How to run Windows 7 faster in the dual-core and multi-core era

Microsoft Windows7 has been sought after by the market and users. Although the kernel of Windows 7 is mainly based on the Vista code, several major advances have gotten rid of the troubles that Vista brings to users and greatly improved the user experience. In the kernel, a major change is how to improve the operation of multi-threaded applications. The benefits of this development are reduced energy consumption, increased scalability, and theoretically improved performance.

To verify the benefits of Windows 7 for desktop PCs, technicians tested it with the most commonly used desktop applications that reflected multi-threading capabilities (ie, graphics-oriented software). For example, Adobe Photoshop and other graphics applications self-detect the load on the desktop system during startup and feature configuration. In this case, graphics software typically uses all of the processor core and almost all of the memory, allowing the system to provide the fastest performance. Technicians typically use Viewperf benchmarks (a comprehensive graphical testing software, issued by SPEC Standard Performance Evaluation) and Cinebench testing (this is a simple graphics rendering benchmarking tool). These two software are free and can be used. Use after downloading from related websites.

The hardware platform tested was the Dell PrecisionT3500 workstation. The Dell T3500 is an entry-level workstation that is typically used by high-end graphics users. It has a quad-core XeonW3540 (Nehalem architecture) processor, 2.93GHz clock speed, a NvidiaFXquadro4800 graphics card, and 4GB of memory. This configuration is about the mainstream level of high-end desktop systems after 12 to 18 months.

In this test, technicians used three identical hard drives, and Dell workstations pre-installed Windows XP Professional, Vista Ultimate, and Windows 7 Ultimate, all based on 32-bit and are up to date. Version of the driver. Then, the technician runs a benchmark test on each operating system. When testing the next operating system, the hard disk used in the previous test will be replaced to ensure that different versions of the operating system run on the system hardware each time. The final test results are as follows:

Performance benchmark results for three versions of Windows

Test benchmarks (good values) Windows XP SP3 Windows Vista SP2 Windows 7 Ultimate
SPEC Viewperf 10 (SMT off) 95.84 142.95 139.35
SPEC Viewperf 10 (SMT on) 93.45 145.30 138.80
Cinebench 10 (SMT off) 3.43 3.40 3.48
Cinebench 10 (SMT on) 3.98 4.07 4.09

These results indicate that performance should be considered a reasonable reason to upgrade from Windows XP to Windows 7 when considering whether to deploy Windows 7, but the performance factor is not enough to make Vista users turn to Windows 7. Windows 7 based on the factors of the Vista kernel makes the performance of the two close.

Surprisingly, the multithreading changes in Windows 7 did not provide more performance gains. The explanation for this is how Windows 7 manages hyperthreading. Major changes in Windows 7 multithreading capabilities include increased processor affinity and Windows kernel scheduling lock changes. This technique, called "glasseye", is at the heart of modern operating systems, how the kernel prevents two threads from accessing the same data or resources at the same time.

Any time a thread wants to access data that another thread may need to access, it must use a locking mechanism to ensure that only one thread can modify the data. Prior to Windows 7, when a thread needed to acquire or access a lock, it needed to pass a global locking mechanism. This mechanism handles the proposed requirements through kernel dispatch locks. But because it is unique and global, it can handle the demands of thousands of Windows hardware platforms. Therefore, this scheduling lock has become a major bottleneck. In fact, this is the main reason why Windows Server can only run on up to 64 processors for a long period of time.

New Locking Mechanism

Windows 7 includes a completely new mechanism to get rid of the concept of global locking and push lock management into locking resources. This allows Windows 7 to scale to 256 processors without performance penalty. In systems with only a few processors, the performance of this new mechanism for desktop and small servers is not obvious because the old kernel scheduling locks are not overburdened.

Although the SMT runtime is not visible, it is expected that the benchmark will utilize all available resources. When running a four-threaded Viewperf benchmark with SMT functionality, the results remain essentially unchanged.

The Cinebench benchmark is a tool that measures how fast multiple threads can run faster than a single thread. Whether or not the SMT function is activated, Cinebench has only a negligible difference in testing the performance of the three operating systems. But unlike Viewperf, the results of the three Windows systems scored significantly better when SMT was turned on than when it was not turned on. That is to say, no matter which version of Windows, Cinebench rendering runs at 8 threads (SMT on) and performance is 20% higher than 4 (SMT off) threads. The difference between the two aspects of the benchmark test makes it necessary for the user to test before deciding whether the existing application is to enable the SMT function.

Power Consumption

Windows 7 provides several techniques for running threads in the same execution pipeline, enabling the underlying Nehalem processor to turn off transistors that are less used or not working. The main advantage of this feature is to reduce energy consumption. To quantify the benefits of this feature, the tester ran a 4-threaded version of the Viewperf program with SMT functionality. This configuration means that about half of the pipes will not see any activity. The data in the table below reflects the energy savings of Windows 7.

Three versions of Windows Viewperf benchmark results

Energy survey items Windows XP SP3 Vista Ultimate SP2 Windows 7 Ultimate
Power consumption (three test average) 247 watts 248 207 watts

As seen from the table, Windows7 has a great advantage, which can save about 17% of energy consumption. And this test is run on the same workstation, and the hardware configuration is exactly the same, only the Windows version is different. This is a significant energy savings, and there are good reasons to believe that other software will also benefit from the power of Windows 7, and take full advantage of the magic of Intel processors.

Combining with Intel Processors

Microsoft operating systems and Intel processors have become standard on desktop systems for years. Although not every cycle of iterative products brings a lot of benefits to end users, the close relationship between Microsoft and Intel has been called a virtuous circle by some insiders. This time, the new product linkage cycle does offer some key advantages: the Nehalem processor outperforms its predecessor and is already available on the market. The new processor has many energy-saving technologies that can be effectively utilized. On the Microsoft side, Windows 7 is launched at the right time, and it is equally good at energy saving.