« Software developers must clean up their act | Main | Pin and chips »
How accurate is Sysmark?
The results of your tests show that the AMD Athlon64 was faster in all but one video encoding test
compared to the Intel CPU where it was 20% slower.
During some tests this year, a commerical Windows 32bit application was built using Intel compilers with optimisation enabled. The same application was then built using a Sun compiler and run on the same dual CPU AMD hardware but with no optimisation flags enabled.
In each case the PC had a clean install of 64bit Windows or Solaris installed, and the same test data
was used.
The Sun 32bit compiled code was between 140% - 180% quicker. Thus, in any test comparing AMD and Intel a vendor-neutral compiler is essential. Given that Intel produces faster compilers than Microsoft it is fair to assume that the Sysmark tests are built with the Intel compiler.
Paul Smith
Unfortunately you're barking up completely the wrong tree: Sysmark 2004 is not a synthetic benchmark and hence has no need of a compiler - it's a script that runs a workload of tasks using 17 real-world, full applications including MS Office, Photoshop and Dreamweaver.
Any compiler optimisations in the individual applications is beyond the control of Bapco, and any such optimisations would in any case reflect what users would experience when running those applications.
For a technical white paper explaining how Sysmark 2004 SE works, go to the Bapco web site at http://www.bapco.com/techdocs/SYSmark2004SEWhitePaper.pdf
Kelvyn Taylor
Deputy Editor, PCW
Posted by Kelvyn Taylor | July 8, 2005 12:13 PM