Looking on CPU/GPU standard Optimizations taking part in Galaxy S 4
By hook or by crook both Anand and I ended up with international versions of Samsung’s Galaxy S 4, equipped with the principal generation Exynos 5 Octa (5410) SoC. Anand bought an international exemplar GT-I9500 while I held prohibited pro the much cooler SK Telecom Korean exemplar SHV-E300S, plus Samsung’s own SS222 LTE modem able of working on orchestra 17 (AT&T LTE) and orchestra 2,5 WCDMA taking part in the US. Both of these came from Negri Electronics, a transportable device importer taking part in the US.
Pro individuals of you who aren’t familiar with the Exynos 5 Octa taking part in these campaign, the SoC integrates four ARM Cortex A15 cores (1.6GHz) and four ARM Cortex A7 cores (1.2GHz) taking part in a significant.Modest configuration. GPU duties are handled by a PowerVR SGX 544MP3, able of running on up to 533MHz.
We both had diplomacy to get something done a deeper dive into the power and performance characteristics of individual of the principal main smartphone platforms to wastage ARM’s Cortex A15. In the function of for all time, the insane march of transportable got taking part in the way and we both got pulled into other things.
Other recently, a job finished on Beyond3D from @AndreiF gave us motivation to dust rotten our international SGS4s. Through about expert old fashioned benchmarking, the poster alleged with the purpose of Samsung was no more than exposing its 533MHz GPU travel clock to clear benchmarks - all other apps/games were restricted to 480MHz. Pro the previous hardly any weeks we’ve been asked by many to look into this, what did you say? Follows are our findings.
Characterizing GPU Behavior
Samsung awesomely exposes the current GPU travel clock with no requiring delve access. Simply run the following commandment finished adb and it’ll return the current GPU frequency taking part in MHz:
Adb shell cat /sys/module/pvrsrvkm/parameters/sgx_gpu_clk
Let’s optimism this doesn’t understand plugged, as it’s really an enormously of use level of transparency with the purpose of I want other transportable platform vendors would offer. Running with the purpose of commandment taking part in a round we can understand real era updates on the GPU frequency while applications run not the same workloads.
Running a few games, even the nearly all demanding titles, returned a GPU frequency of 480MHz - only this minute like @AndreiF alleged. Samsung on no account publicly claimed pass with flying colors GPU frequencies pro the Exynos 5 Octa (our in a row came from inner sources), so veto mischief veto foul along these lines far.
Firing up GLBenchmark 2.5.1 however triggers a GPU travel clock not on hand elsewhere: 532MHz. The same is loyal pro AnTuTu and Quadrant.
Interestingly sufficient, GFXBench 2.7.0 (formerly GLBenchmark 2.7.0) is unaffected. We incorrigible with Kishonti, the makers of the standard, with the purpose of the low level tests are identical flanked by the two benchmarks.
We be supposed to get the message roughly an 11% mount taking part in performance taking part in GLBenchmark 2.5.1 finished GFXBench 2.7.0, and we finish up as a spot other than with the purpose of. The motivation pro the difference? GLBenchmark 2.5.1 appears to be there singled prohibited in the function of a standard with the purpose of is permissible to run the GPU on the privileged frequency/voltage setting.
The CPU is and Affected
The novel job on B3D paying attention on GPU performance, but I was strange to get the message if CPU performance responded similarly to these benchmarks.
Using logic overseer I set aside an eye on CPU frequency while running the same tests. Firing up GLBenchmark 2.5.1 causes a switch to the ARM Cortex A15 cluster, with a default frequency of 1.2GHz. The CPU clocks on no account dump beneath with the purpose of, even after only this minute sitting idle on the menu screen of the standard.
Run GFXBench 2.7 however and the SoC switches finished to the Cortex A7s running on 500MHz (250MHz virtual frequency). It would appear with the purpose of no more than GLB2.5.1 is permissible to run taking part in this privileged performance mode.
A quick check across AnTuTu, Linpack, standard Pi, and Quadrant reveals the same behavior. The CPU chief is fixed on a clear purpose after either of individuals benchmarks is launched.
Interestingly sufficient, the same behavior (on the CPU side) can be there found on Qualcomm versions of the Galaxy S 4 in the function of well. Taking part in these pick benchmarks, the CPU is setting to the upper limit CPU frequency on hand on app launch and stays in attendance pro the duration, all cores are plugged taking part in in the function of well, nonetheless of load, in the function of soon in the function of the submission starts.
Hint with the purpose of the CPU behavior is not the same from what did you say? We proverb on the GPU boundary however. These CPU frequencies are on hand pro all apps to wastage, they are simply affected to upper limit (and taking part in the folder of Snapdragon, all cores are plugged in) taking part in the folder of these benchmarks. The 532MHz pass with flying colors GPU frequency on the other worker is no more than on hand to these restricted benchmarks.
Why this Matters & What’s after that
Not any of this ultimately impacts us. We don’t wastage AnTuTu, BenchmarkPi before Quadrant, and motivated rotten of GLBenchmark 2.5.1 in the function of soon in the function of 2.7 was on hand (we dropped Linpack a while ago). The remainder of our suite isn’t impacted by the aggressive CPU chief and GPU frequency optimizations on the Exynos 5 Octa based SGS4s. What did you say? This does mean however is with the purpose of you be supposed to be there judicious re comparing Exynos 5 Octa based Galaxy S 4s using a few of the affected benchmarks to other campaign and drawing conclusions based on with the purpose of. This seems to be there purely an optimization to foodstuffs repeatable (and high) results taking part in CPU tests, and liberate the highest workable GPU performance benchmarks.
We’ve believed pro years right now with the purpose of the transportable revolution has/will mirror the PC industry, and along these lines it’s veto shock to get the message optimizations like this employed. Only this minute as we’ve seen things like this come to pass taking part in the previous however doesn’t mean they be supposed to come to pass right now.
It's motivating with the purpose of this is sort of the reverse of what did you say? We proverb GPU vendors get something done taking part in FurMark. Pro individuals of you who aren't familiar, FurMark is a stress trying tool with the purpose of tries to understand your platform to obtain in the function of much power in the function of workable. Taking part in order to stay away from creating a setting somewhere thermals were privileged than they'd be there while singing a standard game (and to stay away from detrimental graphics cards with no thermal protection), we proverb GPU vendors limit the travel clock frequency of their GPUs after they detected these power-virus way of apps. Taking part in a transportable device I'd expect even greater sensitivity to something like this. I suspect we'll eventually understand to with the purpose of purpose. I'd and add with the purpose of only this minute like we've seen this sort of fixation many period taking part in the PC pause, the same is likely loyal pro transportable. The exertion is taking part in recognition after something astonishing is on offer on.
What did you say? Samsung needs to get something done on offer brazen is either sweeping up these settings pro all users/applications (e.G. Offer a configurable setting with the purpose of fixes the CPU chief taking part in a exalted performance mode, and unlocks the 532MHz GPU frequency) before remove the optimization altogether. The gamble of burden nothing is with the purpose of we finish up taking part in an arms battle flanked by all of the SoC and device makers somewhere non-insignificant amounts of era and engineering effort is spent on gaming the benchmarks more accurately than civilizing user experience. Optimizing pro user experience is all that’s needed, expert benchmarks benefit indirectly - individuals with the purpose of don’t determination eventually grow to be irrelevant.
Related : http://yaplog.jp/battery14/
Tags : Samsung , Galaxy S4
