Zelda Twilight Princess too slow with latest Dolphin

[Windfisch]

Hi

I used dolphin 3.5-487 (iirc), and the game ran absolutely fine.
However, using the latest Dolphin-3.5-one thousand and something, the game only runs at half speed or so (same machine!)

Can you please help me to get it work fast again?


Alternatively: 3.5-487 worked fine, up to a gentoo upgrade. But now i'm getting glitches, for example when visiting the cemetry at Kakariko village: when i enter the cemetry, all "grass" textures become black. This persists when leaving the cemetry again, however can be fixed by entering certain (but not all!) buildings.

Can you possibly help me fix this?


Edit: Gentoo Linux, Intel Core i5 3570, NVIDIA GTX 560Ti, 16GB of RAM

Greetings,
The Wind Fish

[Shonumi]

IIRC, you were using a build before the GLSL merge. Try enabling the Hacked Buffer Upload in the Hacks tabs of the Graphics config screen. Also, try disabling VSync if it's enabled. It's gotta a lot more demanding to use VSync with OpenGL since the GLSL merge.

[pauldacheez]

A Gentoo user complaining about performance? Typical. :D

If you're using nouveau, you're almost inevitably gonna have problems since it doesn't have perfect support for all the GL features Dolphin currently uses. Use the binary Nvidia driver, it implements most GL 3.x/4.x features that its Windows counterpart supports and it usually runs a good bit faster. (In the (admittedly incredibly-rare) case that you're a stereotypical http://funroll-loops.info/ kind of Gentoo user, keep in mind that no amount of ridiculous CFLAGS can make up for broken/missing GL feature support or incomplete reverse-engineering in a GPU driver.)

[Windfisch]

yes, the fast working version was before the GSGL merge.
with the hacked buffer upload and vsync disabled, the game becomes playable again, though still a bit slower than before.

i use the binary nvidia driver

and i actually do want vsync, otherwise i'm getting tearing which disturbs my visual experience :/

is there a way of getting it fast AND beautiful again?

btw, my CFLAGS are -march=native -O2 -pipe, this should not do any harm. or should i try more aggressive options?
i'm asking because i get a fairly high CPU usage as well (2 cores fully used), and i'm not sure whether decreasing the render resolution actually speeds it up. (in order to break the stereotype: i don't think this helps)

[pauldacheez]

-O3 should be fine. Feel free to add more you think are safe, since Dolphin's ridiculously CPU-heavy. -flto helps massively on Clang on OS X, but doesn't seem to be doing anything on GCC on my Arch install (it doesn't lengthen linking at all, shouldn't it take forever?).

Lack of vsync is only really noticeable on Linux (goddamn xorg) – I've never noticed any tearing or dropped frames without it on OS X, but on Linux I get tons of dropped frames without it. If other graphics settings don't slow it down at all, by all means, turn 'em up – no reason to let your GPU idle while your CPU does everything. (In fact, some GPUs tend to stay at idle clockspeeds while running Dolphin; upping the graphics settings might kick it hard enough to increase performance.)

Also, if you're using DSP LLE (which you should use, otherwise audio's buggy), check "DSP on Dedicated Thread" to use another of your four cores.

In the case of Twilight Princess, though, you're inevitably gonna have some speed issues on Hyrule Field (especially later in the game), even in the best conditions. With an i5 2500K, you could easily just overclock it enough (to, like, 4.5-4.7 GHz, I think?) to get past that, but a normal 2500 isn't overclockable unless you have one of those ASRock motherboards that does No-K Overclocking, and even then you can only go up ~300 MHz or so before it gets unstable...

[degasus]

There is usually an option for fast vsync: triple buffering
You could also try to enable threaded driver. Nvidia disable it by default on linux.
As both are driver setings, I guess nvidia-settings offers this flags - else google :-P

But in the end, with efb2ram, there will still be a performance drop after glsl merge :-(

[noah]

Newer Dolphin revisions are a lot more demanding on the CPU. I can run a bunch of games with 3.5 release, but with the latest build Wind Waker won't even run at a constant speed.

[Windfisch]

:/

is that going to be fixed, or was it a redesign which slowed it down that much?

would you then recommend using an older version?

[degasus]

It was a redisn in our ogl backend. But it should speed up your setup, so I'm confused. Be sure to try the hacked buffer.

The new design should use much less driver calls and also should be faster, but it has a bigger delay (eg 0.1 frame). So normal renderung speeds up, but efb2ram or efb access (functions which has to wait until the gpu is ready) needs to wait longer :-(

[Windfisch]

hmmm. i just retried it, and it seems to now run fast o.O. actually i have changed nothing. mysterious.

btw, EFB access is the thing which enables the HUD-map on zelda, right? this map laggs as hell and sometimes stops working; it just "playbacks" a short "movie" of this map (i.e. the yellow triangle passes the same way again and again)

(using efb to ram, no caching)

is that fixable?