Can you challenge X86 key application arm Server metrics analysis

Since the end of last year, the ARM processor has been a very hot word, because according to ARM's propaganda, the next generation Cortex A50 processor will not only be limited to the mobile terminal area layout, but also to enter the server market. Even Cortex A50 is far away, and some servers based on existing arm architectures seem to be on the horizon. What can arm do in the server market? And to what extent? About arm not only has too much content not public, even if the public part also has many doubts. In this paper, the arm in the server domain of some of the indicators can not understand the interpretation to see if arm is really so good.

calxeda:http://www.aliyun.com/zixun/aggregation/20522.html "> test data not to elaborate

Last July, Boston Server released a arm server viridis, should be regarded as the industry's more advanced arm server products, according to information, the server's power consumption compared with the X86 server has a huge advantage, while the performance is very good, as the opening, This is the first viridis and related indicators to read.

Viridis is a server based on the ARM processor developed by the Calxeda company. Calxeda Company currently launched ARM processor products for the ECX-1000, the processor frequency of 1.1GHz or 1.4GHz, based on ARMV7 architecture, chip technology for TSMC 40nm. Because it is a SOC chip, the processor itself integrates the DDR3 memory controller, 4 PCIe 2.0 channels, and the SATA3 controller.

It can be said that Calxeda's ECX-1000 is a more typical server SOC, not only viridis, including HP, many ARM server research and development projects, and Calxeda have cooperation. According to Calxeda's claim, the processor has an absolute advantage over X86 in some performance, and the following figure is the test data released by Calxeda.

According to the official introduction, ECX-1000 in AB (Apachebench) and Xeon e3-1240 compared to the performance of each watt has a staggering 15 times times the amazing advantage. But aside from the AB test CPU performance is suitable for the doubt, the simple test itself there are some problems did not explain clearly. Apache is a more comprehensive web tool, which means that different test environments will have a great impact on the results. For example, the use of static or dynamic pages in a test environment is highly likely to produce exactly the opposite result for a processor that has a very large gap between arm and Xeon.

Other details not described clearly include time taken for tests and time/Request two important reference data are not disclosed; there is a virtual standard for the power consumption of the Xeon platform (e3-1240 TDP is 80W, Run AB This biased throughput test does not have a full load at all; instead, the main frequency difference between the two is Calxeda in the Open data, but the great difference of the frequency is not particularly significant to the results of AB test. In short, the concealment and exaggeration of key indicators in the Calxeda test data completely weakened the credibility of the test.