PowerVR SGXMP3: Crazy Fast, But Underutilized Today
Evaluating mobile graphics performance is a challenge because there simply aren't many quality benchmarks. The challenge is further complicated by a mobile landscape dotted with different SoCs, operating environments, native resolutions, and real-world games optimized for different architectures. Trying to perform cross-platform comparisons is outright tricky, and if you want to go that far, GLBenchmark is really the only game in town. Although we don't like leaning on just one title to evaluate hardware, a lack of options forces our hand here.
GLBenchmark allows us to normalize things to a certain extent. The suite includes a number of tests that run off-screen, pinning the output at 1280x720. And while that helps us isolate the capabilities of each SoC, we also run on-screen measurements in order to evaluate the performance of individual mobile devices, creating the most applicable benchmark results.
Smartphones and tablets employ v-sync, capping performance at 60 FPS. With this in mind, the iPhone 5 is no faster than the iPhone 4S in GLBenchmark 2.1, which is meant to replicate the demand of today's mobile games and was already pegged by Apple's previous-gen phone. Although is features a third GPU core, the iPhone 5 won't help pave the way for a more compelling gaming experience until developers start utilizing it more completely.
Up against competing phones, however, the iPhone 5 is at a clear advantage. The One X and Galaxy S III can't quite achieve 50 FPS in the Standard Egypt test.
The Pro test is less stressful, and as a result, the One X, Galaxy S II, and S III all perform slightly better. Naturally, though, the iPhones, which were already pegging the Egypt test at 60 FPS, have no problem doing the same thing again.
Setting a consistent 720p resolution and lifting the v-sync cap gives us a better idea of what each graphics engine can do, unobstructed. The iPhone 5's PowerVR SGX543MP3 (with its three GPU cores) really struts its stuff, more than doubling the iPhone 4S' SGX543MP2 (with two GPU cores).
This workload, all on its own, indicates that the A6 has a healthy future ahead of it if ISVs start leveraging the hardware's capabilities more fully. Now that the fourth-gen iPad has Apple's A6X in it, with four GPU cores, it's probable that game development progress will accelerate, benefiting the iPhone as well.
Fill rate remains a bottleneck of mobile graphics performance, and the iPhone 5 dominates this metric, followed by Apple's iPhone 4S.
The GLBenchmark 2.1 benchmark remains useful to us, even though version 2.5 is newer, serving up higher-quality graphics and an off-screen test that runs at 1920x1080. According to Kishonti, it's intended to represent what the next generation of mobile games may demand of graphics hardware.
It's not altogether disappointing that the iPhone 4S, One X, Galaxy S II, and Galaxy S III sustain fairly low performance at their native resolutions, then. Rendering off-screen, the iPhone 5's PowerVR SGX543MP3 takes a tremendous lead, delivering ~2.5x of the iPhone 4S' performance and 2x the One X and Galaxy S III's performance.
Maybe i am reading it wrong ?
Second to last paragraph:
"For example, Samsung's Galaxy S III has been on the market for a while, and its LCD is a market leader"
LCD = AMOLED
My apologies. On page 6 we mentioned that S3 uses AMOLED. I'll make a correction. Thanks for the shout out.
http://news.cnet.com/8301-13579_3-57524291-37/color-me-prettier-galaxy-s3-display-outdoes-iphone-5s/ Check that one out. :)
I'd still like to add an accuracy test at some point. I need to think about how to approach that problem. The way people normally talk about color accuracy (besides deltae and gamma) are sometimes too abstract with color terms that professionals use. I'd like to find a way to present the information in a more intuitive manner.