Dual-core ERA test Pentium D Server application evaluation

Server

The NetBurst architecture has been in its 7th year since its release in 2000. Over the years, the processor in this architecture has been changing, with the core experiencing Willamette, Northwood, Prescott, Cedar Mill, Processor process from the first 0.18 micron to 65 nm, the main frequency from the initial 1.3GHz to the current 3.8GHz, processor cache, front-end bus and such as HT technology, EM64T, eist technology, etc. are added to the NetBurst architecture based Pentium 4 series processor.

Since last year, Intel, plagued by process problems, has slowed the pace of processor speed and failed to allow the processor to break 4GHz in 2005. More importantly, Intel acknowledges that in today's application environment, the simple lifting processor frequency has been unable to meet the user's demand for performance, only a dual-core (DualCore) processor can be used to meet multitasking parallelism.

Pentium D 930 Processor

Pentium D 930 Processor

Intel immediately began to follow the dual-core market. The first issue is a Pentium D 8xx series dual-core processor based on the Smithfield Core, which essentially encapsulates the two Prescott processors, thus continuing to be a continuation of the NetBurst architecture. Intel then further launched the Pentium D 9xx series dual-core processor based on the Presler core, which encapsulates the two cedarmill cores in the Presler core. In addition to the process improvements, the Intel Pentium D 9XX series processors also increase support for VT technology.

Intel dual-core processor specification comparison
	800 Series	840EE	900 Series	955EE
Core	Smithfield	Smithfield	Presler	Presler
Manufacturing process	90nm	90nm	65nm	65nm
Core Quantity	2	2	2	2
Frequency	2.8/3.0/3.2	3.2GHz	2.8/3.0/3.2/3.4	3.46GHz
Front Side bus	800FSB	800FSB	800FSB	1066FSB
L2 Cache	2x1mb	2x1mb	2x2mb	2x2mb
HyperThreading (HT)	Whether	Is	Whether	Is
EIST	Is	Is	Is	Is
em64t	Is	Is	Is	Is
Execute Disable Bit	Is	Is	Is	Is
Vanderpool	Whether	Whether	Is	Is
LaGrande	Whether	Whether	Is	Is
Tdp	95/130/130w	1206	95/95/130/130w	1206
Tcase	64.1/69.8/69.8℃	69.8 ℃	63.4/63.4/68.6/68.6℃	68.6 ℃
Price	178/218/273$	-	241/316/423/637$	-

The Presler core L2 cache capacity is one times the Smithfield core, reaching the 2X2MB. Fortunately, the power did not increase significantly, TDP remained between the 95-130w.

Cpuz 1.31 Reconnaissance Information

Test platform and test methods

Test Platform Configuration

Processor Dual Xeon 3.0Intel Pentium M t2600intel Pentium d 830Intel Pentium D 930AMD Opteron 265 core Noconayonahsmithfieldpresterit Aly process 90nm65nm90nm65nm90nm core voltage -0.900-1.400v1.2-1.4v1.2-1.3375v1.350v frequency 3.0ghz2.16ghz3.0ghz3.0ghz1.8ghz frequency multiplier X15X13X15X15X9FSB or H Tt200mhz167mhz200mhz200mhz200mhz Bus frequency 800mhz667mhz800mhz800mhz-l1 d-cache16kb,8 Road 32kb,8 Road 16kb,8 Road 16kb,8 Road 32KB, 8 Road, 32kb,2 Road, 2mb,8 Road, 2mb,8 Road, 2x1mb,8 Road 2x2mb,8 Road 2x1mb,16 Road We have installed Microsoft Windows 2003 Enterprise Edition (English, 32bit) on the hardware platform of the above configuration. Correctly install the drivers for each hardware to ensure that the server is working optimally. SEPCCPU2000 can evaluate the integer performance and floating-point performance of the processors in the server system, and we are testing the specint_base2000 and specfp_base2000 two projects. Sciencemark v2.0, a software that evaluates the capabilities of the processor's scientific computing, provides the ability to evaluate the memory bandwidth of the system being tested, which can support multithreading very well. Iometer is an I/O subsystem evaluation tool that can be used for a single system or cluster system, and we only use its disk testing capabilities. Webbench and NetBench are two commonly used software in server evaluation, which focuses on the performance evaluation of the server CPU subsystem, while the latter focuses on performance evaluations of IO subsystems.

We used the SiSoft Sandra Pro sr2a to test each system, which is more familiar to the average reader, and can even use the shared version to test its own system directly, so there is a more intuitive concept of our test results.

CPU arithmetic benchmark includes the Dhrystone ALU, Whetstone FPU, Whetstone iSSE2 three subprojects, and the impact of multiprocessor, hyper-Threading technology on this test is obvious. In the ALU test, Intel Pentium D 930 did not show the advantage over Intel Pentium D 830. The FPU and iSSE2 test subprojects also show that 830 and 930 of the processors with the same frequency have no performance differences due to different cache sizes.

The CPU multi-media benchmark test consisted of 2 subprojects, the performance of the dual Xeon platform was the best, followed by Pentium M t2600,pentium D 830 and 930, which remained no different.

Spec CPU2000 v1.2 Test spec is the abbreviation of the standard performance evaluation Company (Standard Performance Evaluation Corporation). Spec is a non-profit organization made up of computer manufacturers, system integrators, universities, research institutes, and consulting companies that aim to establish and maintain a set of standards for evaluating computer systems. Spec CPU2000 is a set of CPU subsystem evaluation software introduced by spec, which includes CINT2000 and CFP2000 two subprojects for measuring and contrasting integer performance, while the latter is used to measure and contrast floating-point performance. The processors, memory, and compilers in the computing system affect the final test performance, while the impact of I/O (disk), network, operating systems, and graphics subsystems on the spec CPU2000 is very small. We have installed the Intel C + + 8.1 Compiler, Intel Fortran 8.1 Compiler The two spec CPU2000 required compilers on the tested server, and Microsoft Visual Studio 2003 is also installed. NET provides the necessary library files. According to Spec, we edited the new config file according to our own situation, we can meet the base test, and the following test results are 2-4 users, specint_rate_base2000 and Specfp_rate_ under the Hyper-threading condition of open dual Xeon processor base2000.

specfp_rate2000 multiple instances (instance) at the same time in the testing process, measuring the ability of the system to perform compute-intensive floating-point operations, such as CAD/CAM, DCC, and scientific calculations can be used to refer to this result. Specint_rate 2000 simultaneous execution of multiple instances (instances) and the ability of the system to perform multiple compute-intensive integer operations at the same time can well reflect such things as database servers, Performance of multiprocessor systems such as e-mail servers and Web servers that are based on integer applications.

The floating-point performance of Intel Pentiu D 930 is very good, with an obvious advantage over the other three platforms, with integer performance comparable to them. Since we do not have the test results for this part of the Pentium D 830 processor, we cite a set of data published by Spec to illustrate that int_rate and fp_rate are respectively 31.0 and 31.4, similar to the performance of Pentium D 930.

Sciencemark v2.0 Membench

Sciencemark v2.0 is a software used to test the performance of the system, especially the processor in a scientific computing application, Membenchmark is a functional module designed for processor caching and system memory that can test system memory bandwidth, L1 cache latency, L2 Cache latency and system memory latency, in addition to testing the performance differences of different instruction sets.

Sciencemark Membench
	AMD Opteron 265	Intel Pentium D 830	Intel Pentium D 930	Intel Petium M T2600	Dual Xeon 3.0GHz
Memory bandwidth (MB/s)	4817.58	4429.8	4065.62	3444.99	4073.71
L1 Cache Latency
Bytes Stride	3 Cycles/1.67ns	4 Cycles/1.33ns	4 Cycles/1.33ns	3 Cycles/1.38ns	3 Cycles/1.00ns
L2 Cache Latency
4 Bytes Stride	3 cycles/1.67 NS	6 cycles/2.00 NS	4 cycles/1.33 NS	3 cycles/1.38 NS	6 cycles/2.00 NS
Bytes Stride	5 cycles/2.79 NS	cycles/4.33 NS	9 cycles/3.00 NS	5 cycles/2.31 NS	cycles/4.33 NS
Stride Bytes	cycles/9.47 NS	cycles/9.67 NS	cycles/9.33 NS	cycles/6.46 NS	cycles/9.00 NS
256 Bytes Stride	cycles/6.69 NS	cycles/9.33 NS	cycles/9.00 NS	cycles/6.46 NS	Num cycles/8.67 NS
Stride Bytes	cycles/7.24 NS	Num cycles/8.67 NS	Num cycles/8.66 NS	cycles/6.46 NS	cycles/8.33 NS
Memory Latency
4 Bytes Stride	3 cycles/1.67 NS	7 cycles/2.33 NS	5 cycles/1.67 NS	4 cycles/1.85 NS	6 cycles/2.00 NS
Bytes Stride	cycles/6.69 NS	cycles/5.00 NS	cycles/4.33 NS	cycles/6.00 NS	cycles/5.00 NS
64 Bytes Stride	48 NS	43 NS	48 NS	53 NS	49 NS
256 Bytes Stride	103 NS	270 NS	298 NS	202 NS	376 NS
Stride Bytes	cycles/59.07 NS	284 cycles/94.66 NS	309 cycles/102.97 NS	205 cycles/94.61 NS	395 cycles/131.66 NS

The L1 latency of the 5 platforms is 3-4 cycles, but due to the different clock bus frequency of the processor, the actual delay time gap is obvious, the shortest 1ns,pentium D and Pentium m of the Dual Xeon platform are 1.33NS,AMD Opteron 265 the longest is 1.67ns.

The L2 delay test shows that the test results for the Pentium D 830 platform and the dual Xeon platform are very approximate, while the results of the Pentium M T2600 platform and the AMD Opteron 265 platform are approximate-the architecture they employ is obviously more efficient, Pentium D The L2 latency of the 930 platform is better than that of Pentium D 830 and the dual Xeon platform, but there is a significant gap with the Pentium M T2600 platform and the AMD Opteron 265 platform.

The memory used by these 5 servers is not exactly the same, AMD Opteron 265 is configured with two 1GB DDR400 memory modules, Pentium D 830 is configured with two 256MB ddrii 533MHz memory modules, and Pentium D 930 is configured with two 512MB Ddrii 533MHz Memory module, Pentium M T2600 is configured with two 256MB DDR2 533MHz memory modules, Dual Xeon 3.0 is configured with 4 512MB ddrii 400MHz memory modules. Different memory types, timing, and capacity can have a significant impact on memory bandwidth and latency.

Webbench v5.0 Test Webbench is designed to test the performance of the server as a Web server, we have installed the IIS6.0 component on the tested server to provide the Web services required for testing. In the test we opened the network Lab 32 PCs (Celeron 1.7/256mb/40gb/100ethernet) as the client, using the Webbench 5.0 built-in dynamic CGI dynamic page script to test the server.

The performance of Intel Pentium D 930 is very low, even lower than the performance of Intel Pentium D 830, which we tested earlier, which is similar to the dual Xeon 3.0GHz platform that turns off Hyper-threading. In contrast to the outstanding performance of AMD Opteron 265 duo processors and Intel Pentium M T2600, I would like to lament that the NetBurst architecture has come to an end.

Evaluation Center View: Say goodbye to NetBurst.

As we have said, the Intel Duo Core T2600 processor and the AMD Opteron 265 processor are very well worth considering from a purely performance perspective, and their frequencies are around 2GHz, But it has the same performance as a dual-core processor (or two-way processor) with a NetBurst architecture with a 3.0GHz processor. In particular, the TDP of the Intel Duo Core T2600 processor is only 31 watts and is ideal for intensive deployments if applied to server applications.

The test showed that Intel's NetBurst architecture was so senile that even a two cache increase of one-fold did not make its performance a little bit better. The only attraction is the VT technology, but it will not be accepted by the home user until the application is more usable. Webbench test Results Let us discard the idea of recommending users to apply them to the server.

Intel obviously has an earlier arrangement for the fate of NetBurst. From the above roadmap, we can learn that Intel plans to introduce the core Conroe of the new architecture in the third quarter of next year, and will not introduce new products based on NetBurst architecture.

（