Monday, September 11, 2017
Geekbench 4 Cross Platform Testing Analysis
Geekbench 4 Cross Platform Testing Analysis
Overview
Geekbench 4 was released at the end of August and has proved to be a welcome improvement over the previous revision 3, having been praised for accurately reflecting the industry standard SPEC results through a benchmark suite that is readily available on different platforms and is easy and quick to run.
As Geekbench 4 is comparable across platforms and is available for Windows, Android, iOS and others, I thought it would be interesting to test my server, desktop, laptop, tablet and phone to see how they perform relative to each other.
I am also interested in how different processors compare across platforms, so compare the scores of the fastest processors available in different platforms and segments, including the fastest Core i7, Core M, Apple A10 and Exynos 8890 / Snapdragon 820 on Android.
Single vs Multi-core scores
Before getting to the results of my test, it is worthwhile discussing the two results that are returned from the benchmark - the Single and Multi-core values
The Single-Core result provides a ranking of which processor has the fastest CPU core in a single threaded situation. Having a fast core indicates that an application will perform well in single-threaded situations. Writing well-threaded applications is more long winded and complex than writing a traditional single-threaded application, so developers are unlikely to put in the extra effort required unless a performance issue crops up in their testing, or their application is particularly suited to multi-threading.
The Multi-Core result shows what can be achieved if an application is both suited to, and optimised for, a multi-threaded environment. The Operating system itself and many core applications are generally well threaded and will take advantage of multi-core processors. Applications and Services running in the background also make good use of multiple cores.
Which result, therefore, is the most reflective of real world performance? Generally speaking, having a high single core score will ensure that all applications perform well, no matter how they are coded. Conversely, having more, slower cores may perform better in some situations but will also perform worse in others.
I would suggest, therefore, that having a high single core score will ensure performance consistency, and having a high multi-core score will show the maximum performance you can expect in ideal circumstances.
The operating system design also plays a part in this - the higher degree of true multi-tasking the OS provides, the more important the multi core score and the less important the single core. In terms of todays common platforms, the multi core score is most important in Windows, important in Android and less important in iOS.
In conclusion, neither the single or multi-core scores provide the whole picture - the real answer lies somewhere in the middle, with the bias being dependant on the host OS and the suite of applications you frequently use.
My Results
Below is a table of my main computing devices in the house - an older desktop PC, a modern Server and Laptop, an iPad Pro and a recent phone.
The results below are ordered by their Single Core result from lowest performance to highest.
| Device | Platform | CPU | Single-Core | Multi-Core |
|---|---|---|---|---|
| HP xw6400 (Desktop) | Windows 10 | 2 x Xeon E5345 | 1362 | 5802 |
| Xiaomi Mi Max (Phone) | Android 6.0.1 | Snapdragon 652 | 1467 | 3304 |
| HP Microserver Gen 8 (Server) | Windows Server 2012 R2 | Core i3-3220T | 2399 | 5118 |
| iPad Pro 12.9" (Tablet) | iOS 9.3.5 | Apple A9X | 3019 | 4915 |
| Dell Inspiron 13 7000 (Laptop) | Windows 10 | Core i5 6200U | 3053 | 5836 |
The table shows just how quickly mobile SoCs have caught up with older and lower power Intel Core processors.
To the layman, my desktop PC would appear to be the fastest compute device in the house - it certainly looks the part, being a hefty HP Workstation. It is a 2 processor, 8 core Xeon Workstation. However, its age is showing and the single core performance (Core 2 era Clovertown) is below even my mobile phone. Of course, having 8 cores means it is still a workable machine even today.
The iPad Pros Apple A9X, featuring 2 x 2.25Ghz Twister cores, is on a par with my 9 month old Ultrabook in terms of single core performance, and not too far off in terms of its multi-core score.
Here I have chosen a set of the top-performing processors in terms of Desktop, Laptop, Fanless laptop, iOS and Android to see how they compare.
| Processor | Cores | Platform | Single-Core | Multi-Core |
|---|---|---|---|---|
| Intel Core i7-6700K | 4 x 4Ghz Skylake | Windows (Desktop) | 5329 | 17023 |
| Intel Core i7-6950X | 10 x 3Ghz Skylake | Windows (Desktop) | 4462 | 29797 |
| Intel Core i7-6650U | 2 x 2.2Ghz Skylake | Windows (Laptop) | 3792 | 7357 |
| Apple A10 | 2 x ? + 2 x ? | iPhone 7 | 3379 | 5495 |
| Apple A9X | 2 x 2.25Ghz Twister | iPad Pro | 3019 | 4915 |
| Intel Core m7-6Y75 | 2 x 1.3Ghz Skylake Y | Windows (Fanless) | 2854 | 6030 |
| Samsung Exynos 8890 | 4 x M1 2.6Ghz + 4 x A53 1.6Ghz | Galaxy S7 / Note 7 | 1809 | 5232 |
| Qualcomm Snapdragon 820 | 2 x 1.8Ghz Kryo + 2 x 1.36Ghz Kryo | Multiple | 1660 | 4015 |
It is also interesting to see that it is faster than the fastest Android SoC, as found in Samsungs S7 & Note 7 phones, in the multi-core score as well as being faster, as expected, in the single core benchmark.
This Anandtech article provides a great overview of how well used multi-core SoCs are on Android
This XDA article provides an overview of the changes between Geekbench 3 and 4 and provides some analysis on how scores have changed between common Android SoC cores in the process.
This XDA article is an interview with the Geekbench CEO about the new benchmark & the rationale behind the changes from version 3.
download file now
Saturday, September 9, 2017
Groningen fork analysis
Groningen fork analysis
Commits from August 11th 2013 to February 2nd 2014
download file now
Saturday, September 2, 2017
Tools for Android malware analysis
Tools for Android malware analysis
This post gives some tools for which I managed to find some info and think that can be useful for dynamic and static malware analysis.
When doing malware analysis, it is necessary to prepare safe and reliable environment for running the malware in order to observe its behavior or perform procedures of reverse engineering. Android malware, as much as other platforms, demands such environments which is bit hard to find. One reason is because malware analysis task demands high processor performance and memory capacity. Also, emulators provided by official Android SDK are very slow and we don�t want to run malware on real Android device. This makes dynamical malware analysis very annoying and tedious. Some tools like CopperDroid address this problem very nicely. Specifically CopperDroid provides web interface where you can upload apk file which is then analyzed and results of dynamic analysis are given. Downside of this approach is that it maybe won�t find all threats.
Following list enumerates some tools for Android malware reverse engineering, debugging, monitoring and emulating behavior of an Android device:
- Androguard � Tool written in Python which can be used to analyze android application. Some of the features are: disassembly/decompilation/modification of DEX/ODEX/APK format, diffing of Android applications, measure the efficiency of obfuscators,checking if an Android application is present in a database (malwares), risk indicator of malicious application
- APKinspector � Also written in Python but provides a GUI to aide analysis of APKs and their DEX code. APKinspector can help to generate reports for permissions used by the application, search and filter strings, classes and methods and more. Some of the current models used by APKinspector are based on Androguard
- Droidbox � Sandbox for dynamic analysis of Android applications. This tool gives hashes for the analyzed package, incoming/outgoing network data, file read and write operations, started services and loaded classes through DexClassLoader, information leaks via the network (files, SMS), circumvented permissions, cryptographic operations performed using Android API, listing broadcast receivers, sent SMS and phone calls.
- CopperDroid � out-of-the-box dynamic behavioral analysis of Android malware. To this end, CopperDroid presents a unified analysis to characterize low-level OS-specific (e.g., writing to a file) and high-level Android-specific behaviors (e.g., sending an SMS, placing a phone call). Based on the observation that such behaviors are however achieved through the invocation of system calls, CopperDroids VMI-based dynamic system call-centric analysis is able to faithfully describe the behavior of Android malware whether it is initiated from Java, JNI or native code execution. In addition, CopperDroid features a stimulation technique to improve code coverage, aimed at triggering additional behaviors of interest. Our initial experiments show that a proper malware stimulation strategy (e.g., sending SMS, placing calls) successfully discloses additional behaviors on a non-negligible portion of the analyzed malware.
- Android-apktool � Tool for reverse engineering Android APK format using Java runtime environment.
- Smali - assembler/disassembler for the DEX format used by Dalvik, Androids Java Virtual Machine implementation. Supports the full functionality of the DEX format, e.g. annotations, debug info, line info.
- Android-x86 - provides a ready-to-use virtual machine disk which can be simply mounted and used to run original Android on VirtualBox. The advantage of this method is that the experience is 99% like an actual Android device, but with higher processor performance, physical memory and storage.
download file now