Autor: Tomek Lutelmowski (natenczas_at_energoprojekt.pl)
Data: Thu 11 Oct 2001 - 16:04:23 MET DST
had intended my next article in this column to be a look at the Winstone
benchmarks (which is in the works), however the news and reaction to the
SysMark 2001 'bug' caught my attention. Essentially, SysMark 2001 has an
inherent problem that seemingly puts the Athlon XP, Athlon MP and Duron
(w/Morgan core) processors at a disadvantage in the benchmark. It turns out
that Windows Media Encoder 7, which is used in the benchmark, checks
specifically for the CPUID string of 'GenuineIntel' in order to determine if
it should use the SSE instruction path or not (just as we thought developers
had learned their lesson about such things).
This, of course, has reignited accusations that Intel is influencing BAPCo,
and that the benchmark is biased. Some have suggested that BAPCo should have
put out a patch, and that by not doing so they have 'proven' their bias. AMD
did put out a patch that replaces the check for 'GenuineIntel' with a check
for 'AuthenticAMD', enabling the SSE code path for the Athlon XP. Many
reviewers and users have claimed that this now makes the benchmark a
realistic and fair benchmark. However, I can make the argument that it makes
the benchmark biased towards AMD.
całość na:
http://www.realworldtech.com/page.cfm?section=columns&AID=RWT101001162806
To archiwum zostało wygenerowane przez hypermail 2.1.7 : Tue 18 May 2004 - 23:05:20 MET DST