: There was another interesting example in this month's Byte.
: They compared WinNT 4.0 on a 6x86 chip and a Pentium Pro. In all the beta
: versions, the two chips had similar performance and NT 4.0 was _faster_
: than NT 3.51 and Win95.
: In the release version, the above was also true of the Pentium Pro.
: Interestingly, on the 6x86 system, there was a _huge_ slowdown in all the
: tests. Microsoft admitted it and said something about inserting code to
: correct for the instability of the 6x86.
For some reason, NT 4.0 doesn't quite like the Write Back cache on the
6X86's L1. So it would work for 99.9999% of the time, but apparently fom time
to time it would hiccup, (excuse the extremely technical discussion :)
So code with MS inserted checks for 6x86 cpus, and if it found one, disabled
the L1 cache to do Write-Through only. This would account for the ~20%
performance hit reported.
Apparently NT4.0 is heating up demand for the P6 as well, the demand for
P6 has now outstripped supply, and Intel is putting its customers on
allocation for P6's now. Cyrix, OTOH, has plenty of CPUs, and has just
announced a big price cut. Its 133 Mhz 6x86, (P166+) now goes for $156.
: I sure hope the DOJ is watching this one, too.
: Regards, Joe Ragosta