Guide:- Maximizing the performance of PPSSPP on your Android/Windows device.

[Nickie]

(10-19-2015 05:01 PM)fivefeet8 Wrote:  Unfortunately, the problem is mostly with QC OpenGL drivers.

yeah but that's my problem:

if the bottleneck is on OpenGL, then it means that psp graphics rendering is too much for the adreno ?

if so that's simply not right, the tablet can render much better graphics with more polygons and stuff, and there is virtually no increase or loss between PSP native resolution and 1080p ,
as farahjr there's a couple more frames at best, to me that would indicate that the CPU maxes out before the rendering part....

[Terr0rGam3r]

Wow Thanks This Is Helpful

[fivefeet8]

(10-26-2015 07:29 PM)Nickie Wrote:  If the bottleneck is on OpenGL, then it means that psp graphics rendering is too much for the adreno ?

It means that Adreno OpenGL drivers are bottlenecking the GPU from having more rendering performance. Adreno devices should be just as fast as Nvidia Tegra devices(Tegra4/K1), but they are hindered by drivers. The hardware specification is actually pretty decent otherwise.

(10-26-2015 07:29 PM)Nickie Wrote:  if so that's simply not right, the tablet can render much better graphics with more polygons and stuff, and there is virtually no increase or loss between PSP native resolution and 1080p ,
as farahjr there's a couple more frames at best, to me that would indicate that the CPU maxes out before the rendering part....

I remember Henrik stating that PPSSPP is actually pretty optimized for ARM CPU's.

[Nickie]

(10-29-2015 04:45 AM)fivefeet8 Wrote:  It means that Adreno OpenGL drivers are bottlenecking the GPU from having more rendering performance. Adreno devices should be just as fast as Nvidia Tegra devices(Tegra4/K1), but they are hindered by drivers. The hardware specification is actually pretty decent otherwise.

yeah i get that, but thesn't that imply that even without the whole emulation layer behind the scenes, the same games would be equally slow due to their graphics complexity alone?

that the part i find hard to get, given that even locoroco has big slowdowns at times :|

[fivefeet8]

(10-29-2015 08:27 AM)Nickie Wrote:  yeah i get that, but thesn't that imply that even without the whole emulation layer behind the scenes, the same games would be equally slow due to their graphics complexity alone?

that the part i find hard to get, given that even locoroco has big slowdowns at times :|

If that were the case, then they'd be slow on Tegra devices as well, but they aren't. LocoRoco is full speed on my Tegra K1 and X1 devices. The CPU in Adreno 330/420 devices are similar to the ARM A15/A57 CPU's in Tegra devices.

[LunaMoo]

@fivefeet8 nvidia K1 and X1 are much bigger shifing from fat tablets to stationary console than adreno which sits in very slim low profile devices, comparing them is like saying pomelo is mandarin, worse actually since it's different architecture and we can't really get any oranges out of it aka will have different bottlenecks. Size can pack up more power, marketing BS and syntetic benchmarks also show power, only one of those is real through.

//==============

@Nickie you don't get it, because you look at wrong things.
You can't really tell how complex game graphics is by just looking at the result, you simply can't and there are many reasons for that, from games itself lacking optimization(look at modern indie games for example:]), for them to be made in completely different way on consoles than on device you're using maybe even abusing some proprietary tech which does exist only on psp. Loco Roco for example since you already mentioned it, abuses bezier curves, not sure how light they are on psp, but certainly quite heavy to emulate, you can change accuracy of them in graphics settings, but round shapes will turn into squares.

Could games that would display exactly same result graphics be much lighter? Sure if they would be written specially for that particular hardware and being optimized for it, but that's no longer an emulation.

You're not seeing much difference with resolution change, because you're cpu limited and gpu driver might be a huge part in that. There are other things which could add to it like "multithreading" which most people confuse with free speed boost that in reality will often slow things down especially on some mobiles. Or even your OS turns on some power saving mode.

If you're unhappy with performance I would recommend you to wait until your device get's Vulkan and ppsspp get's a support for that api or at least when this get's finished, which could also help in such situations.

[[Unknown]]

There's a super easy way to tell if you're GPU or CPU limited.

Turn frame skipping up to max and disable auto frame skip.

If your FPS goes to 100%, and audio sounds smooth (the graphics will obviously be terribly choppy), then you are GPU limited. Frame skipping does NOT skip the CPU emulation, only the GPU emulation.

If it's still not able to play the audio at full speed, then you're CPU limited.

That's it; it's easy. It's hard to argue with the fact that frame skip only skips graphical operations (though this includes the driver, etc., which runs on your phone's CPU.)

But yeah. For some weird reason, PSP game authors made the very strange and bizarre choice of optimizing their games for the PSP fixed function hardware, rather than optimizing it for OpenGL ES. Go figure. I guess that's weird in the same way that Android games that run well on tablets are strangely optimized for OpenGL ES, instead of being optimized for the PSP graphics chip. What can you do with game devs? If only they would build their entire business strategies around emulation, right?

-[Unknown]

[fivefeet8]

(10-29-2015 02:50 PM)LunaMoo Wrote:  @fivefeet8 nvidia K1 and X1 are much bigger shifing from fat tablets to stationary console than adreno which sits in very slim low profile devices, comparing them is like saying pomelo is mandarin, worse actually since it's different architecture and we can't really get any oranges out of it aka will have different bottlenecks. Size can pack up more power, marketing BS and syntetic benchmarks also show power, only one of those is real through.

Adreno devices are also in tablets and larger ones than the K1 Tablet as well.

https://gfxbench.com/compare.jsp?benchma...eld+tablet

I know the GPU architectures are different, but the a15 CPU in the Shield Tablet is not so different to the Krait 400 with Snapdragon 80x SOCs. There probably is different bottlenecks that is affecting GPU performance, but from various sources , Qualcomm openGL drivers are comparatively crappy which could be a reason there is more CPU overhead with devices that use them.

[farahjr]

(10-29-2015 04:04 PM)[Unknown] Wrote:  There's a super easy way to tell if you're GPU or CPU limited.

Turn frame skipping up to max and disable auto frame skip.

If your FPS goes to 100%, and audio sounds smooth (the graphics will obviously be terribly choppy), then you are GPU limited. Frame skipping does NOT skip the CPU emulation, only the GPU emulation.

If it's still not able to play the audio at full speed, then you're CPU limited.

That's it; it's easy. It's hard to argue with the fact that frame skip only skips graphical operations (though this includes the driver, etc., which runs on your phone's CPU.)

But yeah. For some weird reason, PSP game authors made the very strange and bizarre choice of optimizing their games for the PSP fixed function hardware, rather than optimizing it for OpenGL ES. Go figure. I guess that's weird in the same way that Android games that run well on tablets are strangely optimized for OpenGL ES, instead of being optimized for the PSP graphics chip. What can you do with game devs? If only they would build their entire business strategies around emulation, right?

-[Unknown]

I could hardly believe that my 2.3GHz are not enouth to run digimon adventure at last playable on my nexus 5, if it wasn't for the fact that the code running on the android is different from the code running on pc.

My I3 3.1 GHZ 2GB RAM could render digimon adventure on 1080 plus acelerate to over 300%.

My Nexus 5 2.3 GHz just can't play this specific game for 2 years now since a bought the device there was no updates that can fix this.

Alread tried low hardware res and low rendering res. Got terrible graphics and no performance upgrade at all....just 1-5%.

That is the heaviest game i have tried so far with ~= 1.5GB The last playable game i finished was Ragnarok and this one had 1GB and could run just on 95-100% and coudn't acelarate it witout greate fps loss.

May be this have something to do with the dificult of Digimon Adventure too.

I'm looking for this to get better since ever. I'm a dev but not a pro at all so I can't help.

[fivefeet8]

(11-12-2015 06:23 PM)farahjr Wrote:  My I3 3.1 GHZ 2GB RAM could render digimon adventure on 1080 plus acelerate to over 300%.

My Nexus 5 2.3 GHz just can't play this specific game for 2 years now since a bought the device there was no updates that can fix this.

2.3 Ghz on a Nexus 5 is not the same as 2.3 Ghz on the i3. They are different architecture CPU's with differing IPCs.

[[Unknown]]

Well, there are some things the 2.3Ghz Krait can do better than a 2.3Ghz i3 could. There are other things it's worse at. In general, it's probably net worse.

But again, most important is the graphics drivers. The fact that you can get faster speeds with frameskip again proves - your problem isn't (specifically) your CPU speed. It's probably the drivers. If the same driver code was running on your desktop, you might get just as bad performance.

-[Unknown]

[the end]

There some new peature on latest build like *combo key stub
But i just dont get it how it work

[farahjr]

(11-12-2015 09:12 PM)fivefeet8 Wrote:  2.3 Ghz on a Nexus 5 is not the same as 2.3 Ghz on the i3. They are different architecture CPU's with differing IPCs.

I know this perfetcly well.

(11-13-2015 05:45 AM)[Unknown] Wrote:  Well, there are some things the 2.3Ghz Krait can do better than a 2.3Ghz i3 could. There are other things it's worse at. In general, it's probably net worse.

But again, most important is the graphics drivers. The fact that you can get faster speeds with frameskip again proves - your problem isn't (specifically) your CPU speed. It's probably the drivers. If the same driver code was running on your desktop, you might get just as bad performance.

-[Unknown]

But why when I reduce the hardware resolution i can't achive a good performance when my i3 can propably rape the emulator if i put the resolution at a low level and them acelerate the game....i can get more then 1200% spd.

In other hand hardware resolution does nothing at my Nexus 5. May be the emulato can't make the Nexus stop using 1080 on native android resolution?

As far as i aways thinked my Nexus 5 have a native 1080P res wich probably can slow down things, but when I saw the new Hardware Res option on the app I Tought that this option would surpress my 1080P screen.

I was propably wrong. But is there a Way to do that?

[sonic68]

hi guys, i have noticed a decrease of the sound quality since 1.0, playing on the previous version had a better audio with less stutering, did you noticed that ?

[[Unknown]]

(11-17-2015 03:31 PM)farahjr Wrote:  But why when I reduce the hardware resolution i can't achive a good performance when my i3....i can get more then 1200% spd.

From a technical perspective, resolution and mhz are just numbers. For example, if I said "all tall people are ALWAYS faster runners than anyone shorter than them," you would know I was wrong. That's simply not true.

Running involves a lot of things - stride (which is affected by height), breathing, strength, endurance / stamina, diet, practice, even posture. It's not just one thing.

So just because you lower the resolution (affect one factor), doesn't mean that you'll suddenly have the game run so much faster. That's still only one thing - and what your graphics driver is having problems with probably ISN'T resolution.

Note: SOME drivers do have issues with resolution. Some video cards have issues with resolution. But just like some runners could improve in different areas, so too can some drivers and video cards improve in different areas. Don't zero in on resolution as the only thing that can be a problem.

(11-19-2015 12:20 PM)sonic68 Wrote:  hi guys, i have noticed a decrease of the sound quality since 1.0, playing on the previous version had a better audio with less stutering, did you noticed that ?

If you are not getting 100% stable, you will not have good sound quality. First, turn on frameskipping with auto, as it will level out the ups and downs.

As far as it getting worse, go back and try some older git builds from the download page. Try to find where it got worse - what we really need is the first one it was slower in, and the last one it was faster in. That way we can see what was changed between those two builds.

That being said, it might just be that we were ignoring some effect it sometimes rarely did, and now we aren't. Or it might also be that you go back, and it's still slower - this would mean it's probably some setting you changed and forgot about (or didn't realize would impact speed.)

-[Unknown]