37:442 06.41 00:19:628 12.23
translate animation quality low 00:06:764 35.48 00:04:095 58.61
medium 00:19:782 12.13 00:10:652 22.53
high 00:35:020 06.85 00:18:245 13.15
best 00:52:201 04.60 00:27:293 08.79
Total 17:50:319 08.07 11:22:786 12.65

Intel Core 2 Duo

Here follows the same test made with Flash Player 9.0.45.0 and Flash Player 9.0.115.0 on the same computer but working with an Intel Core 2 Duo E6850@3.6Ghz (3Ghz overclocked...). I think it is not possible to deactivate the second core on the Core 2 Duo architecture because they share the same memory controller. The bios of this computer did not offer this option whatever.

Intel Core 2 Duo E6850@3.6Ghz Flash Player
9.0.45.0
640x480		 Flash Player
9.0.115.0
640x480
shape move quality Duration FPS Duration FPS
vector rotate animation quality low 00:01:935 123.51 00:01:938 123.32
medium 00:03:265 73.51 00:02:005 119.7
high 00:05:926 40.50 00:03:435 69.87
best 00:05:920 40.54 00:03:446 69.65
rotate dropshadow quality low 00:16:843 14.25 00:09:767 24.57
medium 00:19:760 12.15 00:11:196 21.44
high 00:22:501 10.67 00:12:650 18.97
rotate blur quality low 00:14:995 16.01 00:16:923 14.18
medium 00:19:855 12.09 00:26:669 09.00
high 00:24:817 09.67 00:37:219 06.45
scale animation quality low 00:02:296 104.97 00:02:020 119.31
medium 00:03:969 60.22 00:02:672 89.45
high 00:06:632 36.19 00:04:155 57.76
best 00:06:644 36.12 00:04:171 57.54
translate animation quality low 00:06:375 37.65 00:04:023 59.66
medium 00:03:817 62.88 00:02:634 91.12
high 00:06:374 37.65 00:04:024 59.64
best 00:06:378 37.63 00:04:030 59.55
bitmap rotate animation quality low 00:02:585 92.84 00:01:942 123.58
medium 00:06:423 37.37 00:03:972 60.42
high 00:11:420 21.02 00:06:897 34.80
best 00:11:437 20.98 00:06:906 34.75
rotate dropshadow quality low 00:18:319 13.10 00:10:611 22.62
medium 00:21:413 11.21 00:12:148 19.76
high 00:24:296 09.88 00:13:651 17.58
rotate blur quality low 00:17:075 14.06 00:18:138 13.23
medium 00:21:994 10.91 00:27:888 08.61
high 00:26:930 8.910 00:38:405 06.25
scale animation quality low 00:02:441 98.32 00:01:940 123.71
medium 00:06:365 37.71 00:03:752 63.97
high 00:11:209 21.41 00:06:470 37.09
best 00:12:587 19.07 00:06:976 34.40
translate animation quality low 00:02:595 92.49 00:01:943 123.52
medium 00:06:891 34.83 00:03:762 63.80
high 00:12:120 19.80 00:06:464 37.13
best 00:16:602 14.46 00:08:695 27.60
Total 06:51:004 21.02 05:33:537 25.90

The overall result is "only" a 23% rendering speed increase for the Flash Player 9.0.115.0 upon the Flash Player 9.0.45.0. By comparing first results made with the Dual Core architecture and the Core 2 Duo architecture we can suppose that the Flash Player 9.0.45.0 already benefits Core 2 Duo architecture but not the Dual Core architecture and 23% less than the Flash Player 9.0.115.0. I really want to investigate more this result. Please let me know if you have any clue.

Strange result too, notice that the blur effect lost in speed with the player 9.0.115.0 in this test.

source of the benchmark. The archive include two Flash projectors made one with the Flash Player 9.0.45.0 and the other with Flash Player 9.0.115.0.

You can even give it a try by your own here.

Rem : Thread will evolve, I will add ActiveX/Plugin test. Overall and each test results will be traduced in percentage.

[ Monday, December 17, 2007 ]
1371 reads - permanent link - trackback

Comments

nice and all but you should have compared the new release player with the previous beta, then you´d see how much worse the release player acts performancewise compared to the beta.

[ 21 décembre 2007 13:49 ]
tomsamson

I have thought about that but I prefer not to give too much information for people that will read the post next months because they will not care the beta and want clear statistics about official releases.

I could give it a try and let a word about it. Do you have any clue that the official release has lost in performance compared to the beta on multi-core or single-core systems ?

[ 21 décembre 2007 14:41 ]
Tek

yeah, i have a dual core and tested lots of content and have partially a performance loss of over 40 fps with the release player (compared to the beta).
Sevearl developer mates have single cores and they moan that they experience a quite big performance loss with the release player (again compared to the previous beta) ,too.

[ 22 décembre 2007 12:51 ]
tomsamson

40fps performance loss ? No error while typing ?

Could you give me which is your preferred export framerate for animations ?

I will try to make some tests next week after christmas.

[ 22 décembre 2007 14:08 ]
Tek

yeah, sadly no error while typing.

The amount of fps lost with release player compared to beta player wasn´t the same for all types of content of course, it varies depending on various things (for example on how much the performance heap is on graphic handling side or how much its on code execution side in that specific content) but yeah, what was pretty consistently noticable throughout all my tests was that there was a noticable performance loss with the release player.
I´m also not the only one noticing that by far, see posts and comments on several sites/blogs (like here:
http://blog.gamingyourway.com/PermaLink,guid,af7f2db9-2604-4159-8433-6c40b49838d8.aspx , here: http://www.kaourantin.net/2007/12/adobe-flash-player-9-update-3-released.html#comments etc).

Regarding your question which target fps i usually choose:
For export i normally test content on several average systems and see what runs well without eating up too much cpu, but yeah, to run performance tests i usually set the fps way high to be able to see what it can really achieve (like to 120 and above)

And yup, nice, looking forward to your tests :)

Happy Holidays time :)

[ 22 décembre 2007 19:55 ]
tomsamson

Tomsamson, I have made the test that compares ActiveX player 9.0.64.0 (last moviestar public beta version of the player) and ActiveX player 9.0.115.0 and did not notice any difference in rendering speed.

Note that I only test graphics rendering, I did not use any code from virtual machine VM1 or VM2. Try to do your test again and tell us the result if any interesting.

[ 30 décembre 2007 13:05 ]
Tek

hey there,
sorry for the late reply, haven´t checked the post for a while.
yeah, my test results are still the same, previous two beta players achieve way higher and more constant fps in the browser than the release one for lots of content.
i don´t have time to post an as in depth comparison as you have here now as i´m in the middle of a client project but meanwhile there are way more comments like mine on both emmy´s and tinic´s blogs (see somments here http://weblogs.macromedia.com/emmy/archives/2007/12/flash_player_9_10.cfm and here http://www.kaourantin.net/2007/12/adobe-flash-player-9-update-3-released.html#comments ), also on one of those comments moaning about drastic performance loss of release player compared to beta player Emmy replied this:
“@ Squize - I’ve filed a bug - you’re right.”

[ 11 janvier 2008 8:31 ]
tomsamson

Tomsamson> I could do the test for you, but I need to know :
- the exact version of the beta player you used to compare to the 9.0.115.0 one
- if it was on a mono-core or multi-core architecture (exact configuration is the best)
- if you compare the two version of the player in a web browser or not
- are you using wmode=”opaque”, or wmode=”transparent” init properties of the player
- does the test you’ve made use lot of code to run or it was mostly rendering use

Overall rendering speed is quite better on 9.0.115.0 than 9.0.45.0, I’m pretty sure that the problem cames from changes in code execution not in rendering.

[ 11 janvier 2008 10:02 ]
Tek

Thanks for all these interesting datas !

[ 12 janvier 2008 19:46 ]
Paramoteur

Leave a comment

Javascript must be activated to post

[english] [français] [2007] [2006] [2005] [2004] [2003] [2002]
  wordpress rss rss français rss english xhtml 1.1 css 2.0 wdg