cpu clock's impact on fps [GT-I9300]

[zarles1]

Hello everyone,
I've been curious about the influence my cpu clock is having on the actual fps ingame. So I made a little list.

Testing conditions:
-Galaxy S3 GT-I9300
-CF-rooted
-SiyahKernel s3-v1.9.1
-Stock Android international 4.1.2.
-Underclocked with NS Tools
-Untouched GPU Clock
-PPSSPP build: v.0.7.6-805-g2ef4a05

Settings:
-Skip Frames: no
-Enable Sound: yes
-Vertex Cache: no
-Hardware Transform: yes
-Stream VBO: no
-Media Engine: yes
-Linear Filtering: no
-Mipmapping: no
-Buffered Rendering: no
-Display Raw Frambuffer: no
-Anisotropic Filtering: no
-Texture Scaling: no

-Dynarec (JIT): yes
-Fast Memory: yes
-Encrypt Save: no

How I proceeded:
-freeze unnecessary apps via App Quarantine (such as Youtube, gmaps, gplay, g+, WhatsApp,etc.)
-kill tasks with ES-Taskmanager
-clear RAM in Android
-wait a few seconds

I ran every test twice loading the same savestate:
Practically I just ran around for 2-3 minutes dodging this monster as I did here:
http://www.youtube.com/watch?v=Pu1bRZ6YHHw

Here's the list:
500 MHZ - 35-42 fps - avg. 40
800 MHZ - 40-45 fps - avg. 43
1000 MHZ - 48-53 fps - avg. 51
1200 MHZ - 58-62 fps - avg. 60

As I'm limited to 60fps it didn't make sense to clock at stock 1,4 or even 1,6-1,8 GHZ. The "unlimited fps"option doesn't really work authentically, I guess.

As a result:
People always claim that OCing is totally useless as a mean to improve emulation speed but this little test showed that there actually is a noticable effect.
I'm not a developer so please correct me if I'm wrong.

[sfageas]

Interesting results and i agree overclocking helps a bit(4-5 vps boost)keep in mind that emulator is still in beta and there's still room from improvements in JIT as developers said
I have galaxy s2 4.1.2 siyah kernel overclocked to 1.4(stable)Bloatware free
Good video

[solarmystic]

Great experiment. Thanks for sharing the results.

On windows with a laptop Core 2 Duo CPU @ 2.8GHz, and FPS unlocked to unlimited, the FPS in that same area with Tigrex reaches the hundreds (150-300) though lol.

There's still quite a performance disparity between ARM and Intel processors.

[cloud1250000]

lol, solar, the graphic card in your pc is way better then what we can find in a phone xD
This explain why the performance is better on intel processeur (and the higher clock)

[xsacha]

(05-26-2013 01:28 PM)zarles1 Wrote:  As a result:
People always claim that OCing is totally useless as a mean to improve emulation speed but this little test showed that there actually is a noticable effect.
I'm not a developer so please correct me if I'm wrong.

1200MHz is a very low clock. This is like 800MHz on a Krait or Cortex-A15.

Also, I'm not sure if you made sure of this.. but in a lot of mobile SoCs, the CPU clock is linked to the GPU clock. Overclocking the CPU can overclock the GPU. This would make sense since you only seem to get a 50% performance improvement from a 140% overclock.
However, I think it's simply that the game you tested doesn't use the GPU very much.

In almost every game I play (decent graphics), the CPU is able to generate 60 frames (so you can get 60 VPS with frameskipping) but the GPU is not able to render the 60 frames. So, without frame skipping, you get something lower than 60 FPS.
Examples off the top of my head: Tekken 5, Tekken 6, Soul Calibur.
In these games, I get the same FPS (under 60) at 1.2GHz, 1.4GHz, 1.6GHz and 1.7GHz when GPU clock is fixed at 650MHz.

[Henrik]

Who claims that overclocking is useless? I wouldn't say that It will consume more battery though.

Also, if the OpenGL drivers are a bit sucky, which often is the case on Android, then even the drawing can be CPU limited. Then overclocking the CPU without overclocking the GPU can actually help what looks like "GPU speed". Counterintuitive for sure, performance is complex.

Thanks for your experiments.

[arg274]

very detailed test report indeed!

[zarles1]

Thanks to all of you for your replies.
I just tried out a few Benchmark apps (Quadrant, Nenamark, Basemark ES2 Traji) and with both overclocking CPU(to 1,6 GHZ) and GPU (from 440 to 666 MHZ) using STweaks the results improved by 10-15%.
@XSacha: Thanks for the advise. I checked with STweaks whether CPU-OCing with NS-Tools affects the GPU-clock but it didn't. I'll give other games a try to experiment with OCing^^

Just one more question: Henrik said something about OpenGL drivers on Android. I guess, they're always bound to the Firmware (like Jelly Bean 4.1.2) or is there a way to get enhanced drivers?

[kenny43]

(05-29-2013 05:36 PM)zarles1 Wrote:  Thanks to all of you for your replies.
I just tried out a few Benchmark apps (Quadrant, Nenamark, Basemark ES2 Traji) and with both overclocking CPU(to 1,6 GHZ) and GPU (from 440 to 666 MHZ) using STweaks the results improved by 10-15%.
@XSacha: Thanks for the advise. I checked with STweaks whether CPU-OCing with NS-Tools affects the GPU-clock but it didn't. I'll give other games a try to experiment with OCing^^

Just one more question: Henrik said something about OpenGL drivers on Android. I guess, they're always bound to the Firmware (like Jelly Bean 4.1.2) or is there a way to get enhanced drivers?

GPU drivers can be found inside the kernel, there's suposed to be a newer driver on 4.2.2 ROMs (CM, SlimBeam), as these roms won't display anything without the new driver, I could test that by myself using the older siyah Kernels which worked before the new driver implementation on the newer roms, have in mind that multi-rom kernels as siyah, googy, Devil, will choose the driver to be used depending on the rom you're on, so if you're on 4.1.2 you'll be using the old driver, then switching to something newer as CM 10.1 or SlimBeam, or anything else which needs it, will then trigger the newer Mali driver. (GM had stated how he removed one of the drivers from his kernels do drop CM10 support, as he refused to keep three Mali drivers on a single kernel).

Hope that's useful for you.

[zarles1]

Thank you a lot Kenny!

[kenny43]

You're always welcome