Android Tablets Forum banner
21 - 40 of 43 Posts

·
Registered
Joined
·
6,731 Posts
Citadel looks gorgeous (how gorgeous an engine could look on a mobile platform, I'm used to PC gaming so every game sucks on tablets in my point of view).

I guess that game doesn't have that powervr gpu in its database and when it queries libegl it answers, "Yo man! I'm POWERVR_SGX544", that games answers "Sorry man, you're not on the VIP list"


Anyway A31 had a noticeable smoothness increase with "sdk 3.2" kernel/hwcomposer/surfaceflinger combination, now it's working good, but for now only Ainol Spark (with a leaked unofficial build) and Onda (kudos to them because they are really fast in updates) have a 3.2 based firmware.

You can run that kernel (you need it because of some updated kernel stuff that do the magic) on another device but of course something specific might not work (touchscreen and camera mostly because every customer makes slightly changes to the reference board code, for example spark gt82x kernel driver is modified to load a firmware, and ainol's driver don't load on a generic kernel zImage because of those firmware missing symbols)

Going back to amlogic I really hope they did these two things (I already mentioned it to them via mail, if I ever met them I'm going to tell them face to face as well) because it will be a huge step forward to all the other chinese socs:
- move to openmax video playback (like rockchip and atm do) instead of hacking libmediaplayerservice and the java framework
- most important thing: use a real deep sleep where the socs goes in a very deep sleep power state, where it turns off pretty much everything non needed (sound, touch, everything but ram and usb/sdio for wifi) instead of using an S3 (suspend to ram) approach. Any chinese soc can do that (never tried mediatek stuff but I guess they will support it since they use their soc in phones too), that's why if the device enters deep sleep wifi won't survive unless you hold a wakelock, that anyway consumes more power because the soc is powered and doesn't have a very low power state
 

·
Registered
Joined
·
2,176 Posts
I am now downloading Epic Citadel on my RK3066 Powered N90 II (with 1024x768 screen). It has a proper build.prop mod so that it looks like an SGS II phone (same Mali-400 GPU so it gets the correct version of apps).

First thing to note, the download for the app on my N90 II is 154 MB... On the A31 powered Momo 19 it as 98 Mb. So clearly there are different version of the app for different platforms...

Benchmark results on "ultra-high" quality setting on the Momo 19 = 32.8 FPS average. Considering it is 1024x768 that isn't staggering by any stretch. I am going to bump quality down a notch and try again....

I will also note that the RK3066 installed the bloody app about 5x faster. App installs rely heavily on the processor as it involves unpacking compressed files. As my guess is that the app install process is optimized for dual-core at best and the RK3066 individual cores are > A31 cores that = faster install times.

I also noted that on my RK3066 powered unit, "ultra-high-quality" mode was not available.

What I did was set both units to High Quality Mode and I ran the benchmark side-by-side at the same time and compared visual quality as well as results... Here are the results:

RK3066 - High Quality - 42.4 FPS
A31 - High Quality - 48 FPS

The A31 edges out the RK3066 and to be fair it was slightly smoother all around running the bench. However, considering what is is reported to have done at FHD resolutions I expected 2x the score. I ran it in "performance mode" (i.e. low graphics quality) and got it to 50.4 FPS. Again, very surprised.

The Mali-400 also produces noticeably better visuals. Textures in particular are a little clearer (not as fuzzy, probably better Antiostropic filtering and possibly higher-res textures being used). Contrast also seemed better (shadows and lighting were much better but this might have been due in part to the screen on the N90 II being so much better.)

To be fair to the A31 - I did notice some effects in place not present on the RK3066 - Light-Halo's/Flares and Chimney smoke to be specific. In Ultra quality mode the differences are even greater with things like fire showing up. I would be interested in getting a Ultra Quality setting available on the N90 II. I might switch the build-prop on an SGS III and see what happens.

I really think a lot of the noted visual quality is down to the screen being so much better on the N90 II, that should be said because the screen in the Momo 19 isn't giving a fair showing for the A31.

The big takeaway isn't the visual quality differences though as it is the closeness in benchmark performance. At this resolution I expected the A31 to be breezing through the benchmark at
+ FPS at least on low - high quality.

Regards,

Roman
 

·
Registered
Joined
·
2,176 Posts
Citadel looks gorgeous (how gorgeous an engine could look on a mobile platform, I'm used to PC gaming so every game sucks on tablets in my point of view).

I guess that game doesn't have that powervr gpu in its database and when it queries libegl it answers, "Yo man! I'm POWERVR_SGX544", that games answers "Sorry man, you're not on the VIP list"


Anyway A31 had a noticeable smoothness increase with "sdk 3.2" kernel/hwcomposer/surfaceflinger combination, now it's working good, but for now only Ainol Spark (with a leaked unofficial build) and Onda (kudos to them because they are really fast in updates) have a 3.2 based firmware.

You can run that kernel (you need it because of some updated kernel stuff that do the magic) on another device but of course something specific might not work (touchscreen and camera mostly because every customer makes slightly changes to the reference board code, for example spark gt82x kernel driver is modified to load a firmware, and ainol's driver don't load on a generic kernel zImage because of those firmware missing symbols)

Going back to amlogic I really hope they did these two things (I already mentioned it to them via mail, if I ever met them I'm going to tell them face to face as well) because it will be a huge step forward to all the other chinese socs:
- move to openmax video playback (like rockchip and atm do) instead of hacking libmediaplayerservice and the java framework
- most important thing: use a real deep sleep where the socs goes in a very deep sleep power state, where it turns off pretty much everything non needed (sound, touch, everything but ram and usb/sdio for wifi) instead of using an S3 (suspend to ram) approach. Any chinese soc can do that (never tried mediatek stuff but I guess they will support it since they use their soc in phones too), that's why if the device enters deep sleep wifi won't survive unless you hold a wakelock, that anyway consumes more power because the soc is powered and doesn't have a very low power state
To be totally fair I haven't updated firmware at all on the Momo 19 - just pulled it out of the box and started using it. There is a very real possibility that firmware is the limiting factor in my testing of the A31 and the Momo 19 is probably not a good representative of the chipset.

It is also interesting to note that there are definitely more "visual fx" represented by the A31 than what is being shown on the RK3066.

Real-world use of the application - both felt smooth but the RK3066 did look nicer. However, as noted before, the screen on the Momo 19 really sucks.

Kind Regards,

Roman
 

·
Registered
Joined
·
6,731 Posts
it's a faux myth that build.prop influence a properly written game because the query is done to the EGL library. Build.prop spoofing is useful ONLY for EA/GL games that perform a check on model_id that grants (or not) the download/start of a game.
Some poorly written stuff might rely on build.prop but it's a very bad way to get into that way.
 

·
Registered
Joined
·
2,176 Posts
it's a faux myth that build.prop influence a properly written game because the query is done to the EGL library. Build.prop spoofing is useful ONLY for EA/GL games that perform a check on model_id that grants (or not) the download/start of a game.
Some poorly written stuff might rely on build.prop but it's a very bad way to get into that way.
In my experience, primarily on the A10, I would get completely different packages particularly for Gameloft stuff (as you noted) based on build.prop.

It would be interesting to see if a build.prop change on the A31 to another chipset (what else uses PowerVR on the android side? TI does I think?) would change the download size of the Citadel app.
 

·
Registered
Joined
·
2,054 Posts
When I got into the game I noticed instantly that the graphics absolutely suck. The textures and geometry complexity is ultra-reduced. I have seen this same thing happen on an Action ATM-7029 device I tested which has the GC1000 GPU in it. I tried NOVA 3 for kicks and it was the exact same story. My guess is that the devices get detected by the play store as "openGL - generic" or whatever and it gets the ultra plain version of the game.
ARM came out with ASTC compression which is really great and royalty free. Maybe in OpenGL ES 3? Have to check.
http://en.wikipedia.org/wiki/Adaptive_Scalable_Texture_Compression
http://hexus.net/tech/news/graphics/43397-opengl-43-opengl-es-30-astc-compression/

Only thing that comes to mind is texture compression affecting image quality. Vivante & ARM GPUs only support ETC compression. PowerVR supports their own PVRTC which gives better compression but with greater loss in detail. A31 seems to be downloading and running PVRTC version of Citadel. That would explain why textures look worse (ie. highly compressed) and why it may run smoother (ie. using less memory bandwidth).
http://blogs.arm.com/multimedia/643-astc-texture-compression-arm-pushes-the-envelope-in-graphics-technology/

No idea why textures would look worse on GC1000 though. Could be lack of features? Lack of precision (ie. less math)? I believe one of those maybe causing the difference you see.

@Roman2025, probably best to test both tablets with SGS2 spoof. They will download the generic OpenGL version (without special texture compression) of Citadel which should work on both devices and avoid getting PVRTC (98MB) version for A31. The quality maybe closer if you compare the two with the same game textures.

For comparison,
Yuandao N80 - RK3066, 1024x768
Citadel: High Quality - 30.7 FPS

Just want to show how newer and better optimized firmware, like found in N90-II, can affect 3D gaming performance. ie. N90-II is doing better in Citadel test with exact same specs to my N80.

Also, I cannot say and have no idea if Mali400 & SGX 543/544 match up in terms of features without checking out their specs. Only 3 reasons you would notice difference, different textures, less supported features and/or less precision.

Thanks for comments and testing guys!!!

I find it interesting that Exynos 5250 uses Mali-T604 but latest 5410 will be going with SGX544MP3. Also, OMAP5 is going with PowerVR too but SGX544MP2.
List Of Exynos SOC
http://en.wikipedia.org/wiki/OMAP5#OMAP_5

8726-M8 should do pretty well but I think Mali4XX is getting too dated and prefer seeing something with MaliT604 instead. I just checked Exynos 5410 (SGX544-MP3) VS Exynos 5250 (MaliT604-MP4). 5410 has higher fillrate by 32% but 5250 has higher triangle count by 59%. 5410 & 5250 both bench fairly close to one another in offscreen tests. I avoided comparing onscreen because resolution differences between devices.
http://gfxbench.com/compare.jsp?D1=Samsung+GT-I9500+Galaxy+S4&D2=Google+Nexus+10&cols=2

That tells me that OMAP5 will bench somewhere around Mali450 or Mali400. Only SGX 543/544-MP3, MaliT604 & Tegra4 GPU offer powerful choices today for "tablets".

Qualcomm Adreno GPUs - are only found in Snapdragons and these chips are not used in any tablets
Nvidia Geforce ULP - only found in Tegra chips and mostly used by high end tablets because of cost

That just leaves Chinese chip makers with following GPUs to use:
-PowerVR Series 5XT
-ARM MaliXXX
-Vivante GCXXXX - worst of the mobile GPUs. Should avoid them. No chip makers are using them anymore except for i.MX6. Very hard for them to compete with the other embedded GPU makers.

Edit1:
ASTC appears to be part of OpenGL ES 3 as optional feature for now
http://www.anandtech.com/show/6134/khronos-announces-opengl-es-30-opengl-43-astc-texture-compression-clu/4
 

·
Registered
Joined
·
2,176 Posts
At this point I am wondering if we are going to see a device based on this chipset before the end of the year?
 

·
Premium Member
Joined
·
1,026 Posts
What do you think of this guy MTK 8389, is it a good idea to buy a tablet based on it ?
I'm in the process of getting a few tablets with these chips in my hands. I've also contacted MTK to see if they will help with kernel sources for development. I haven't been active much on these forums lately. Hopefully something kick starts my interest in tablets again. I love Amlogic but I want them to just pop out this chip already. I'm bored with the dual-core. I'm sure CT is too.
 

·
Registered
Joined
·
2,176 Posts
Still not a word on the 8726-M8 and we are are now in the 4th Qtr... Rockchip is slated to release the 32XX series (Cortex-A12, Mali-T624) 2nd Qtr. of next year. If they are able to get the M8 into devices being shipped in the west by Christmas they will only have about 6-months before being thoroughly eclipsed
. Not a strong showing by AmLogic at this point unfortunately.

-Roman
 

·
Registered
Joined
·
122 Posts
I read that they changed the name of the chipset to M802, and they were releasing other versions with other numbers (M801, 803 and so on).
 

·
Registered
Joined
·
2,054 Posts
1.5-2.25X better GPU performance depending on test. 1.5X for Egypt and 2-2.25X for T-Rex & other tests.

Amlogic says max of 2 GHz CPU but tested version was only at 1.7 GHz.

I believe 8726-M8 tablets are due to come out by end of Oct or in Nov. Very sure more info will be provided during the trade show (Oct 12-15).

Also, Amlogic 8726-M8 slightly outperforms the Voyo A15 (MaliT604 - 5250). MaliT604 has very strong triangle count but lower fillrate. Both of which impact gaming performance.

RK32XX will take performance to the next level for Chinese tablets since will use much more powerful T624 GPU & Cortex A12. We will not see any RK32XX tablets until about June 2014.
http://www.tgdaily.com/hardware-features/73250-rockchip-designs-a-quad-core-arm-cortex-a12-rk32xx-chip
 

·
Registered
Joined
·
2,054 Posts
The compare links were not working properly because Amlogic was added in. ie. URL changed. Re-posting them below with new URLs.

U30GT-2 (RK3188 - MALI400) VS M802 K200 (8726-M8 - MALI450)
RK3188 VS 8726-M8

Voyo A15 (5250 - MALIT604) VS M802 K200 (8726-M8 - MALI450)
5250 VS 8726-M8

Edit1: Updated this post
Edit2: Updated my previous post too with better info
 

·
Registered
Joined
·
2,054 Posts
m802 K100 (2048x1536) benchmarks can be seen here:
http://www.padnews.cn/?p=2761

CPU speed set at 1.9 GHz. Final product maybe at 1.6-1.8 GHz?

Antutu 4: 25,050
NenaMark2: 58.3 FPS
Sunspider: 821.9 ms
Velamo: HTML5: 1719 Metal: 739

Prototype tested. 1.5 (or maybe 2) months before see it in actual China tablets?
 

·
Registered
Joined
·
2,054 Posts
Still no 8726-M8 tablets out? Not sure what is going on.

Amlogic has taken very long to release their newest chips. Heard they were trying to improve heat issues to avoid hot chips like 8726-MX.

m802 looked very interesting and just waiting to see which tablet makers will use it. I will give it to late Nov. and hope more m802 tablet news comes out by then.
 
21 - 40 of 43 Posts
Top