Editor’s Note: since CES2012 since Intel unveiled Medfield platform for the mobile market, there were some X86 based Android phones on the market. They can achieve in the test even a single core dual-core, quad-core dual-core, that X86 mobile application and is really so good?
The writer writers, technical talent for love “gun God” @ioncannon.
Most are using Dalvik Android application Java code development. In theory, because the Dalvik code executes in a virtual machine system, no performance loss in the x86 ATOM platform. But due to the limited performance of virtual machines, for some applications of high performance pursuit, Google allows developers to use the native c-language code. Native code is compiled for a hardware platform, x86,ARM, or MIPS has its corresponding binary file. In General, binary code compiled for ARM platform x86 processor is not running. Popular Android products using ARM processors, although a long time ago, Google introduced the Android NDK x86 compiler option, but there are quite a few applications compiled for ARM platform, this creates a x86 Android platform compatibility issues.
In order to solve this problem, the Intel ATOM platform for Android mobile phone system introduced at the beginning, have just introduced a “binary conversion” (Binary Translation) this function, to solve the problem of x86 not run direct ARM. ARM compiled binary code will be translated into code that executes the x86 ATOM processor to run applications that contain native code compiled for ARM.
Through this function, the x86 ATOM is compatible with most of the applications on the market. Latest Lenovo K900 Atom Z2580 processor has their TOP 20000 application of compatible with.
After you resolve compatibility issues, plus x86 Atom single core and dual-core quad-core, dual-core second strong performance (running), x86 seems to have no spots, swept the market just around the corner. But it is a pity, x86 advocated by the “compatible” and “performance”, because of the current limits of eco-environment and cannot eat.
First, we simply look at x86 on the Android ecosystem:
Although the TOP 20000 application would be compatible, but individual statistics peas top 250 apps in the market, which 31% applications without using native code. Other 69% containing native code applications, only 8% includes the x86 library, 61% only the left ARM native libraries. In other words, only 8%, x86 can play to their best performance, and most, needs to convert the binary to be compatible.
Further, in high demand for performance gaming applications, native code is more widely used.
By TOP100 popular game statistics when the music network, we found that only 6% of the game contains no native library is better than 94% games containing native code. In this one, only 5% game x86 native library, remaining close to 90% games, Atom processors are needed by binary conversion runs code compiled for ARM.
Of course, the demand for high performance, many games include the armv7a runtime. Nevertheless, I have Lenovo K800 (Atom Z2460) cell phones can also be performed by binary transformation. But after testing, native libraries that contain Neon SIMD code cannot be executed.
Said this, we understand that quite a number of applications need to be x86 by binary conversions to compatible ARM code from running. So, x86 performance, so powerful, compatible implementation of ARM code, how about the performance? This is a problem that has long been shunned by Intel and manufacturer.
The test object: Lenovo K800 mobile phone, is the first listed in the domestic x86. Processor Atom Z2460 1.6GHz, single-core dual-thread, PowerVR SGX540 400MHz GPU. iPhone case Kenzo
First, we know Bunny, which is one of x86 favorite runs.
Although Z2460 only a single core, but its score was still more than most of the low-frequency dual-core ARM A9, compared to high-frequency dual-core A9/Krait no less. Dual-core four-thread Z2580 is 30,000 shock, pointed to the current top-end 4+4-core processor.
However, after using the ARMv7, Z2460 performance is not satisfactory, General suddenly shrunk two-thirds, documentary A8 CPU score is almost …
Following a series of tests revealed a similar situation: when using x86 native libraries, we can find ATOM single-threaded performance exceptionally tough, especially memory performance
When using compatibility mode, running ARM binary conversion library, performance is greatly reduced, 1.6GHz Atom or even just 1GHz less than Cortex-A7 level.
CoreMark is exactly the same situation, considerable losses
As a compatible price performance loss is one thing. On the other, an increase in power consumption. In the first half of this paper has shown, and present few popular game x86 native libraries, in order to facilitate testing, we selected the ghost Castle (Epic Citadel) as representative of the game.
Epic Citadel is a demo based on the Unreal engine, very beautiful picture, light and shadow effects are also good
The tests we used the Intel GPA System Analyzer to record the current CPU usage and mobile phone
Epic Citadel’s own Benchmark scenarios tested, the results are as follows
Using x86 native libraries (top) received an average score of 45.9FPS, compatible with ARM native libraries for 31.2FPS, about 30% gap.
Also found that throughout the process, executed by binary conversion ARM compared with the original running x86, CPU usage and current consumption on mobile phones, are certain to increase.
By contrast, compatible running ARM compared to original x86 code, performance dropped 30%, average CPU usage increased from 58.9% to 73.4%, and the average has increased from 621mA to 717mA, an increase of approximately 100mA, relatively speaking, CPU power consumption increases about 400mW, is quite substantial.
CPU occupancy rate is only a reflection, of course, since the core frequency control when the default is Ondemand, although occupancy rates there was little difference, but when you execute ARM code binary conversion, CPU running at high frequency for longer, thus leading to increased power consumption.
Concluded as follows, you can see native x86 compared to binary conversion code, loss of performance is very impressive in performance demands, losses can be as high as 50%. In some game applications, will also bring about increased CPU power of 400mW.
Overall, the x86 still have a long road to Android, despite efforts by Intel and manufacturer addresses compatibility and performance issues, but a relatively poor performance and compatibility of the ecological environment can not have both. But regrets of is, Intel in this year said “phone business without sales only play high-end” “despite market Shang sales maximum of is thousand Yuan phone, but currently also not Intel of direction, Intel aimed at made now world Shang performance highest of phone”, in must degree Shang, relative lower of share also is to caused developers of note and mobilization developers of enthusiasm, in ecological environment of improved Shang also is was concerns.
Learn more about the new cool device, please pay attention to @ love machine
Refused to talk mobile GPU reading (a)
Refused to talk mobile GPU interpretation (b)
High-quality audio-video vivo Xplay hands-on experience
Empire Galaxy S4 “4+4 nucleus” full version evaluation
“Super” in? Millet 2S performance testing
Millet Note Hat Edition
Millet Note Hat indeed stronger than other phones with the same processor performance, as to whether the legend of “Xiao long version V2.1 810″ played a role, we have no idea, but there is no doubt that it does have a lot of advantages in kernel optimization. Millet Note Hat voices to remain consistent and Standard Edition, but more on the transient performance of sound excellent. In addition, the hat to dive at low frequencies also will feel in place, also has some strengthening of sound field.
View details of the voting >>