Disable 387 FPU

[Xacalite]

So I'm benchmarking some math software on a 386 with FPU, and want to measure the performance impact of the FPU.
But I don't feel like physically removing the chip - anybody knows a program to disable it?

 

[kgober]

As far as I know there's no way to disable the hardware via a software setting.

But many math libraries had a configuration option to disable their use of floating-point hardware (presumably to allow easy testing without physically removing the chip), for example by setting the "NO87=1" environment variable. Do you have documentation for this math software? It might have such an option.

 

[Agent Orange]

So I'm benchmarking some math software on a 386 with FPU, and want to measure the performance impact of the FPU.
But I don't feel like physically removing the chip - anybody knows a program to disable it?

I have a 386 with an AMIBIOS. Under 'Advanced CMOS Setup' it has a provision for 'Numerical Processor Test', but no way to enable or disable the FPU.

 

[twistedpneumatic]

Write software that ises the CPU and FPU directly? Find some different software?

 

[Stone]

Write software that ises the CPU and FPU directly? Find some different software?

Sorry, but that's way beyond my pay grade.

 

[twistedpneumatic]

How about shifting that i key one to the left. Uses sounds a little better.

 

[Xacalite]

Too bad...
I tried to prevent software from using FPU by zeroing the bit 1 at 0040h:0010h, but it's not enough...
Checkit still detects FPU and benchmarks it as usual.
Mathematica on startup displays a warning about the lack of FPU, but later on works as usual.

 

[Stone]

How about shifting that i key one to the left.

Since you're the OP I would suspect that's your responsibility.

 

[Stone]

Mathematica on startup displays a warning about the lack of FPU, but later on works as usual.

Have you tried a coprocessor emulator?

 

[Xacalite]

Have you tried a coprocessor emulator?

No, but I doubt it would do any change.
Coprocessor emulators are based on some exception generated by CPU on FPU instructions executed without an actual FPU, right?
If the FPU is present, such exceptions should never occur.

 

[Jo22]

So I'm benchmarking some math software on a 386 with FPU, and want to measure the performance impact of the FPU.
But I don't feel like physically removing the chip - anybody knows a program to disable it?

I remember there was a DOS utility, hide87.com, for that purpose.
I believe it was from German c`t magazine. However, I have not really tested it.

 

[Xacalite]

I remember there was a DOS utility, hide87.com, for that purpose.
I believe it was from German c`t magazine. However, I have not really tested it.

There was some problem with the forum, recent posts got lost, so let me reply again...

I tried HIDE87.COM:

"Terminate and Stay Resident code that hooks Int 11h (BIOS equipment check) to fool some programs (notably Windows' Kernel) into thinking there isn't a math coprocessor."

But it's of limited use - eg. Checkit 3.0 still detects FPU and benchmarks it as usual.

 

[Jo22]

Thanks for the feedback, Xacalite!
I almost thought my reply was lost.
Also sorry for the program. I wasn't sure how well it works.

Maybe there are different versions of the program..
The version comes with Winfloat outputs a different string than the c't version I'm familiar with (see attachment).