LLE is not much slower in WII games ! ( lle benchmark thread )

[rpglord]

I was doing some testing and discovered something interesting :
Whenever I enable LLE in a wii game,I seem to not get any significant slowdown,only couple of fps.
However,if I do the same in gamecube games,there is almost always significant fps drop.

I tested with latest revision,increased internal resolution but left AA disabled. Also efb copies are set to textures.
Everything alse is on default,but in some games I enabled "vbeam" since it gives a nice speed boost ( for example Skies of Arcadia and SMG2 ).
When using LLE,I have enabled "LLE on thread" option.
I have also,for the sake of benchmarking,disabled framelimiter.

So these are my results :

LLE vs HLE:

Gamecube:

Tales of Symphonia:
In town 100 fps vs 200 fps
In field : 35 vs 75
This means LLE is 100% slower in ToS !

Fire Emblem Path of Radiance:
80 vs 110

Metroid Prime:
31 vs 40

Skies of Arcadia:
34 vs 84

Zelda TP:
21 vs 28

And now,wii games :

Monster Hunter tri
39 vs 40

Pandora's Tower
26 vs 28

SMG2:
Starship: 60 vs 65
Flufy Bluff : 52 vs 52
Yoshi Star : 64 vs 67

Last story
18 vs 19

Xenoblade:
42 vs 45

I am using i5 2400 and geforce gtx 460.

So,if someone alse is willing to comment this,or even better contribute his own results ( preferably with a quad core so "lle on thread" could be used ) it would be great.

This could be LLE benchmark thread,so we can see how LLE perform in different games and on different systems.

Even if it doesn't have noticeble impact on perfomance with WII games only on my system,then I can figure out why

[LordVador]

I got these games so I can try. I'll communicate the results as soon as possible (that means when I get time )

[rpglord]

I got these games so I can try. I'll communicate the results as soon as possible (that means when I get time )

Ok thanks

[Shonumi]

A few suggestions to make this benchmark a bit more consistent and reproducible. We should use choose a single build (the "latest" is always changing) across every game. I suggest 3.0, but it could be a newer one if we all specifically decide to use it. Additionally, we should use Starscream's Speed Settings so people are on the same page on what to use, with the exceptions that LLE is set to the recompiler, and LLE on Thread is enabled, and Framelimiting is turned off. If you want to rpglord, you can just take screenshots of your current settings and post them here. We should post screenshots (or videos) for every game, else we have people saying whatever they want without backing it up.

I don't have any Wii games dumped, but I can contribute for GC games, in case it turns out the LLE isn't much slower on Wii and GC games.

[delroth]

I tested with latest revision,increased internal resolution but left AA disabled. Also efb copies are set to textures.

Do your benchmark with minimal graphics settings. What you see here is Dolphin being GPU-bound in most of your Wii tests, so CPU speed does not matter all that much.

[rpglord]

I agree we should use single build and same settings.
So I retested with 3.0 and Starscream's speed settings.
Only exceptions are lle and lle on thread of course,and autolimit off.

As for screenshoots, I created six ( 3 hle vs 3 lle ) for SMG2,but I dont see how is this usable in this case,since there is no way to tell from a screenshoot if hle or lle is beeing used ( there is no option in dolphin to display that anywhere as far as I can see )










This is 3.0 and starscream settings. As you can see not much difference,just higher fps.
This also proves SMG2 doesnt require nothing more then 3.2 sandy bridge,not overclocked cpu's.
Lowest it got with LLE is 59 fps which is just 1 fps short of fullspeed.

I tested with latest revision,increased internal resolution but left AA disabled. Also efb copies are set to textures.

Do your benchmark with minimal graphics settings. What you see here is Dolphin being GPU-bound in most of your Wii tests, so CPU speed does not matter all that much.

I just did benchark with minimal settings,so gpu bottleneck is not the case. It just seems LLE doesnt effect perfomance much in wii games,atleast on my system

[Shonumi]

As for screenshoots, I created six ( 3 hle vs 3 lle ) for SMG2,but I dont see how is this usable in this case,since there is no way to tell from a screenshoot if hle or lle is beeing used ( there is no option in dolphin to display that anywhere as far as I can see )

Screen shots are necessary to make sure people who partipate in the benchmark aren't just making up numbers, even if we can't tell if HLE or LLE is used. It's possible to click on the DSP button during gameplay and see which DSP emulator engine you're using (this is greyed out when running a game, can't be changed) so a screenshot that includes that would help it's authenticity. Ideally, everyone would post one screenshot of HLE and another with LLE for every game tested. Posting just the links would make the thread easier to read instead of having a bunch of screenshots filling up the page.

[rpglord]

It's possible to click on the DSP button during gameplay and see which DSP emulator engine you're using (this is greyed out when running a game, can't be changed) so a screenshot that includes that would help it's authenticity.

How come I didnt remember that
Here is one more from me:


Link to bigger version:
http://i.imgur.com/1eJNL.jpg

[AzureKnight]

Well if I remember correctly the Wii uses streamed audio over the sequenced audio of the Gamecube. In laymans terms something like MP3 vs MIDI. I would think that the sequenced audio would be much harder on the emulator than streamed audio. A good example is if I play Pokemon Colosseum and XD, on my old system Colosseum would lag like hell and have horrid FPS with LLE while Pokemon XD would run at a decent speed, Pokemon XD uses a type of streamed audio while Colosseum uses sequenced. Then again i think Nintendo may still be using sequenced audio on their recent Wii titles, Twilight Princess uses it still which means any other Nintendo game based on the same engine such as Mario games may as well. So in theory you wouldn't expect to get much of a performance hit when using HLE vs LLE on something that is less complicated to emulate. Try running the Wii version of Twilight Princess with HLE vs LLE, should see some very different numbers there. Just thought I'd put that out there, correct me if I'm wrong.

[NaturalViolence]

.......

The GC and Wii both use the same DSP architecture. There is little if anything different about audio emulation between them on the hardware side of things.