Suggestions for speed improvement on A8-7600 AMP APU

[necbot]

My computer has a A8-7600 AMD APU as my processor/GPU.  I'm on Arch Linux, so I pulled the very latest version of dolphin (4.0.2.r5396) and I compiled it.  I tried running Mario Kart on it using it's default settings but I was barely able to get above 30 FPS during gameplay.  I set my resolution down from 1080p to 720p and the performance did not improve.  At first I figured that this processor was too weak so I tested the dolphin benchmark from here...

https://forums.dolphin-emu.org/Thread-new-dolphin-cpu-benchmark-no-game-required

I got 10 minutes and 17 seconds as the total time, which puts it on par with a i7-3770k according to the spreadsheet in the benchmarks thread.   So my question is...  Why is the performance so bad with Mario Kart?  Is there a setting I'm missing?  Thanks.

EDIT:
Here is a video of someone playing Mario Kart on an A8-5500, which is a lower spec chip than mine.  He regularly get's up to 60 fps.  I can't get anywhere near this kind of speed.

[Link_to_the_past]

(02-26-2015, 11:05 PM)necbot Wrote: My computer has a A8-7600 AMD APU as my processor/GPU.  I'm on Arch Linux, so I pulled the very latest version of dolphin (4.0.2.r5396) and I compiled it.  I tried running Mario Kart on it using it's default settings but I was barely able to get above 30 FPS during gameplay.  I set my resolution down from 1080p to 720p and the performance did not improve.  At first I figured that this processor was too weak so I tested the dolphin benchmark from here...

https://forums.dolphin-emu.org/Thread-new-dolphin-cpu-benchmark-no-game-required

I got 10 minutes and 17 seconds as the total time, which puts it on par with a i7-3770k according to the spreadsheet in the benchmarks thread.   So my question is...  Why is the performance so bad with Mario Kart?  Is there a setting I'm missing?  Thanks.

EDIT:
Here is a video of someone playing Mario Kart on an A8-5500, which is a lower spec chip than mine.  He regularly get's up to 60 fps.  I can't get anywhere near this kind of speed.

That is not the latest dev version and you can try windows in case there is a linux driver speed issue. You shouldn't use efb to ram for good speed in this game.

[Anti-Ultimate]

Quote:AMD

You'll get more performance on Windows. AMD Linux drivers suck and the Open Source ones are actually on the edge of being better than the proprietary ones, although not as fast as Catalyst on Windows.

[bomblord]

what are your settings such as IR etc? Do you have dual core speedup turned on?

[degasus]

(02-26-2015, 11:05 PM)necbot Wrote: I got 10 minutes and 17 seconds as the total time, which puts it on par with a i7-3770k according to the spreadsheet in the benchmarks thread.   So my question is...  Why is the performance so bad with Mario Kart?  Is there a setting I'm missing?  Thanks.

This doesn't mean your CPU is on pair with the fasted Ivy, this just shows the speedup we had in the JIT in the last year. Just try to build 4.0 and see the difference

But here I think it might really be the GPU thread which slows down. Mesa is mediocre, fglrx/catalyst are a lot faster. Are you using the lastest linux kernel + mesa version? Just post "uname -a" and "glxinfo | grep RENDER". They also did increase a lot in the last time.

[KHg8m3r]

To summarize, you need to grab the latest dev build, not 4.0.2, and try lowering your IR

[necbot]

@Link_to_the_past
I suspect Windows might run this better but I'm going to use that as my nuclear option  Right now all my settings are the default, EFB is set to texture, not RAM.

@bomblord
All my settings are default.  Dual core speedup is turned on.  Internal Resolution is 1X Native.

@degasus
Markus!  Thanks for the clarification. I thought it might be something like this.  The results of that benchmark make sense in hindsight.  I mentioned the i7-3770k because I have another system like this as my primary workstation.  On a side note, after reading your post I went and ran that benchmark on my workstation (which also has an Nvidia GTX 750 Ti and is running Arch linux) and I got 6 minutes 57 seconds!  That is a great speedup from the 11+ minutes on the benchmark spreadsheet.

uname -a gives...

Code:

Linux kodi 3.18.6-1-ARCH #1 SMP PREEMPT Sat Feb 7 08:44:05 CET 2015 x86_64 GNU/Linux

glxinfo | grep render gives...

Code:

direct rendering: Yes
   GLX_MESA_multithread_makecurrent, GLX_MESA_query_renderer,
   GLX_MESA_multithread_makecurrent, GLX_MESA_query_renderer,
OpenGL renderer string: Gallium 0.4 on AMD KAVERI
   GL_NV_conditional_render, GL_NV_depth_clamp, GL_NV_packed_depth_stencil,
   GL_NV_conditional_render, GL_NV_depth_clamp, GL_NV_fog_distance,
   GL_OES_element_index_uint, GL_OES_fbo_render_mipmap,

I will try the catalyst drivers as soon as I can get them working on this computer.  I've run into issues with my system freezing at random after loading the catalyst driver (even if xserver is not running) so I'm over on the arch forums trying to figure out why.

@KHg8m3r
Will do!  I thought the package in the AUR gave me the latest or close to it.

[lok1]

The problem is mesa. I've tested both drivers on my a10-7850k, and my old a8-6600k. With fglrx I had no problem running running Mario Kart Wii at 1.5x, but when I used mesa I got around 30-35 fps at 1x. You should switch to fglrx for dolphin for now mesa just isn't ready yet.

The problem you're having with catalyst might be due to it not yet supporting kernel 3.18. Try switching to 3.16, or wait for a new catalyst to come out. There's supposed to be a new driver this week, and it might support the newer kernel versions.

[degasus]

(02-27-2015, 11:47 AM)lok1 Wrote: The problem you're having with catalyst might be due to it not yet supporting kernel 3.18.

Or Xorg1.17 as you're on arch. AMD seems to not care about fglrx. I like them for caring about mesa through, but the performance isn't that good. Common shader performance is fine (so usual games run fast), but mesa still have a bigger API overhead which is especially bad for dolphin :/

My suggestion is to just use your workstation Also try the nidia __GL_THREADED_OPTIMIZATIONS=1 environment variable configuration. It's the default on windows for quite a long time and it's an impressive speedup there.

(02-27-2015, 10:03 AM)necbot Wrote: I thought the package in the AUR gave me the latest or close to it.

The aur-git package you've used is fine. Just don't use the one without -git

Edit: btw: where did I use the name "Markus" ?

[necbot]

@lok1
Good to know that there is hope of getting this running at reasonable speeds on linux. I may just install the latest version of Ubuntu since I've never had problems with the proprietary drivers on that distro.

@degasus
Yes, I'm using the one with -git. When I install the Catalyst drivers in Arch I'm forced to downgrade Xorg to 1.16 (I'm using Vi0l0's repos as suggested in the arch wiki). I blacklist radeon, generate a new xorg.conf file, remove mesa-libgl, disable kernel modeset, etc. etc. The system boots fine, but it freezes for no apparent reason. It is a new install of arch so I don't have the 3.16 kernel package to downgrade to. Regarding your name... You didn't. I figured that was your name because when I was troubleshooting this I read all about your contributions to the project in the driver hall of fame/shame article so I figured you were Markus, my bad it I got that wrong. Thanks for your help. I think my next step might be to try to compile the catalyst driver for xorg 1.17 and kernel 3.18. I remember the arch wiki saying that there is a AUR package but that I'd have to recompile whenever I update.