Graphics API backend

[Streeter]

DX9 is the fastest, yet least accurate/stable, while OpenGL is the opposite (and DX11 being somewhere in between).

So, does this still hold true with the 3.5 release? How much has changed in terms of performance/accuracy/stability since?



(By the way, this is by far my favorite emulator out of the many that I use. If only I were better at coding!)

[Shonumi]

(01-02-2013, 03:58 PM)Streeter Wrote: DX9 is the fastest, yet least accurate/stable, while OpenGL is the opposite (and DX11 being somewhere in between).

More accurately put, DX9 is generally faster than OpenGL or DX11. Whichever one runs faster for you depends on which games you're playing, your hardware, and the drivers you're using. Sometimes DX11 actually runs faster for some people. I've also seen a handful of games other users play where DX11 performance tanks while OpenGL stays in the middle. Other times OpenGL performance delivers more FPS in certain games over the either DX9 or DX11. In short, it depends; generally, your statement holds true, though not in every case.

(01-02-2013, 03:58 PM)Streeter Wrote: So, does this still hold true with the 3.5 release? How much has changed in terms of performance/accuracy/stability since?

Probably, not much has changed. Accuracy has gone up since 3.0; DX9 is still the least accurate, being behind DX11 and OpenGL. Their overall performances should be relatively the similar, though I remember a few commits that reasonably sped up the DX9 and DX11 backends a few months ago (can't find it atm).

[Streeter]

Oh, on the Just-in-time recompiler, is the experimental engine faster and/or more accurate? I get that it's less stable, but what's the development status on it? Do features from there get branched into the defualt compiler?

Another tidbit, I have a quad core processor, dual threaded (8 logical cores). I do not quite get the advantages of locking the processes instead of letting my OS handle it. In what situation should I enable locking threads to cores?

[NaturalViolence]

Quote:Oh, on the Just-in-time recompiler, is the experimental engine faster and/or more accurate?

Accuracy should be slightly lower. It hasn't been tested as thoroughly as JIT. Although you're unlikely to notice any difference between the two for 99.99% of games.

As for faster, it depends. Usually no. With 64 bit builds JIT is almost always faster by a decent margin. On 32 bit builds JITIL is faster on some games and slower on others. They are generally pretty close to each other performance wise on 32 bit builds.

Quote:I get that it's less stable, but what's the development status on it?

Development Status: He gave up and stopped caring.

It was a failed experiment. He thought it would yield a significant performance improvement, and when that turned out to not be the case it was abandoned.

Quote: Do features from there get branched into the defualt compiler?

What do you mean by that?

Quote:Another tidbit, I have a quad core processor, dual threaded (8 logical cores).

Please avoid using the term "dual threaded" that way. That's not what it means.

Quote:I do not quite get the advantages of locking the processes instead of letting my OS handle it. In what situation should I enable locking threads to cores?

When you want to see how much cpu load each thread is eating up. It's generally a bit slower than letting the OS handle it.

[admin89]

If your CPU is an i5/i7 mobile CPU or locked i5/i7 desktop CPU (can't overclock) , "lock threads to cores" (2 cores ) will allow your CPU turbo boost higher . For example :
i7 740QM turbo boost :
_2.9GHz (1 core active)
_2.5GHz (2 cores active) <- This
_1.7GHz (3 or 4 cores active)
That option also help Donkey kong CR : http://wiki.dolphin-emu.org/index.php?title=Donkey_Kong_Country_Returns

[Streeter]

Considering that I have an i7 2600k @4.2Ghz, and am using the LLE recompiler for accurate audio, DSP LLE on Thread would speed things up, correct?

Also, how well does the OpenMP Texture Decoder work? Is it glitchy/innacurate? How likely/often does it result in a speedup?

[RachelB]

(01-02-2013, 10:21 PM)Streeter Wrote: Considering that I have an i7 2600k @4.2Ghz, and am using the LLE recompiler for accurate audio, DSP LLE on Thread would speed things up, correct?

Also, how well does the OpenMP Texture Decoder work? Is it glitchy/innacurate? How likely/often does it result in a speedup?

Not likely. You can try it yourself if you want, but you'll likely find it makes no difference.

I'm not sure if openmp texture decoder is buggy, but i've never seen a speed increase from it either.

[admin89]

Quote:DSP LLE on Thread would speed things up

Dolphin will use 3 cores if you enable that option
Small speed up , don't hope much

Quote:OpenMP Texture Decoder

Reduce micro-stuttering
Very small speed up in some games .

[DefenderX]

DSP LLE on thread will provide a small speed up indeed, but can cause crashes in games which use Zelda uCode.

[Streeter]

How accurate is per-vertex vs per-pixel depth calculations? I haven't seen any noticeable difference so far.

Also, 3 is the maximum number of cores/threads that dolphin utilizes, right?

And why are certain features like the JITIL recompiler and OpenCL texture decoder even available as options? It sounds like they should be hidden from normal users by default.