EMM386.EXE slows down the computer: Why?

[Wayne Smith]

I recently installed DOS 6.2, and I tried the program called memmaker.exe.
Even though I told it I had no need for expanded memory, it added the
line device=emm386.exe NOEMS (or something like that) to my config.sys.

I noticed that with emm386 that I had >640k of conventional memory
available, but without it I had exactly 640k available. Why?

I ran the processor test from PCBENCH 8.0 with and without emm386,
and the Floating Point emulation dropped from 672 to 429 and the
math co-pro test dropped from 6415 to 5165 (ie it ran slower with emm386).

What's going on here?
Is the benefit of getting a few extra k of ram worth a 25% performance hit?

[Jan Just Keijser]

In article <2uqpbj$k...@falcon.ccs.uwo.ca> wls...@valve.heart.rri.uwo.ca (Wayne Smith) writes:
>Subject: EMM386.EXE slows down the computer: Why?
>From: wls...@valve.heart.rri.uwo.ca (Wayne Smith)
>Date: 29 Jun 1994 03:21:23 GMT

This sounds right. When you install EMM386.EXE (the NOEMS switch indicated
indeed that you do not want EMS support) you get access to the memory region
from 640kb to 1 Mb. Especially when loading network drivers it is nice to be
abel to park some software up there, because you'll have more memory left for
your DOS applications. The performance drop comes from the fact that EMM386
switches the processor to another mode (from real mode to V86 mode) in order
to be able to give you the access to the high memory regions (UMBs). This V86
mode is an emulation of 8086 real mode and is always slower than real mode
itself. The penalty when using EMM386 is a lot larger than when using QEMM386,
however. All 386 memory managers switch to V86 mode and are therefore (almost)
always introduce a penalty compared to real mode.

Is it worth it? Judge for yourself....

Jan Just

[Cyril Chui]

Even if your PC is not on network, you can still enjoy some benefit from
emm386 by loading smartdrv(26k) and doskey(4k) to UMB. If you have network
or other device drivers as TSR, you can save as much as 100K in Conventional
memory. Memmaker should add the line .. emm386 ... in config.sys for you, and
put the prefix devicehigh and loadhigh to config.sys & autoexec.bat resp.

You should have 655360 (640k exactly) with emm386, but less (651264) without.
Don't know why though.

--
|Equity Systems Pty Limited |
|Level 11, 270 Pitt Street, Sydney, N.S.W. 2001, Australia |
|Phone: +61 2 373-7900, Fax: +61 2 267-3484 |

[jki...@crash.cts.com]

> >I recently installed DOS 6.2, and I tried the program called memmaker.exe.
> >Even though I told it I had no need for expanded memory, it added the
> >line device=emm386.exe NOEMS (or something like that) to my config.sys.
>
> >I noticed that with emm386 that I had >640k of conventional memory
> >available, but without it I had exactly 640k available. Why?
>
> >I ran the processor test from PCBENCH 8.0 with and without emm386,
> >and the Floating Point emulation dropped from 672 to 429 and the
> >math co-pro test dropped from 6415 to 5165 (ie it ran slower with emm386).
>
> >What's going on here?
> >Is the benefit of getting a few extra k of ram worth a 25% performance hit?
>

EMM386 Switches the processor in Virtual 86 (Protected mode) in order to do
some of the more fancy memory managment techniques. When the processor is in
protected mode, and a real-mode interrupt handle is trigger, it causes the
processor to swithch back and forth from real to protected mode thus causing
the system to slow down a little. This is a well documented fact.

Jamey

[WU DAVID S]

I tried this experiment (e.g. using himem noems and not) and
I ran the Norton Sysinfo program and there was no differece.

Is sysinfo just messed up?

[Osmo Ronkanen]

In article <CsJ4q...@ecf.toronto.edu>,

WU DAVID S <wud...@ecf.toronto.edu> wrote:
> I tried this experiment (e.g. using himem noems and not) and
>I ran the Norton Sysinfo program and there was no differece.
>

himem noems???

Well EMM386 puts the CPU always in V86 mode, no matter what the
parameters are. Did you try without EMM386.

> Is sysinfo just messed up?
>

Maybe it just won't test features that are slowed down. Get a better
benchmark, like comptest 2.59.

Osmo

[Ian Smith]

In article <2uqpbj$k...@falcon.ccs.uwo.ca>
wls...@heartlab.rri.uwo.ca "Wayne Smith" writes:

It gets worse. Try benchmarking the disk and screen access. They both take
hits with EMM386 loaded. Just another example of "no free lunch". Since
you're using MSDOS 6.2, you can set up several configs, with and without
EMM386, and tune your setup to your requirements.

Regards
Ian.
--

| Ian Smith | "The Moving Finger writes;
| i...@isis.demon.co.uk | and, having writ, Moves on."

[Akopov Mikhail]

WU DAVID S (wud...@ecf.toronto.edu) wrote:
WDS> I tried this experiment (e.g. using himem noems and not) and
WDS>I ran the Norton Sysinfo program and there was no differece.

WDS> Is sysinfo just messed up?
Sysinfo measures the CPU perfomance which is not changed under the
EMM drivers.
"Brakes" are on in interruptions process.
During the real work hardware (any) and software (BIOS&DOS calls)
produces a lot of interrupts, thus the system perfomance lows.

Vale! -Michael Akopov

[Michel Nulion]

WU DAVID S (wud...@ecf.toronto.edu) wrote:

: I tried this experiment (e.g. using himem noems and not) and

: I ran the Norton Sysinfo program and there was no differece.

EMM386 puts the processor in virtual 86 mode (i.e. protected).
And it is widely known that protected mode instructions are slower than
their real-mode equivalents.

: Is sysinfo just messed up?

Maybe sysinfo has some strange way of measuring CPU speed. When
I move the mouse the performance jumps down by 0.8 immediately.