PCMark2002 – A First Look

Pages: 1 2 3 4 5 6

CPU Test Results

PCMark2002 uses six different algorithms to test the performance of the processor. These tests stress both the integer and floating point unit, and include some SSE and SSE2 optimizations. Both MS Visual C++ and Intel C++ compilers were used, and the specific executable used depends upon which processor is being tested (i.e., if it supports SSE, then the SSE enabled executable is used). This ensures that the features providing the greatest performance are utilized for each processor, at least as much as is possible.

The Help file describes it this way: “There is one Intel compilation optimized for CPUs with SSE support and one optimized for SSE2 support. Some tests are compiled faster with the default (MS VC++) compiler, while others get better results with the Intel Compiler. PCMark2002 therefore uses the optimal compilations for Intel CPUs supporting SSE, Intel CPUs supporting SSE2, and AMD CPUs supporting SSE. All other CPUs will run all tests compiled with the default compiler.”

The six specific CPU tests are:

  1. JPEG decompression – This test measures how many image pixels are decompressed in 10 seconds by processing 3 images sequentially, which are 149kB, 771kB, and 889kB in size. The result unit of the test is average Mpixels/s. The help file indicates that the standard JPEG library (version 6b) from Independent JPEG Group (www.ijg.org) is used.
  2. Zlib compression – Three files are compressed as many times as possible in 10 seconds, which are a 887kB JPG image, a 1468kB text file, and a 1280kB executable. This test (and the next one) uses the LZ77 compression method from the open source ZLIB (www.gzip.org/zlib/). Integer performance is the critical factor here.
  3. Zlib decompression – The archive created above is decompressed as many times as possible in 10 seconds. The results for both compression and decompression are given in MBytes/s.
  4. Text search – This test simply searches a large text file for many strings using the Boyer-Moore algorithm for 10 seconds. The result is the number of search loops executed, with each loop consisting of multiple string searches. Again, integer performance is what is being measured.
  5. Audio Conversion – Here, a 500kb compressed MP3 file is decompressed using the Microsoft MP3 decoder and then compressed to Ogg Vorbis format (http://www.gnu.org/directory/oggvorbis.html), intended to emulate MP3 decompression and playback. The result is expressed at time spent in KB/s.
  6. 3D Vector Calculation – In this test, human hair (and movement) is simulated using polygon lines, as might be done in a game or 3D modeling. The test is run for 10 seconds and the test result is the amount of frames completed in that time. This one uses some assembly language for optimization, and no specific routine is mentioned.

The CPU Overall score is calculated using the following formula:

JPEGDecompression*60,6 + (FileCompression*153,8 + FileDecompression*12,4)/2 + TextSearch*4,9 + AudioConversion*11,1 + 3DVectorCalculation* 16,7

According to the details given in the help file, a high end PC (as of the release of the benchmark) should get around 5000 points as the CPU score.

OK, so what do my test results look like…

P4 Willamette

1.4GHz

1.6GHz

1.8GHz

2.0GHz

JPEG

10.5

12.1

13.5

15.1

Zlib Compression

4.1

4.6

5.2

5.7

Zlib Decompression

54

60.9

67.4

73.8

Text Search

187.9

213.7

238.8

263.9

Audio Conversion

55.7

63

70

76.8

3D Vectorize

34

38.8

43.6

48.4

CPU Overall

3398

3859

4314

4766

P4 Northwood

2.0A GHz

2.2 GHz

2.4 GHz

JPEG

15.2

16.8

17.9

Zlib Compression

5.9

6.5

7.3

Zlib Decompression

74

78.3

85.4

Text Search

262.9

285.6

306.7

Audio Conversion

83.4

91.4

100

3D Vectorize

48.7

53.6

58.4

CPU Overall

4864

5311

5764

One of the first things you should notice is that both 2GHz processors get very nearly the same scores. This seems to imply that the CPU tests fit within the 256KB cache of the Willamette, and therefore don’t accurately show the performance of this processor in ‘real world’ applications. However, it may be an indication that the benchmark has effectively isolated the CPU performance without regard to the cache size. It is important to note that we have not yet seen the memory scores – which might provide some interesting information!

This next chart simply shows that the scaling for P4 is fairly smooth, with the lines flattening out between the 2.0 and 2.0A processors.


Pages: « Prev   1 2 3 4 5 6   Next »

Be the first to discuss this article!