Qualcomm Krait & MSM8960 @ AnandTech

[Nebuchadnezzar]

Then Kishonti needs to change testing methods. The medium results aren't much of a help.

That's what the battery performance test is for.

 

[Ailuros]

That's what the battery performance test is for.

* I'd say that the majority these days looks for GLB3.0 results. Battery and long time are with TRex onscreen.

* More than often I see early results of a device that suggest a high amount of throttling in that one and then after some time it just disappears and in the end not a single device seems to throttle after all.

(Frankly it could mean that these are early results where the vendor still makes final sw adjustments, but still I wouldn't suggest that then these results should be allowed to get published...)

Here's one example: http://gfxbench.com/device.jsp?benchmark=gfx30&os=Android&api=gl&D=Samsung%20SM-G850

Yes it's just one run for the long time performance test, but 23.7 there vs. 48.2 fps in T-Rex offscreen suggests a helluva lot of throttling.

My suggestion would be that the battery test should be from the main test (in this case Manhattan) both on - and off screen and no result to be allowed to get published without those. Yes it takes it's time obviously to get run and especially twice.

Even then the battery test is an indication IMHO since I don't trust battery readings on some Chinese devices f.e.

 

[Ailuros]

http://semiaccurate.com/2014/12/04/qualcomm-shows-off-mare-parallel-api-runtime/

I like it; it has a Mantle*kickintheahem*DX12 kinda taste

 

[OlegSH]

I'd rather say it has a *wecantwriteagoodgldriverpleaseuseouryetanothernoexpiriencedapiifyoudontwanttofightagainstthewallwithourshittydriver* kinda taste I hate it

 

[Laurent06]

Indeed the proliferation of proprietary graphics API is a concern. The next step will be assembly programming for everyone on the CPU.

 

[Deleted member 13524]

I personally see it more like a "here's something we can use as a better performing API until OpenGL NG gets finally released" initiative.

 

[Ailuros]

I personally see it more like a "here's something we can use as a better performing API until OpenGL NG gets finally released" initiative.

Along that line yes...I wonder if it would actually bring something for todays applications....

 

[Laurent06]

I personally see it more like a "here's something we can use as a better performing API until OpenGL NG gets finally released" initiative.

The problem is that each HW vendor is providing his own API that matches better his own HW (or worse an OS+HW combination as for Apple Metal). So as a SW developer what will you do? Port to each API? Or hope your preferred middleware supports them all? But if these initiatives force Khronos to move faster, that's nice

Don't get me wrong, I love low-level stuff, it's just not productive any more.

Has there been some feedback from middleware vendors?

 

[liolio]

I'd rather say it has a *wecantwriteagoodgldriverpleaseuseouryetanothernoexpiriencedapiifyoudontwanttofightagainstthewallwithourshittydriver* kinda taste I hate it

I guess it is a cheap PR cover up for their bad drivers, it is easy at this time as there are really few really demanding games and few serious means to measure performances.
If more serious software/games hit mobile devices appstore that type of cover up won't help against Nvidia. Nvidia has a ace up its sleeves.