Intel Core i3-550 benchmarks
Core i3-550 microprocessor was planned for the second quarter of this year, and is not officially released yet. The processor is based on Clarkdale core and combines two separate dies into a single package:
While the i3-550 microprocessor is not officially introduced yet, it is sold in many stores. Current prices for this model are in the range $150 - $170 USD, but probably will be reduced once the processor is officially released. Preliminary launch date of this CPU is June 1, 2010. I expect that Core i5-655K and Core i7-875K will be introduced at the same day.
For detailed pictures of the microprocessor and the box please visit Intel Core i3-550 specifications page.
For benchmarks I selected all models from Core i3 family. Although it's easy to predict performance of 530 and 540 models in comparison with the i3-550, the benchmarks can serve as an additional confirmation that the benchmarks were properly done, and should be helpful for those who cannot decide which Core i3 model to get. To see how the 550 model stacks against more expensive Intel processors, I included two slowest Core i5 microprocessors - i5-650 and i5-660. AMD counterparts in this review are represented by budget triple-core Athlon II 440 and quad-core Athlon II 635 CPUs, both are considerably cheaper than the i3-550, and 4 Phenom II models - X2 555, X3 720, X4 925 and 955. I considered adding Athlon II X2 255 and Pentium G6950 to the review, but they were hopelessly slow in comparison with the i3-550.
For reference, I put all features of tested processors in the table below. For more detailed characteristics please click on the processor name in the left column of the table:
"HT" in the "Other features" column stands for HyperThreading, and "TB" stands for Turbo Boost. Prices for Intel and AMD processors as of May 25. No doubt the prices will change after the Core i3-550 part is launched.
Tests of socket 1156 processors were performed on ASUS P7P55D motherboard, equipped with 2 GB of dual-channel Corsair TR3X3G1600C9 DDR3 memory. Tests of socket AM3 microprocessors were done on ASRock M3A780GXH/128M motherboard, equipped with 2 GB of dual-channel Corsair CM3X1024-1333C9 DDR3 memory. Socket 1156 and AM3 systems had 32-bit Windows XP installed, and used the same type of hard disk drive and video card. The graphs in this review do not show absolute numbers returned by benchmarks. Instead, all results are provided as relative value, in percent, that shows how much faster or slower current processor is in comparison with core i3-550 CPU.
Sandra Dhrystone test, part of SiSoftware Sandra 2004 Service Pack 2 (SP2b) suite, runs integer performance test on all logical processors. The values are measured three times, and the final value is an average of the three results.
The test depends on: CPU frequency, Integer performance, The number of cores / threads
We'll start integer benchmarks from Sandra Dhrystone synthetic benchmark. The Sandra benchmark uses all processor cores, therefore it's not a surprise that quad-core CPUS have the best scores here. The benchmark does not depend much on the size of L2 and L3 caches, this is why Athlon II 440 and Phenom II 720 perform better or almost as good as the i3-550 CPU. Turbo Boost feature, incorporated in Core i5 models, helps i5-650 to beat 550 model by 4.4%. This value is close to the maximum for multi-threaded applications. As we will see later, in single-threaded applications the Turbo Boost feature in core i5-6xx series of processors may increase performance by up to 8%.
MySQL - selecting data
MySQL benchmark shows sustained number of SQL requests per second served by MySQL server v5.0.27. The test starts multiple processes, one per each logical processor. Each process connects to a MySQL server, running on the same computer, and executes a series of SELECT statements. All processes retrieve data from the same database table, but use different parameters in SELECT statements. Queried table has all required indexes to avoid excessive disk access. The processes do not disconnect between requests.
The results depend on: CPU frequency, Integer performance, The number of cores / threads, Size of caches, Memory bandwidth
This benchmark takes into account the size of caches and memory bandwidth, and, as a result, Core i5 and i3 processors look better here than in Sandra integer test. Still, quad-core AMD processors, even cheaper 635 model, outdo more expensive Intel counterparts.
Apache web server - static pages
Apache benchmark shows an average number of static pages served per by Apache 1.3.41. The test starts 64 threads. Each thread opens a connection to the same Apache web server, running on the computer, and fetches the same static page as many times as possible during predefined time interval. The static page is a copy of Test Page, that comes with Apache web server. The test has much greater error margin than other tests in our suite.
The results depend on: CPU frequency, Integer performance, Number of cores / threads, Size of caches, Memory bandwidth.
The Apache test uses relies heavily on a memory bandwidth, which partially explains why Intel processors are in the lead here. Another reason of excellent performance of Intel CPUs here is much better efficiency of HyperThreading for this type of test. In this test the HyperThreading technology boosts performance by more than 40%, while in MySQL benchmark the HT increases performance by 16%.
WinRAR compressing / decompressing benchmark
WinRAR compressing / decompressing benchmark reports compression / decompression speed, in KB per second, of random data by WinRAR archiver. The test is a part of WinRAR software.
The results depend on: CPU frequency, Integer performance, The number of cores / threads, Size of caches, Memory bandwidth.
WinRAR benchmark relies more on the number of cores and the size of processor caches, than on CPU frequency. This explain the results - quad-core CPUs are firmly in the lead, followed by triple-core and Core i3 / i5 processors. HyperThreading feature on core microprocessors actually results in 50% performance increase, but even that increase is not sufficient to outrun Phenom II X3 720.