Kishonti GFXbench

Why not just a lower-clocked OMAP4470? It looks like the performance that a ~300MHz SGX544 would have.

Or maybe a Nova A9540? Samsung has a couple of handsets using SoCs from ST-Ericsson (Galaxy Ace 2, Galaxy S Advance).

By the time you assume that it could be any other SoC not from Samsung with lower CPU or GPU frequencies than usual it sounds likelier that it's a lower clocked engineering sample than anything else. The Nova A9540 is supposed to have its 2*A9 at up to 1.85GHz if memory serves well and I'd be very surprised if they'd clock the 544 at only 300-350MHz; that under 32nm. Novathor L8580/28nm wil have its dual A9 clocked at up to 2.3GHz and the SGX544@600MHz.
 
If it's some sort of engineering sample it wouldn't run at final frequencies at least parts of the SoC. Given the GPU results its probably running at somewhere around 350MHz which is quite a bit apart from the target frequency I'd expect for it. The latter has obviously nothing to do with the CPU frequency. We'll see.
My point is that until now now all their modern CPUs have APLLs capable of providing "clean frequencies" and are usually configured so, instead of like frequencies like 1456MHz, which is more what we've seen from Qualcomm and TI. I'm familiar with their platform source code and what's been flying around lately, and this is not any of it.

My bet is on some OMAP4 variant.
 
By the way battery life tests are starting to show up slowly in GLBenchmark2.5. So far the ones I've seen are for iPad3, iPhone4S and Galaxy SIII.

The iPhone4S battery life under as heavy 3D seem comparably low especially compared to the SIII. The latter is more or less comparable to the iPad3 battery life, except that on the SIII with 50% brightness for the display the battery life is less than will full brightness. Anyone have a reasonable explanation for the latter, since I might have missed something considering the display technology on SIII?
 
By the way battery life tests are starting to show up slowly in GLBenchmark2.5. So far the ones I've seen are for iPad3, iPhone4S and Galaxy

I also see ios6 tests appearing for the iphone4s. Unfortunately there is no way I can find to see older tests, but it doesn't look like ios6 with the latest drivers brings much change to the bench tests. I do note that pvrtc2 is not an exposed extension in ios6
 
And here I thought the iPhone4S battery life is too short. By the way Kishonti also started to use a score system based on battery benchmarks. Way too early of course for conclusions due to results for battery being still to few yet still:

http://www.glbenchmark.com/compare....1&D1=Apple iPhone 4S&D2=HTC One X (EndeavorU)

Doesn't the T3 based HTC OneX EndeveaurU have a 1800 mAh battery while the iPhone4S a 1432 mAh battery? If yes it doesn't look all that good for the first in that particular test despite comparable performance in GLBenchmark2.5.
 
S3-4412: Egypt 2.1 60fps 50% : 3h40min

7 minutes less than the website score, ironically, I have no idea how to upload my result, there's absolutely no option for it, or maybe it uploaded by itself. Edit:

GLBenchmark 2.1 Egypt Battery - 50% Brightness 60 FPS Max : 3:41 [h]

It uploads by itself, so I guess I'm in the database, but can't publish it.

I'll run the next tests over the following days.
 
Last edited by a moderator:
I don't think the GL bench battery test makes much sense as (afaik) it doesn't involve setting the brightness of the tested device to the same level using a light meter i.e. one deviced 50%setting != another devices 50% setting etc. Basically this type of test can only work by setting the display brightness with a light meter... Or maybe turning the display of completely..
 
I don't think the GL bench battery test makes much sense as (afaik) it doesn't involve setting the brightness of the tested device to the same level using a light meter i.e. one deviced 50%setting != another devices 50% setting etc. Basically this type of test can only work by setting the display brightness with a light meter... Or maybe turning the display of completely..

For which I assume that it's also valid for 100% brightness there brightness values not being equal either between different devices?
 
Copied from the iphone5 speculation thread, as it might be more relevant here:-

If the gl2.5 results in the following tables are to be believed:-

http://www.pcmag.com/article2/0,2817,2410034,00.asp

Iphone5 is getting slight better offscreen 1080p results that the ipad3(27 v 25.2). Given that reports todate suggest that the A6 doesn't have as wide a memory interface as the A5X, does it suggest that they are clocking @ a good bit over 300mhz, maybe 325mhz or is the custom arm processors helping in the benchmarks in some way ?

A more direct comparison is 4s v 5.

4s gets 11fps offscreen, 5 gets 27fps, and improvement of 2.45. Iphone4s runs 543mp2@200mhz. If iphone5 runs its MP3@325mhz, then performance increase would be x2.43.

4s gets 19fps onscreen 5 gets 38fps, an exact doubling. However iphone5 has 18% more pixels than iphone4s, so a doubling of onscreen performance would be exactly what you would expect from an MP3@325mhz.

The above of course assumes there are no other improvements in the soc that would affect gpu benchmarks.


Update, first result for iphone5 hits the database proper. Only score is off-screen hd, and indeed it is outperforming the ipad3 that is running mp4@250mhz, by about 12%

http://glbenchmark.com/phonedetails.jsp?D=Apple+iPhone+5&benchmark=glpro25
 
Yep. Riterating my post from the iPhone thread, 300 or 325 make the most sense as Tangey also figured from a different angle.

MP3 with a higher clock speed (and higher than Anandtech's guess of 267 MHz considering the fill rate it's achieving) versus MP4 accounts for the slightly lower fill rate but still marginally higher overall performance in GLBench 2.5 considering the slight loss in performance scaling that happens when increasing the core count.
 
Yep. Riterating my post from the iPhone thread, 300 or 325 make the most sense as Tangey also figured from a different angle.

MP3 with a higher clock speed (and higher than Anandtech's guess of 267 MHz considering the fill rate it's achieving) versus MP4 accounts for the slightly lower fill rate but still marginally higher overall performance in GLBench 2.5 considering the slight loss in performance scaling that happens when increasing the core count.

There could be also other factors that could play a minor yet collective role for the higher performance.
 
Kishonti will be releasing another, more detailed GL ES 2.0 benchmark with particular usefulness for comparing to device performance under the DirectX API, in the few cases where an applicable device could be tested.

The new GL ES 2.0 test uses the same "dinosaur chasing a biker" scene and assets from Kishonti's DXBenchmark 2.7 (likewise, a DX version of GLBenchmark 2.5 is also being released.)

http://glbenchmark.com/benchmarks.jsp?benchmark=glb27

On a different topic, indications from the GLBenchmark results and from rumors around are that the next Nexus phone will be an Adreno 320 equipped (APQ8064, presumably) LG super phone with performance near that of the iPhone 5.
 
Last edited by a moderator:
Kishonti will be releasing another, more detailed GL ES 2.0 benchmark with particular usefulness for comparing to device performance under the DirectX API, in the few cases where an applicable device could be tested.

The new GL ES 2.0 test uses the same "dinosaur chasing a biker" scene and assets from Kishonti's DXBenchmark 2.7 (likewise, a DX version of GLBenchmark 2.5 is also being released.)

http://glbenchmark.com/benchmarks.jsp?benchmark=glb27
We seem to be running out of 2.x version numbers. :D

Direct DirectX vs OpenGL ES comparisons will be interesting. It'll add fuel to the fire of DirectX vs OpenGL (ES) API debates and Windows/OS X/Linux OS/kernel debates but differences in status of the drivers on DX vs GL ES will probably be the bigger determining factor, at least initially.

What version of DirectX is DXBenchmark 2.5/2.7 written against? I guess DX9.0c/SM3.0? Since Microsoft is pushing desktop class graphics APIs into mobile I wonder how long having a dedicated mobile OpenGL ES API is necessary if most silicon meets DirectX/OpenGL requirements in order to be Windows/Windows Phone compatible (which negates ES' transistor advantage)? Does OpenGL ES really offer significant power savings vs OpenGL and Microsoft is just accepting a power disadvantage by using desktop DirectX? If it's just a matter of avoiding legacy cruft, the OS could just make only Core profile available and perhaps the silicon could by focused on Core profile as well for transistor savings. I suppose OpenGL ES will still have a role outside of mobile in other embedded applications.

On another note, what happened to the results database for GLBenchmark 2.1? Both GLBenchmark 1.1 and 2.1 seem to been removed from the App Store and Google Play but the GLBenchmark 1.1 database is still available so why not GLBenchmark 2.1? 720p is still the most relevant resolution for smartphones although I suppose GLBenchmark 2.5 is better at capturing the graphics used in newer smartphone games.
 
The 2.5 benchmark suite includes 2.1 tests (just not at 720p for offscreen but 1080p).

***edit:

What version of DirectX is DXBenchmark 2.5/2.7 written against? I guess DX9.0c/SM3.0?

Both benchmarks are coded for win8 it seems and are aimed to get released around the OS release time. DX9.0L3 (DX11 certified DX9.0) sounds likelier than anything else.
 
GLBenchmark2.7 and DX Benchmark2.7 are marked now as "available" on GLBenchmark's homepage. I wonder when the first results will be available.
 
Back
Top