This really sucks!

[RJARRRPCGP]

Switch to Intel

But the thing is, I thought that later Windows versions, especially Windows XP were even more pro-Intel! I heard that it's best to use an earlier Windows version for AMD systems!

[LLXX]

Don't forget that in the days of Win98, AMD was not as popular as it is today, and thus the majority of code was optimised for Intel CPUs such as the Pentium II.

[somewan]

L1,L2,L3 caches are managed by the hardware. The software need not intervene.

Regarding size of cache, I have a circa. 1993 80486dx2-66 with 512KB of L2, and that was around before 98se...

But, this may be because of another problem. It's possible that Windows 98 SE assumes that it only has 256 KB of L2 cache, because most Athlon processors have 256 KB of L2 cache. (before "Barton") The Windows 98 SE processor driver may be getting confused. It may not like the fact that the L2 cache is integrated. Most processors at the time Windows 98 SE was released didn't have integrated L2 cache!

Operating systems do not "select" how much cache to

use - they either use all or nothing of it, and disabling

the cache would only be useful for troubleshooting.

The fact that Win98 is so much more compact than XP/2K

should affect performance positively as far as the cache

is concerned, because a larger portion of the system will

fit into it.

There other CPU-related features that affect performance,

of course, such as the Memory-Type Range Registers that

Pentium Pro and newer Intel CPUs support, and that can

be used to select caching strategies for portions of the

address space, as appropriate. I assume AMD processors

have something similar.

For example, enabling maximum read/write caching for the

video memory can greatly enhance performance, because

it allows the CPU schedule accesses to it in whatever order

it finds optimal.

However, using the same caching strategy for self-modifying

code (e.g. JIT compilation) or for data segment that need to

be accessed in the order assumed by the compiler or

programmer, would crash either the application in question,

or whole the system.

The default settings programmed by the BIOS at startup,

or the CPU's power-on defaults, are likely to be selected

with stability and compatibility in mind, and of course, the

settings will remain that way until reprogrammed - and

that's where device-, chipset- or CPU-specific drivers

come in.

For example, the DOS kernel basically regards any CPU as

an 8086, and it requires drivers like HIMEM, QEMM or

386MAX to take advantage of more advanced functionality.

Windows 95 is written to run on a 386 (I've tested).

I think Windows 98 demands a 486 at minimum. However,

the setup program installs more specific drivers where

available.

For example, the VMM32.VXD-archive on my Celeron 400

system includes a module called MTRR.VXD. I imagine that

the absence of such processor-specific support could

adversely affect performance to significant degrees.

[LLXX]

I think Windows 98 demands a 486 at minimum.

For more speed, but a 386 is all that's needed... I've tried a 33MHz 80386DX and 98se does indeed run on it.

[RJARRRPCGP]

I'm wondering if most games aren't affected?

[LLXX]

Anything running on the CPU will be affected by the microcode updates...

[RJARRRPCGP]

Anything running on the CPU will be affected by the microcode updates...

Likely not if the application wasn't compiled with a pro-Intel compiler. Super Pi was unfortunately compiled with a pro-Intel compiler, AFAIK.

Super Pi, at least the non-modded one was written in 1995! Back then, AMD wasn't popular at all!

In 1995, I didn't even have a PC at all!

Edited February 21, 2006 by RJARRRPCGP

[RJARRRPCGP]

I have gotten wording about the following:

Intel made it's complier so that it's slower on purpose if a non-Intel processor is detected. Even if the processor supports the functions required.

[LLXX]

I have gotten wording about the following:

Intel made it's complier so that it's slower on purpose if a non-Intel processor is detected. Even if the processor supports the functions required.

Makes sense. It is the Intel compiler after all... so it'd be optimised only for their processors.

[Lunac]

All this talk about a possible performance gap? Gap that might manifest itself under certain uncommon conditions (very uncommon conditions) such as running certain stress-test/bechmarking software dating from mid 90s on a particular OS. (I say might because who knows what configuration of hardware/software dear RJARRRPCGP is running, and that is if RJARRRPCGP is not BS'ing us to begin with)