@recedent

Quote:

aFlopsWOS: Amiga Flops Benchmark (Double Precision), MFLOPS(1) = 283.2340 MFLOPS(2) = 205.4596 MFLOPS(3) = 331.6121 MFLOPS(4) = 466.1064

What do these results mean?
How do you read this?

Does this mean 466 MFlops?

@Gunnar

Honestly wasnt the problem here what it means exactly...
The important thing was to have for everyone the same metric compaarison value...
So mflop tflop or picoflops dont cares...
The important thing was to use for eveyone the same bench tool, so to have comparable values.
No matter if wasnt good values for all the world...

_________________
Simone"Tuxedo"Monsignori, Perugia, ITALY.

@Tuxedo

Maybe because there are only a few beta testers who own one.

also it does say "2018 Edition!" perhaps start new "2024 Edition"?

Last edited by NutsAboutAmiga on 10-Mar-2024 at 10:16 AM.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

@Tuxedo

But how do you read it?

Does this means 466 MFlops?
Or what?

Because if it means this ...
then it makes no sense ...

As the G4 values should be 3 times this number..
So I wonder what the bench measures of if the bench is broken?

@Tuxedo

Must be careful with tools that’s not native WarpOS, is perhaps PowerPC, but it can be misaligned.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

@Gunnar

I honestly only wanted to compare results made with same program on different systems, no matter the number itself, but the result obtained on different systems that maybe comparable.

_________________
Simone"Tuxedo"Monsignori, Perugia, ITALY.

@Gunnar

Can be many things, EClock was changed in AmigaOS4.1. it tricks too fast compared to Amiga1200.

Last edited by NutsAboutAmiga on 10-Mar-2024 at 10:28 AM.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

@NutsAboutAmiga

Quote:

Maybe because there are only a few beta testers who own one.

No problem here...no hurry...

Quote:

also it does say "2018 Edition!" perhaps start new "2024 Edition"?

no have time to create a new post at all...and here we have already some values.

_________________
Simone"Tuxedo"Monsignori, Perugia, ITALY.

@ALL

BTW....
That wasnt a thread to create flame or whaterver...
Just try to use the tool that can be used and see what happens...

For example simply use only tools (like lame for example) that have a native version and compare them...possibly with same version

_________________
Simone"Tuxedo"Monsignori, Perugia, ITALY.

@Tuxedo

Quote:

For example simply use only tools (like lame for example) that have a native version and compare them...possibly with same version

This makes a lot sense..

Of course as long the tools not broken by design.

@NutsAboutAmiga

Quote:

Can be many things, EClock was changed in AmigaOS4.1. it tricks too fast compared to Amiga1200.

Is the source of the bench anywhere so download?

Maybe we can explain it then?

@Gunnar

No idea, the decompiles has improved I know, it perhaps get fixed, buy someone with the skills.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

@Tuxedo

Is this the source?

https://github.com/AMDmi3/flops

@Gunnar

I always wanted something like MacOS7.x.x speedometer for AmigaOS, yep its pretty simple tool, but tells a lot about the difference aspect of system disk, io, math, etc..
the best thing about it is that easy to compare different systems. But I’m pretty sure its 680x0 only, so does give some misinformation about PowerPC machines, what’s good and what’s bad. The information is only relevant if a MacOS 68K program running on the hardware nothing else.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

@NutsAboutAmiga

Yes I know what you mean.


I think measuring performance is very difficult.

A CPU has many different instructions and a coder has often many different ways to solve a problem.

The FPU benchmark code for example ... is not well coded.
FPU instructions have latency and the IBM and Motorola programmers manual explain you very clearly how to write reasonable FPU code. Looking at the source, this test is not written how coders should write FPU code.
This explains also the terrible score.

If you goal is to measure how fast the CPU can go with very bad written code...
Then the test does a perfect job.

It depends on what you want to measure.
If you want to measure how fast the CPU could go with good code, then you need to use another benchmark.

@Gunnar

dunno the source, I only downloaded the aMiga-like versions ready of the tools, maybe in the readme files was written...

_________________
Simone"Tuxedo"Monsignori, Perugia, ITALY.

@Gunnar

Quote:

If you goal is to measure how fast the CPU can go with very bad written code...

That's the typical real life scenario.

_________________
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC

@kolla

Gunnar:
Quote:

If you goal is to measure how fast the CPU can go with very bad written code...

Kola:
Quote:

That's the typical real life scenario.

Maybe yes, maybe no.

I agree that some coder write bad code.

But with FPU code in general you very often have workloads where you have to bulk process a lot of data.
E.g. do matrix transformation over thousands of vertices for a 3D program.
These FPU workloads can often be written very good and optimal.
Also C Compiler like GCC can on do on such code pretty decent results.
And "good" CPU will allow to several FPU instruction parallel.

Now this benchmark code look relative bad written,
and the way its written will nullify many good features in good CPUs.

So not sure if its useful to run it.


When we you look at cars then typically you say that a good driver with free street can get this car can do max 280 KMH.

What value would have a sentence like:

This Porsche with a bad driver with some traffic jam might reach 50 KMH ?