Articles: CPU
 

Bookmark and Share

(29) 
Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 ]

At first I would like to say a few words about the 64bit performance. Unfortunately, our opportunities were pretty limited because there were hardly any applications compiled for native 64bit mode. The only available benchmark for Win64 is a 64bit version of SiSoft Sandra 2004 SP1. This test package includes short synthetic benchmarks for the CPU and system memory. Due to these tests we get at least some idea of the system performance in Windows XP 64-Bit Edition for 64-Bit Extended Systems. At the same time I would like to say that SiSoft Sandra 2004 SP1 also exists for Win32. So, we get an excellent opportunity to compare the results of this test during 64bit code processing in 64-bit Windows XP with those during 32bit code processing in 64-bit and 32-bit Windows XP versions.

SiSoft Sandra 2004 SP1

Win64
64-bit exe

Win64
32-bit exe

Win32
32-bit exe

CPU Arithmetic Benchmark

Dhrystone ALU

10213

10441

10387

Whetstone FPU

4277

3797

3790

Whetstone SSE2

4547

4960

4854

CPU Multimedia Benchmark

Integer MMX/SSE

17729

22913

22888

Floating Point SSE2

25451

24631

24611

Memory Bandwidth Benchmark

Int Buffered

6051

5633

5668

Float Buffered

6006

5613

5629

The results are pretty ambiguous. In some tests the shift to 64bit mode leads to a notable performance gain, while in some cases the performance drops significantly. However, first of all I would like to say that 32bit benchmarks from the Sandra package were run equally fast in 32bit and 64bit operation system. This way, I can conclude that WOW64 emulation doesn’t cause any delays of the 32bit code processing.

Returning to the results in 64bit SiSoft Sandra 2004 in Win64 I would like to point out that the shift to 64bit mode allows AMD64 processors to speed up the work with the system memory quite tangibly. However, this is almost the only case (besides Whetstone FPU benchmark) when 64bit technology has a positive effect on the performance, according to SiSoft Sandra 2004. In all other cases the performance in 64bit mode is either the same or lower. Unfortunately, we can’t state with all certainty why the 64bit performance is considerably lower than 32bit performance in Multimedia Integer MMX/SSE and Whetstone SSE2 benchmarks. The most probable explanation here is the poor optimization of the test algorithms for AMD64 technology. All exiting compilers generating the code for Windows XP 64-Bit Edition for 64-Bit Extended Systems are only available as alpha- or beta-versions, that is why no one can guarantee their efficient work today.

At the same time, if the code of the 64bit SiSoft Sandra version is not very well optimized for AMD64 architecture (for instance, it uses nearly none of the additional general purpose registers), we can also face a noticeable performance drop but for a different reason. In 64-bit mode Athlon 64 processors perform certain instructions (namely integer multiplication and division) somewhat slower than in 32-bit mode. However, this explanation suits only for Dhrystone ALU benchmark, because Athlon 64 performs all SSE2 instructions equally fast in 32-bit and 64-bit modes.

Unfortunately, besides SiSoft Sandra 2004, there are no other applications at our disposal today supporting 64-bit mode and offering some benchmarking opportunities. That is why let’s pass over to a more indepth investigation of the 32bit applications performance in Windows XP 64-Bit Edition for 64-Bit Extended Systems.

 
Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 ]

Discussion

Comments currently: 29
Discussion started: 03/18/04 12:32:06 PM
Latest comment: 01/22/08 02:40:42 AM

View comments

Add your Comment