Saving the long boring explanation, on Arm CPUs much 32bit software does not use the Maths FPU, or uses it very little. So performance sucks on Maths heavy Apps.

What is a Maths heavy App? Bitcoin mining obviously. CAD? Graphics? Compression/Decompression? Being a hardware head, I really have no idea.

EDIT: I meant to say, I have a 'hard float RazPi 4 so I can do this stuff better, because it has an OS compiled using the full FPU

Have you heard about Mandelbrot? Or about mathematical models of nuclear physics? Brute force algorithm(s)? Solving differential equations?
Or just calculating a few million digits of pi (or e or any other similar number). Face recognition? Fast Fourier transformation?
Just to name a few.

Anything that runs on the GPU is both "maths-heavy" and massively parallel.

Also: I'm not sure if you were asking how to exercise an FPU, but if you were, then my first suggestion would be a polygonal DOS game (flight simulator or Quake) running in FreeDOS or DOSEMU.

To name just a few:
MatLab (from the MathWorks) or its open source equivalent GNU Octave (www.gnu.org/software/octave/)
Formula manipulation programs like Mathematica or Maple, a open source implementation is maxima (sourceforge.net/projects/maxima/)
Fast Fourier Transform like fftw (www.fftw.org/)
All kinds of heavy statistical calculation works, like the ones in the R package (www.r-project.org)

Thanks for the replies, guys. You're spot on, but most of those apps have the obvious name give away. Even if the calculation time is going to be increased by 10, the big delay is (Like scientific calculators) is going to be on input, isn't it?

Anyone trying to run games on an Arm cpu is in for a rough ride indeed. Quake? Doom?


What I was hoping for was things not so obvious. I've used Matlab in University - the time went into I/O & rebooting windows. It was a "Knowledge Bulimia subject in my Electronice degree

Knowledge Bulimia = Cram it for the test & forget it after .

To refine my question: what ordinary processes are maths intensive? We've got Graphics; There's not going to be Cad, Matlab or any emulator at all. No VMs. Arm software is hard and slow to build. A distro kernel could take 2 hours.But I want to know where I will be bitten in the rear end unexpectedly. Image conversion, or decompression, perhaps?

?
in every case the calculation will depend on the input. There is no "ordinary" process (what should it mean?) and no process will use any resource[s] by itself. Including compilers, games, conversion, decompression or whatever you could imagine.

I was thinking of Something like jpeg to png conversion, zooming stills. You tell me basically none of them are Maths Heavy. That's my answer, and thank you. I just didn't know that, as I rarely read code unless I want a headache.If I knew some process were notably better with an FPU, I'd avoid 32bit apps using them. Most Arm OSes are 32bit. When you look for 64bit, I've found 64bit DIY stuff, One complete one (in Slackware), and Fedora with serious dev work pointed in the totally wrong direction for me.

DOOM doesn't use the FPU at all.

Really? I thought all that gpu stuff would have some maths content. Shows you what I know about software.

I'm no gamer. Anyhow, I think only losers would select a lightweight cpu/gpu with only 4G of ram for games