Top secret: whether hyper-threading and dual-channel are used or not!

Recently, I have found an interesting phenomenon in various major markets. All the users who come to the Computer City to crash, whether beginners, game players or masters, chose intel p4c series processor (supporting hyper-threading) and the i865 chipset (supporting dual-channel memory technology) motherboard, it seems that the selection of P4 with hyper-Threading Technology and supporting dual-channel chipset is a outdated, price drop performance. But how many of them really understand these two technologies, and how many people finally bring their performance into play?

Hyperthreading and dual-channel memory control are two different technologies. Of course, both of these technologies can find their own position and value at different application levels. In order to give you a thorough understanding of the two technologies, I believe that only the corresponding analysis and vertical comparison tests of the two technologies can find the answers we need. Of course, only in this way can we be "targeted" when we are "busy" to avoid the waste of "Silver" in our pockets.

1. What is hyper-Threading Technology?

1. simple definition of "hyper-threading" Technology

The so-called hyper-Threading Technology is to use special hardware commands to simulate two logical kernels inside the multi-thread processor into two physical chips, so that a single processor can "enjoy" the thread-Level Parallel Computing processor technology. Multithreading technology can effectively enhance the processing capability of the processor in multi-task and multi-thread processing on the operating systems and software supporting multithreading.
Hyper-threading technology allows multiple threads of the operating system or application software to run on a single hyper-threading processor. The two internal logic processors share a group of processor execution units, add, multiply, and load operations in parallel. This improves the processing capability of the processor by 30%, because the applicationProgramYou can make full use of the various computing units of the chip.

For a single-thread chip, although it can process thousands of commands per second, it can only process one command (single thread) at a certain time, the result is that other processing units inside the processor are idle. Hyperthreading technology enables the processor to concurrently process more commands and data (multiple threads) at a specific time point ). It can be said that hyper-threading is a technology that can fully "mobilize" Resources temporarily idle inside the CPU.

2. How does hyper-threading work?

When processing multiple threads, each logic processor in the multi-thread processor can independently respond to the interrupt. When the first logic processor tracks a software thread, the second logic processor also began to track and process another software thread.

In addition, to avoid CPU resource conflicts, the logic processor responsible for processing the second thread uses a processing unit temporarily idle when the first thread is running. For example, when a logic processor executes a floating-point operation (using the floating-point operation unit of the processor), another logic processor can perform an addition operation (using the Integer Operation Unit of the processor ). In this way, the CPU processing unit utilization and the corresponding data and command throughput are greatly improved.

3. Five prerequisites for hyper-threading

(1) CPU support required:

Currently, CPUs supporting hyper-Threading Technology include pentium4 3.06 GHz, 2.40c, 2.60c, 2.80c, 3.0 GHz, 3.2ghz, Prescott (pentium5) processors, and some Xeon processors.

(2) Motherboard chipset support:

The main model of the Motherboard chipset that officially supports hyper-Threading Technology includes Intel's 875 P, e7205, 850e, 865pe/G/P, 845pe/GE/GV, 845g (B-Stepping ), 845e. 875 P, e7205, 865pe/G/P, 845pe/GE/gv chipset can be used normally with hyper-Threading Technology, the earlier 845e and 850e chipset only need to upgrade the BIOS to solve the support problem. Sis includes sis645dx (B edition), sis648 (B edition), sis655, sis658, and sis648fx. Via has p4x400a, p4x600, and p4x800.

(3) motherboard BIOS support is required:

The motherboard vendor must support hyper-threading In the BIOS.

(4) Operating System Support is required:

Currently, Microsoft only supports this function in Windows XP, And the hyper-threading Support Plan on Windows has been canceled.

(5) application software support is required:

Generally, software that supports multi-processor can support hyper-Threading Technology, but in fact there are not many such software, and they prefer graphics, video processing, and other professional software, game software is rarely supported. Application software includes Office 2000 and Office XP. In addition, later versions of Linux kernel 2.4.x also support hyper-Threading Technology.

2. What is "dual-channel" memory technology?

The dual-channel memory technology is to create two memory controllers in the North Bridge (also known as gmh) chipset. These two memory controllers can work independently of each other. In these two memory channels, the CPU can separately address and read data, so that the memory bandwidth can be doubled, and the data access speed can also be doubled (theoretically ).

Currently, the popular dual-channel DDR memory architecture is built on two 64bitddr memory controllers. The bandwidth can reach bits, but the working method is different from the single-channel memory control technology of bits. Because the two memory controllers in the dual-channel system are independent and complementary Intelligent Memory controllers, both memory controllers can operate at the same time without waiting for each other. For example, when controller B is preparing for the next access to the memory, Controller A reads/writes the primary memory, and vice versa. This complementary "nature" of the two memory controllers can reduce the effective wait time by 50%, thus doubling the memory bandwidth.

The two memory controllers of dual-channel DDR functions exactly the same, and the timing parameters of the two controllers can be set separately. This flexibility allows you to use two dimm memory disks with different structures, capacities, and speeds. At this time, the dual-channel DDR can be adjusted to the lowest density to achieve bit bandwidth, allows the dimm memory stick with different density/wait time features to operate reliably together.

In short, dual-channel technology is a technology related to the Motherboard chipset and has nothing to do with the memory itself. As long as the manufacturer integrates two memory controllers inside the chip, it can constitute a dual-channel DDR system. The motherboard manufacturers only need to divide dimm into Channel 1 and Channel 2 according to the memory channel. Users also need to insert memory in pairs, just like RDRAM. If you only insert a single memory, only one of the two memory controllers will work, so there will be no dual-channel effect.

The dual-channel memory control technology can effectively improve the memory bandwidth, especially the software that requires frequent data exchange with the memory and the chipset that integrates the graphics core (integrated graphics card. The high bandwidth brought by the dual-channel memory control technology can help the Integrated Video Card achieve higher data bandwidth when dividing the primary memory as the video memory, the data bandwidth of the video memory restricts the performance of a video card.

For the motherboard that integrates the graphics core, its memory not only needs to frequently change data with the CPU, but also shares the integrated graphics core as the Display memory. At this time, the memory will change data frequently, which is a severe test for limited memory bandwidth.

The dual-channel memory control technology is a Motherboard chipset technology. Only the chipset that supports the dual-channel memory control technology can build a dual-channel memory platform, intel's idks include i850, i875p, i7205, i865pe, i865g, sis655, sis655fx, via pt600 (p4x600), via pt800 (p4x800), and via pt880, the amd camp only supports nforce2 and nforc4chipset.
Iii. Advantages and Disadvantages of hyper-Threading Technology

1. Advantages of hyper-Threading Technology

(1) hyper-threading has excellent performance in applications in many server fields such as Web Services and SQL databases.

(2) Mainstream desktop chipsets support hyper-threading, and you do not need to spend any extra time.

(3) Windows XP has been optimized to improve performance when running multiple programs that do not support multithreading. Even if it brings losses, it will appear mild.

(4) the performance of some software applications that support multithreading is improved by about 30%, such as 3 DSMAX, Maya, office, and Photoshop. Intel even achieved an improvement of 90% in a test.

2. disadvantages of hyper-Threading Technology:

(1) The popular Windows 2000 does not support hyper-Threading Technology. You must install Windows XP, which may be unsatisfactory.

(2) Enable hyper-threading for post-processing single-threaded applications, and sometimes degrade processor performance.

(3) There is a lack of common application software optimized for hyper-threading, so the performance is not fully reflected.

In general, through the comparison of the above advantages and disadvantages, we have learned that hyper-Threading Technology can indeed improve system performance when processing multiple tasks. When a single task is processed, the advantages of multithreading cannot be shown. Once hyper-threading is enabled, the internal cache of the processor is divided into several regions to share internal resources with each other, as a result, the performance of a single subsystem is reduced. The author believes that there is no need to enable hyper-threading when performing single-task operations. Only multi-task operations can enable hyper-threading in a timely manner and enjoy the benefits of hyper-Threading Technology.

Iv. Advantages and Disadvantages of "dual-channel" Memory Control Technology

1. Advantages of dual channels

(1) It can bring about twice the memory bandwidth, which can greatly benefit those software that frequently exchanges the necessary memory data, for example, spec viewperf, 3 DMAX, IBM data explorer, lightscape, etc.
(2) When the onboard video card shares the memory, the high memory bandwidth brought by the dual-channel technology can help the video card get smoother speed in the game. Take 3dmark2001se as an example, the score gap can be increased to 15-40%.

2. dual-channel disadvantages

(1) it must be structured on a motherboard that supports dual channels and have two identical memory disks of the same capacity and type. Intel dual-channel has high requirements on memory type and capacity, and the two memory modules must be completely consistent. The two-channel motherboard of SIS and via allows different capacities and types of memory to coexist, as long as there are two memory disks.
(2) dual-channel memory control technology is widely used in common games and applications, with a very small gap with single-channel.
(3) You need to purchase a motherboard that supports dual-channel memory control technology and two memory modules, which requires more costs.
(4) The dual-channel connection method is very important for beginners. Once the connection method is incorrect, the dual-channel connection will not work.
(5) The dual-channel memory architecture is difficult to overclock, which is not suitable for friends who like DIY overclock.

V. Competition between "hyper-threading" and "hyper-threading"

Next, we will take a P4 3.0g (supporting HT) processor as the testing object, and compare the performance gap between the two by enabling and disabling its hyper-Threading Technology.
Test Platform, visible (table 1 ).

Under this testing platform, we will use ZD business Winstone 2002 1.0 (commercial office performance testing), sisoft sandra2003 (hardware information and Performance Benchmark Testing), 3dmark2001 Se (famous 3D graphics performance testing software), lame (audio compression), flaskmpeg + DivX 5.0.2 codec (video compression), 3D Studio MAX 5 (3D rendering) photoshop 7.0 (2D image), 3D mark2001 Se + super Pi (multi-task test) to verify the power of HT.

Test 1. ZD business Winstone 2002 1.0

This software is used to test the commercial office performance of the system, including Word, Excel, access, PowerPoint, FrontPage, WinZip, Norton AntiVirus, Lotus Notes, and Netscape; ZD content creation Winstone 2002 1.0.1 is used to test the system's Web, graphics, multimedia, and other content creation performance, includes Photoshop, premiere, ctor, Dreamweaver, ultradev, Navigator, Windows Media Encoder, and Sound Forge.

In ZD business Winstone 2002 1.0, the performance gap between enabling and disabling HT is very small. After HT is enabled, there is almost no effect on the commercial office performance of the entire machine (only 0.2% ). Of course, this has a lot to do with whether office software supports hyper-threading. Therefore, if you are just a commercial or office user, there is no need to choose hyper-threading.

Test 2 Use sisoft sandra2003 to separately test the processor subsystem

General CPU computing performance

In the testing of general CPU computing performance, when we use HT, the performance of the processor increased significantly. The integer operation test score is increased by 7969 from 9336 to 17%, and the Integer Operation Performance of the processor is increased.

In the unique isse2 command operation test of P4, after HT is enabled, the score has also increased from 4046 to 5624, and the performance has increased by about 39%.

CPU multimedia computing performance

In the multimedia computing performance test, the integer computing score after HT is enabled increased from 12216 to 14059, and the performance is improved by about 20%. In the isse2 instruction set score, when we enable HT, the test score is immediately increased by about 40%. It can be seen that as long as there are games and related multimedia processor software to optimize Intel's hyper-Threading Technology, the performance improvement it brings is conceivable. It can be said that the first problem encountered by super Cheng technology is the support of software vendors. The reason why the performance test score of sisoft sandra2003 processors can be greatly improved is that sisoft sandra2003 supports hyper-threading. It can be seen that hyper-threading processors are only supported and optimized by application software, to realize its potential.

Test 3 3 dmark 2001se performance test

3dmark2001 se demonstrates the d3d performance of the system. After 1028*768 and 32 Color tests, enabling and disabling HT has not improved the system performance. Instead, after enabling HT, the performance of the processor has a small loss. It can be seen that once the software does not support hyper-Threading Technology, sometimes it will lead to unexpected performance degradation.

Test 4-tone and Video Compression

Audio Compression and video compression have extremely strict requirements on the performance of the processor, especially for video collection and compression. Without a video capture card, the performance of the processor has a significant impact on the compression duration of videos and audio.

First, the test of audio compression. When lame (audio compression software) compresses A 97.2mb WAV file into a 8.82mb MP3 file, enabling HT is exactly the same as disabling ht, this should not be the case, but the two are not competitive because the lame is not optimized for hyper-Threading Technology. Compression of audio and video files will test the processor performance more than compression of pure audio files. When we use flaskmpeg + DivX 5.0.2 CODEC to compress a dvd video file into MPEG4 files, divX 5.0.2 codec that supports hyper-Threading Technology gives full play to hyper-Threading Technology. The compression time was shortened from 349 seconds to 297 seconds.

Test 5 2D image processing and 3D rendering

Photoshop 7.0 and 3D Studio MAX 5 represent 2D image processing and 3D animation processing respectively. Through the performance of these two software, you can determine the role of hyper-Threading Technology in graphics and animation processing.

Because Photoshop 7.0 supports multi-processor and hyper-thread technology, it can easily take advantage of this advantage, with the help of multi-thread technology, performance is improved by about 10%.
This is a comparison of the time consumed after enabling and disabling ht when performing the despeckle operation in Photoshop. After HT is enabled, it takes only 6.8 seconds. After HT is disabled, it takes almost one second. It can be inferred that if an operation previously took 1 hour, it would save 7-8 minutes after enabling ht. The 3D Studio MAX series is the most popular 3D modeling software in the 3D graphics field. In the final rendering process of the scenario, the processor is used at full load, and the computing capability of the CPU is fully investigated, 3D Studio MAX 5 has greatly optimized Pentium 4 to greatly improve rendering performance. The 3D Studio MAX Series also supports multi-threaded operations to minimize the rendering time by taking advantage of multi-processor. The multi-processor system can always achieve great performance advantages. However, because 3D Studio Max is excessively dependent on floating point units during rendering, the multi-thread technology does not help much. In this case, other units of the processor cannot help. At this time, all the floating point units of the CPU are fully loaded, while other units cannot be "inserted ".

Test 6 multi-task parallel processing test

What is the help of HT for processing multiple transactions at the same time? It is decided to run super Pi, a software that puts pressure on the computing performance of the processor. In addition, the software is running, and another software 3dmark2001se that has strict requirements on CPU computing performance is synchronously running. when dealing with multiple tasks, super PI and 3D mark2001 Se are neither easy tasks for the system alone, not to mention running at the same time, which is a very rigorous test. When hyperthreading technology is disabled, Pentium 4 at 3.06ghz runs 3D mark2001 se, which takes a long time to call in the test scenario, and the frame drop occurs obviously during the test. After hyperthreading technology is enabled, the transfer time in the 3D mark2001 se test scenario has been significantly shortened. It seems that the basic tasks are similar to running a single task, and frames are not dropped during the test. From the actual test results, we can also see that both super PI and 3D mark2001 se have greatly improved the parallel computing performance after hyperthreading technology is enabled.
Summary:

Through the above tests, we found that as long as the application software supports hyper-Threading Technology, users can benefit from 1%-40%. Especially for servers and graphics workstation users, of course, this benefits from the server, graphics, 3D animation, and video production software, and has made special optimizations and support for multithreading technology. The hyper-Threading Technology of P4 is "kinship" with multithreading technology, so software supporting multithreading technology will naturally benefit from the hyper-Threading Technology of P4. For General Office, commercial, and game users, as the software in the application field does not support hyper-threading technology, it is not necessary to purchase P4 that supports hyper-Threading Technology During the purchase, after all, the purchase of CPU supporting hyper-Threading Technology will pay more costs and costs.
Here we take the mainstream P4 2.4c as an example. In terms of market positioning and price, P4 2.4c is only about 100 yuan more expensive than P4 at the same frequency that does not support hyper-threading. In exchange for the performance improvement of about 6% at a price difference, I think it is worth something. In addition, in the market, we can easily obtain a motherboard that supports hyper-Threading Technology at a price of around 500 yuan. Of course, if you really cannot get a little benefit from hyper-Threading Technology, you naturally don't need to spend more money.

6. Comparison between the P4 platform and the dual channel and the Single Channel

To keep pace with P4, Intel has released several mainstream MHz dual-channel chipsets, including i865pe, i875p, and i865g. Sis and via are not weak, and sis655fx and via pt800 are eliminated. However, while introducing the main dual-channel chipset, Intel also launched a single channel i848p, while SIS and via also released a single channel 648fx, via pt800 chipset. How much is the difference between the mainstream dual-channel and low-end single-channel in martial arts? We can only use a comparison method for demonstration.
Here, we will use i875p to pick out several other single-channel players that do not support the dual-channel technology (i848p, pt800, and sis648fx. For the test platform, see table 2 ). The most powerful geforce FX 5900 is selected here.

Comparison projects: sisoftware Sandra, Max, ZD business Winstone 2002, 3dmark2001, spec viewperf, etc.

Test 1 sisoftware Sandra max3 (memory bandwidth benchmark test)

In the memory Benchmark Test of sisoftware Sandra max3, i875pe, a dual-channel, takes the lead, leaving other competitors behind with a 40% advantage. It can be said that the dual-channel is indeed a very good solution for improving the memory bandwidth. However, whether the 6.4 Gbit/s memory bandwidth brought by the dual-channel ddr400 can be fully utilized by the current operating system and application software to achieve its real performance, other software is required for testing and judgment.

Test 2 ZD business Winstone 2002 (commercial performance)

ZD business Winstone 2002 is a comprehensive system testing software that simulates the efficiency of actual commercial software to reflect the overall efficiency of the system, from the test, the performance of the single-channel memory chipset is relatively poor. This shows that the dual-channel performance will be improved when running commercial software.

Test 3 ZD content creation 2002 (Multimedia performance)

ZD content creation 2002 is used to test the performance of multimedia applications in the system. It simulates the running of flash, Photoshop, dreamwave, and other software to reflect the system performance. This computation involves multimedia computation, and higher memory bandwidth provides better performance. The test results also prove this point.

Test 4 tmpgenc (video compression performance)

Let's take a look at the performance of video compression. During the test, we use the tmpgenc software for testing. The compression format is MPEG2. The video file with the tested clip size of 136 MB is compressed by default. The size of the generated file is MB. The test result is time-consuming. We can see that the dual-channel still has some advantages, although not obvious.

Test 5 WinRAR (File compression performance)

In the WinRAR compression test, we selected a folder with a capacity of 716 MB (3979 files in total) for compression testing. The compression format is standard RAR format, and the compression ratio is normal, the compressed file size is 345 MB. Because the compression test is also a test on the floating point and bandwidth performance of the processor, we can see that the dual channel has a high bandwidth advantage in File compression.

Test 7 spec viewperf (3D rendering performance)

Spec viewperf is a commonly used OpenGL professional testing program, mainly for real 3D professional application environments, such as 3 DSMAX, IBM data explorer and lightscape. In this test, the dual-channel i875p makes full use of the bandwidth advantage brought by the dual-channel, because software such as 3dsmax needs to frequently exchange data with the memory, therefore, higher bandwidth can effectively shorten the rendering time.

Test 8 comanche4 (kamanqi 4) Game Performance

This is a flight simulation game developed by novalogic. It uses Microsoft's dx8 as a standard graphics engine. In addition to testing the performance of graphics cards, it can also reflect the overall performance of the system. During the test, because the game has very high requirements for memory bandwidth, It is very busy to exchange data with the memory (large scenario ). Therefore, we can see that the dual-channel i875p chipset uses its high bandwidth to finally improve performance by about 10%.

Conclusion: Same! For applications that require frequent data exchanges with memory, the high bandwidth brought by dual-channel memory can bring about 1% ~ Performance Improvement of about 15%. I think this is mainly related to the application software because the performance has not improved by 100%, after all, the 800 Gbit/S band width of the 6.4 MHz and dual-channel dd400 is not fully enjoyed by any software.

For video production (compression), 3D animation rendering, File compression, high-end games and commercial purposes, the high bandwidth brought by the dual-channel memory architecture can bring some help. Therefore, for these users at the application layer, it is indeed necessary to use a dual-channel memory architecture to increase the memory bandwidth. In addition, if you are using an 865g chipset that integrates the graphics core, we believe that you will be able to enjoy the smoothness of the dual-channel technology in the game.
VII. Summary:

I believe that for professional video processing, 3D animation processing and 2D image processing, we can consider both hyper-Threading Technology and dual-channel technology, because both technologies can bring tangible performance improvements. If this machine is only used as a server, I think the hyper-Threading Technology that supports multithreading and multitasking will be more useful than the dual-channel memory control technology. For mainstream 3D Game users, dual-channel technology will be more useful than hyper-threading technology, because it is not necessary to add hyper-threading support to the game, however, the current demand for system memory bandwidth for games is extremely urgent, from the upgrade of the AGP bus from agp4x to agp8x, we can see how much mainstream games will pay attention to data transmission capabilities in the future. The inherent advantages of dual-channel memory technology in data transmission will inevitably bring strong support to mainstream games in the future.

In addition, from the processor collocation and market positioning, it is also very exquisite. The P4 2.4c (800 MHz) processor is a 6.4 MHz Front-End bus and must satisfy the G/S band width of P4 2.4c, non-dual-channel ddr400 memory architecture (875 P, 865pe ). Although the same-frequency P4 2.4b + 845pe architecture is the same as P4 2.4c in terms of operating frequency, however, it has lost the powerful support of 300 MHz Front-End bus, hyper-Threading Technology and dual-channel technology. Although, it seems that the cost is about 100 yuan (CPU 100 yuan, motherboard yuan, memory yuan ), however, for those server platforms that support hyper-Threading Technology and professional graphics, images, and video users, the performance impact is far greater than the cost gap of about 6%.

For ordinary users, the author believes that the low-end P4 (about 1000 yuan), C4 (about 500 yuan), CIII (about 300 yuan), and AMD athlon XP (about 600 yuan) and the new Poison Dragon (about 300 yuan) processor will be more suitable. These CPUs are enough for users who do not have "multi-threaded applications" or "Big Data Exchange" to enjoy the benefits and satisfaction of "inexpensive.