"At the press conference last September, Apple executives deliberately compared the a9x chip on the iPad pro with the PC platform. According to apple, the CPU performance of a9x exceeds 80% of the portable PCs released in the previous year, and the GPU performance exceeds 90% of the models. Needless to say, such a comparison has naturally caused great controversy.
Fans who support Apple's views tend to prove Apple's conclusions with the test data of geekbench and gfxbench, while opponents scoff at the fairness and representativeness of such cross platform tests. At the beginning of this year, the famous IT website AnandTech compared the performance of a9x and skylake platform super books with the industry standard test set spec CPU 2006. The conclusion is that the performance of a9x is close to the level of core m 6y30, which seems to put an end to this dispute.
However, the spec CPU test set used by AnandTech is more used in the server field and is not highly representative in the general household and commercial markets. Moreover, Anand only tested the CPU integer items and did not involve the comparison of floating-point items; As for GPU, AnandTech still uses traditional evaluation methods such as gfxbench and 3D mark. As for the performance of the iPad Pro equipped with a9x in real tasks, major media and evaluation institutions rarely involve it.
For users, what really determines the performance feeling of the equipment is not the score result of the test program, but the execution speed of the actual task. In order to more intuitively compare the task performance of iPad pro and traditional PC, the author decided to conduct a comparison based on real applications to explore what kind of performance level iPad Pro has in real scenes.
IPad Pro Performance Test
Test platform parameters
Due to the failure of the newly purchased skylake platform PC, the author decided to use an old PC to compete with the iPad pro. The specific configuration of both parties is as follows:
PC——
CPU: Core i7 920 @2.66GHZ, Turbo mode on
RAM: 6GB DDR3 1333
GPU: Geforce GTS 250
SSD: Sandisk V300 120G
OS: Windows 10 Build 10586
iPad Pro
128G Wifi/iOS 9.2.1
The biggest trouble of cross platform performance comparison is that the test software is difficult to unify. In particular, there are few cross platform porting applications on IOS and windows platforms. It is basically impossible to measure the performance of the same software on the two platforms. Therefore, the author can only take the second place and select the relatively mainstream applications for each task on the two platforms for testing. Obviously, such comparison results will inevitably be affected by the optimization efficiency of different software, so it can not accurately reflect the real ability of different platform hardware. However, for ordinary users, this test method has more practical significance and reference value.
The comparison items selected by the author are web page JS engine efficiency, file decompression, picture format conversion, video software decoding and video recoding, which basically cover the categories with the highest performance requirements in IOS platform applications. Unfortunately, there are no 3D rendering applications on IOS platform, and there are no large-scale games transplanted from PC platform. Therefore, these two important projects cannot be compared with PC. The author will try to make up for this shortcoming when conditions permit in the future.
Next, let's take a look at the performance of iPad pro in real application scenarios.
Web page JS engine performance
Comparing web page performance is a very easy task. Mozilla's Kraken test is used here. JS test results are mainly affected by CPU single thread performance and browser engine efficiency, and the latter has a greater impact.
The Core i7 920 is a very old CPU and the dominant frequency is not high, so it performs poorly in this test. Now mainstream low-voltage mobile CPUs such as core i5 6300u have much better performance, with a score of about 1000ms, which is obviously better than iPad pro.
File decompression test
There is no easy-to-use file compression and decompression tool on IOS platform. The most popular application is izip pro, which only has relatively simple decompression and compression functions. The author uses a 1g image compression package (RAR format) with a compression rate of 50%. The results show that on iPad pro and PC platforms, the file decompression speed only depends on the performance of SSD. The speed of iPad Pro processing this compressed package is about 85MB / s, while the speed of contrast PC is only about 70mb / s due to poor performance SSD. On another PC equipped with high-performance SSD, the speed of decompressing the same file reaches 200MB / s.
The author has also used different compression rates and different formats of compressed packets to test, and the results are basically the same. It is difficult for us to encounter compressed documents with extremely high compression rate and performance test in our daily applications. Therefore, the main parameter determining the decompression performance is the file replication speed of the storage system. In this regard, the performance of the iPad Pro is roughly equivalent to that of a PC equipped with an older SSD.
The more valuable task is file compression test. Unfortunately, the file compression tools on IOS platform only have the option of simple packaging, that is, generating a compressed package with the same size as the source file, which does not involve CPU calculation. Therefore, the file compression performance can not be compared.
As a reference, in the spec CPU 2006 integer project test conducted by AnandTech, the file compression performance of a9x is close to that of core m 6y30 processor with the same frequency.
Picture format conversion
Converting a large number of pictures to different formats is a task that consumes CPU resources. There is an application called imageconverter on IOS platform, which can convert pictures in batch between JPG and TIFF formats. Here, use this application to convert 100 jpg images to TIFF format; In contrast, Photoshop CC 2015 (updated in November 2015) was used on the PC platform to perform the same task. The results are as follows:
Video decoding test
In this link, the author selects a 92 second video for test. The video resolution is 3840x2160, the encoding is x264 hi10p, the frame rate is 120fps, and the video size is 293mb.
The PC platform selects the latest version of potplayer in 2015, and the decoder is ffmpeg (the measured performance of LAV decoder is much worse); On the iPad pro, oplayer HD is used, and GPU acceleration is turned off in software decoding mode. During both tests, close the loop filter and count the actual playback time of 92 second video.
In this test, i7 920 CPU is fully loaded, and the occupancy rate of 8 threads is as high as 98%. From the final result, the iPad Pro is roughly equivalent to the four core Nehalem platform with reduced frequency to 1.8g. Considering that the a9x of the former has only two 2.26g cores (the frequency will be reduced to 2.16G when running under full load for a long time), this achievement is very amazing.
However, the efficiency of IOS platform video player playing the latest hevc encoded video is relatively poor. The performance of iPad Pro decoding 4K hevc video is only equivalent to the dual core same frequency Nehalem platform. However, when dealing with 1080p / 120fps hevc video, iPad Pro performs well and can play smoothly as compared with PC.
In this test, neither Photoshop nor imageconverter can make the processor work at full load, and the image conversion speed depends more on the single thread performance of the processor.
Video coding test
The author uses iMovie to convert a 52 second 4K H.264 video into 1080p format on iPad pro, and Adobe Premiere CC 2015 on PC to complete the same operation. The results are as follows.
Such a big gap is obviously not right. Finally, the author found that the mainstream video editing applications on IOS platform use GPU to assist in computing when dealing with such tasks, which greatly improves the execution speed. Therefore, the author started the GPU acceleration of premiere on another core i3-6100 PC and compared it. The result is still inferior to the iPad pro, and the conversion time has reached 57s. It should be noted that the GPU assisted coding efficiency of Apple's professional video software FCP x on OS X platform is also higher than that of Adobe Premiere. IMovie for IOS is likely to inherit this advantage.
IOS platform also has a video transcoding application iconv based on CPU operation. The results of transcoding test using this application are very unstable: when encoding video into MPEG4 and wmv8 formats, the speed of iPad Pro is roughly half that of i7 920, but the performance of the former is less than 1 / 5 that of the latter in x264 encoding tasks. However, iconv uses an older version of x264 coding component, and it is unknown whether it has been optimized for the arm platform.
Interesting details in the test
It is quite difficult to select applications suitable for comparison with PCs in the app store. In addition to the applications listed in this paper, the author has also tried more than a dozen other software, but it is difficult to get reliable results.
In the web page performance test, if you use the split screen function on the iPad pro and open two browsers at the same time to perform the kraken test, the score will drop to about 4400ms; In contrast, opening multiple browsers on the PC platform and performing tests at the same time will not reduce the performance. It can be seen that the performance of IOS's split screen function still has a lot of room for optimization. Dual core processors should process at least two tests at the same time with the same performance.
In the file decompression project, the document decompression speed of some file management applications on IOS can be called "snail crawling", and it takes more than one second to decompress a picture. A compressed package containing thousands of images needs to be processed for an hour, which is amazing;
In the video playback test, the decoding performance of several popular applications in the app store varies greatly, and the efficiency of decoding hevc video varies several times at most;
Most video editing applications on IOS platform are highly optimized for GPU, and the efficiency is almost the same. Surprisingly, their output video picture quality is very high, which is difficult to distinguish from the results of CPU coding. Unlike GPU coding applications on Windows platform, they are always inferior in output picture quality;
The image processing application of IOS platform has a very smooth performance on iPad pro. The response speed of batch filter operation on multiple pictures is very fast, which is no worse than Photoshop under windows. Only because it is difficult to quantify the results, this test does not include such items;
No matter what application scenario, the power consumption performance of a9x is excellent. In the video transcoding task, the iPad Pro runs at full load for 6 hours under 10% screen brightness and 10% power, which can be called the champion of mobile device endurance.
summary
From the comparison of the above tasks, the comprehensive performance of iPad pro in real use scenarios is comparable with that of mainstream PCs. However, due to the influence of software optimization and other factors, these tests can not accurately reflect the CPU and GPU performance level of a9x chip.
In some application scenarios, iPad Pro performs very well, but in other cases, its performance gap with mainstream PCs is very different; There are many things that PC can do, and there are no suitable applications on IOS platform. Only when the large software of Windows platform is completely transplanted to IOS platform, or apple mac system is migrated to its own chip, can we comprehensively and accurately measure the performance of Apple's self-developed chip.
However, we should also realize that the situation of mobile chip = low performance has become history. Processors based on ARM instruction set are now qualified to compete with low-power x86 chips. Perhaps not for a long time, we can see the comprehensive confrontation between arm platform and X86 platform. Let's witness how high the arm camp can reach in the future.
The official account of WeChat is searching for "love plate network". With the latest development boards, intelligent hardware, open source hardware and activities, you can master all your hands. Recommend attention!
[wechat scanning can be followed directly]
Recommended reading:
ADI five series design tools - my favorite award-winning activity
Do you know these facts about MEMS?
MSI can be assembled to release the one modular motherboard
Xiaomi will launch a mobile phone prototype with self-made chips at the rice noodle festival on April 8“
Our other product:
