Krait Benchmark Updated, and a question (tin hats welcome)

[stormkroe]

Just pointing out that there was an update to the Rightware Basemark benchmarks that put the Adreno 225 well ahead of the Mali400. This is only two benchmarks updated in the suite, but coupling this with the fact that Krait beats Exynos in Glbenchmark Egypt (despite the massive resolution increase) makes me wonder if there is some software trickery happening on the part of Samsung. Both Apple and Samsung scores SKYROCKET going to the offscreen render, despite the fact that the resolution is dramatically higher (especially in Exynos' case). Apple's score could be chalked up to vsync crippling the native resolution score, but not Samsung. Unless someone can tell me how a score of 34.6 fps @ 800x480 can equate to 42.5 fps @ 1280x720? All other devices in the test turn in lower scores when the resolution is stepped up, just like we'd expect.
Well?

 

[ChronoReverse]

It's not so much software trickery but rather a flaw in the GLbenchmark for offscreen for testing PowerVR's style of GPU.

From what I've read, the actual rendering to screen is something that has non-trivial and significant cost but is entirely negated if only rendering offscreen.


The SGS2 has HDMI out at 720p right? Does it actually render at that resolution? Someone should run GLbenchmark through HDMI if it does to see how it really performs at 720p

 

[stormkroe]

Where can I read about the flaw specific to SGX gpu's, I'm quite interested in this. Also, SGS 2 isn't a powerVR gpu anyway so Apple might be off the hook, my conspiracy-sense is tingling for the way samsung handles the test with it's Mali.

 

[ChronoReverse]

Oh I don't think it's a conspiracy but rather a coincidence that the rendering style gets an unexpected bonus in GLBenchmark's method of off-screen.