| Apollo Hardfpu Results | page 1 2 3 4 5 6 7 8
|
|---|
|
|---|
| | 
Gunnar von Boehn
(Apollo Team Member)
Posts 6219
07 Oct 2017 09:56
| Lorenzo Pistone wrote:
|
There is a program to test the real power of the cpu/fpu
|
NO, not on AMIGA.
MINIBENCH is very good for measuring CPU features and strength in detail.
You have to understand that the old 68K FPUs are old and very very slow compared to a modern designs.
APOLLO FPU is a modern design and capable to reach speed 10 times more than 68060 - if proper code is used.
All AMIGA software is not able to fully utilize a good FPU.
This is like cars and streets.
You can have a race car - but to fully reach its potential you also need a good street - running the race car on a potato acre - will not reach its full potential.
So frankly all those FPU screenshots that we posted are results of a race car - running on potato acres.
| |
| | 
Kolbjørn Barmen
(Needs Verification)
Posts 219/ 2
07 Oct 2017 10:16
| Asaf Ayoub wrote:
| | Some people use this FPU test, it is part of HPC Challenge Benchmark.
wget EXTERNAL LINK mv linpackc.new linpack.c
gcc linpack.c -o linpack -lm
./linpack
EXTERNAL LINK |
Right, but there is no hand crafted 68k assembler in there, you cannot expect test binaries produced from sloppy gcc to make any meaningful numbers.
| |
| | 
Michal Warzecha
Posts 209
07 Oct 2017 10:32
| Because next step is to change this rotten potato acres to race road. Amiga world try to do this since many years. NG was try of it. We all know how it's looking today. Community is splited, even NG part of it. Apollo is attempt to back to the roots with another way to go forward. And I'm glad all about this. We cannot ask team do to that much work only for possibility to use some old benchamrks and close road to new modern things.
| |
| |
Kolbjørn Barmen
(Needs Verification)
Posts 219/ 2
07 Oct 2017 10:50
| Michal Warzecha wrote:
|
Because next step is to change this rotten potato acres to race road.
|
That cannot be done using Amiga OS, and there are way more shiny race roads elsewhere. AROS you say? Yes, but then there are way faster systems on which AROS runs much better. But no compatibility with Amiga software you say? Well, that is the damn potato acre.
| |
| |
Gunnar von Boehn
(Apollo Team Member)
Posts 6219
07 Oct 2017 10:55
| Kolbjørn Barmen wrote:
|
Michal Warzecha wrote:
|
Because next step is to change this rotten potato acres to race road.
|
That cannot be done using Amiga OS, and there are way more shiny race roads elsewhere. AROS you say? Yes, but then there are way faster systems on which AROS runs much better. But no compatibility with Amiga software you say? Well, that is the damn potato acre.
|
What you say is NOT true.
Of course one could write excellent FPU code on AMIGA OS It was just not done yet.
Also the benefit of doing it was not yet there - as no 68K did had a good FPU before.
| |
| | 
Mr Niding
Posts 459
07 Oct 2017 10:55
| Kolla, I do love your avatar :)
| |
| |
Kolbjørn Barmen
(Needs Verification)
Posts 219/ 2
07 Oct 2017 11:01
| Gunnar von Boehn wrote:
|
Also the benefit of doing it was not yet there - as no 68K did had a good FPU before.
|
No good MMU, no good FPU, no good this, no good that... what was good about 68k at all?
| |
| | 
Nixus Minimax
Posts 416
07 Oct 2017 11:05
| Kolbjørn Barmen wrote:
|
No good MMU, no good FPU, no good this, no good that... what was good about 68k at all?
|
The Instruction Set Architecture. Motorola always sucked at making good CPUs.
| |
| |
Gunnar von Boehn
(Apollo Team Member)
Posts 6219
07 Oct 2017 11:08
| Maybe it makes sense to explain this more clearly. Old FPUs needed very many cycles to execute 1 FPU instruction. Lets make an example:
Lets say you want to multiply a number of vector with 1 scalar So your FPU code would look like this:
lea array,A0
fmove scale,Fp0
moveq #100-1,D0 -- 100 vector to process
LOOP:
FMOVE.D (A0),Fp1
FMUL Fp0,Fp1
FMOVE.D Fp1,(A0)+
DBRA D0,LOOP
This code will execute 3 FPU instruction per data element to process.
These 3 operations are depending.
Each of them needs to wait for the result and finish of the previous
instruction. Old FPU needed several clock for each FPU instruction - and could only do 1 FPU instruction at the same time. Modern high end FPU can process several FPU instructions in parallel. As long as those instructions are INDEPENDANT. This means if we rewrite the code like this
lea array,A0
fmove scale,Fp0
moveq #100/50-1,D0 -- 100 vector to process
LOOP:
FMOVE.D (A0),Fp1
FMOVE.D 8(A0),Fp2 FMUL Fp0,Fp1
FMUL Fp0,Fp2
FMOVE.D Fp1,(A0)+
FMOVE.D Fp2,(A0)+
DBRA D0,LOOP
We now enable a much higher paralism.
This is how code is written since 15 years.
If you write for modern CPUs then you write the code to allow parallel execution - and reach this way several times higher performance.
| |
| | 
Thierry Atheist
Posts 644
07 Oct 2017 11:14
| Kolbjørn Barmen wrote:
| | That cannot be done using Amiga OS, and there are way more shiny race roads elsewhere. AROS you say? Yes, but then there are way faster systems on which AROS runs much better. But no compatibility with Amiga software you say? Well, that is the damn potato acre. |
Who's with me and Kolbjørn?Let's give up on this fiasco and get a cheapo raspberry PiIi!!
| |
| | 
Peeri the Sunlight
Posts 71
07 Oct 2017 11:15
| Gunnar von Boehn wrote:
|
moveq #100/50-1,D0 -- 100 vector to process
|
Should it be instead
moveq #100/2-1,D0
| |
| |
Kolbjørn Barmen
(Needs Verification)
Posts 219/ 2
07 Oct 2017 11:16
| Gunnar von Boehn wrote:
|
If you write for modern CPUs then you write the code to allow parallel execution - and reach this way several times higher performance.
|
Right, but how does it help old Amiga software?
| |
| |
Gunnar von Boehn
(Apollo Team Member)
Posts 6219
07 Oct 2017 11:18
| Kolbjørn Barmen wrote:
|
Gunnar von Boehn wrote:
|
Also the benefit of doing it was not yet there - as no 68K did had a good FPU before.
|
No good MMU, no good FPU, no good this, no good that... what was good about 68k at all?
|
The 68K FPU were good at its time. 30 years ago.
So for example the 68040 FPU was good compared to an x486.
But many years has passed and CPU development made many inventions since then. So the 68060 FPU is not as good as the Pentium 1 FPU.
And the 68060 FPU is weak compared to more modern design. The 68080 does fix this now.
| |
| |
Gunnar von Boehn
(Apollo Team Member)
Posts 6219
07 Oct 2017 11:22
| Kolbjørn Barmen wrote:
|
Gunnar von Boehn wrote:
|
If you write for modern CPUs then you write the code to allow parallel execution - and reach this way several times higher performance.
|
Right, but how does it help old Amiga software?
|
BRAVO!
Finally you understood it!
FACT: The 68080 FPU is MUCH more better than the 68060 FPU
FACT: old AMIGA FPU software is not able to utilize this advantage.
You see this in all the benchmark screenshots that we printed.
68080 scores little bit over 68060@80MHz FPU speed in SYSINFO and SYSSPEED scores.
But with proper software 68080 could score like an 68060@500 MHz
| |
| |
Kolbjørn Barmen
(Needs Verification)
Posts 219/ 2
07 Oct 2017 11:27
| Everything about Amiga is weak compared to modern design, it was a product of its time - 30 years ago. Keeping the silly car analogy going, building a modern engine into an old car does not make it a better car. Just faster, and a heck more dangerous to drive.
| |
| |
Mr Niding
Posts 459
07 Oct 2017 11:27
| Basically, how I read it; Apollo core will boost old program execution somewhat, but not to its max potential unless said programs are patched. So you get a quicker cpu, fpu and memory, but unless developers program for it spesifically, there are severe limits to the performance advantage. Thats understandable.
| |
| |
Kolbjørn Barmen
(Needs Verification)
Posts 219/ 2
07 Oct 2017 11:28
| Gunnar von Boehn wrote:
|
But with proper software 68080 could score like an 68060@500 MHz
|
Right, so all this is just "MUCH BETTER" nonsense is just... masturbation.
| |
| |
Mr Niding
Posts 459
07 Oct 2017 11:32
| Kolbjørn Barmen wrote:
|
Gunnar von Boehn wrote:
|
But with proper software 68080 could score like an 68060@500 MHz
|
Right, so all this is just "MUCH BETTER" nonsense is just... masturbation.
|
You couldnt keep it civil. Good way of living up to your avatar.
| |
| |
Thierry Atheist
Posts 644
07 Oct 2017 11:35
| Mr Niding wrote:
| | You couldnt keep it civil. Good way of living up to your avatar. |
+1(I lied in my previous post.)
| |
| |
Kolbjørn Barmen
(Needs Verification)
Posts 219/ 2
07 Oct 2017 11:38
| Mr Niding wrote:
|
You couldnt keep it civil. Good way of living up to your avatar.
|
Keeping it civic was exactly what I was doing, there are other ways to describe it that are not considered civic. And the avatar is something Gunnar has put on me, I would have chosen something from Kittelsen, not this weird thing.
| |
|
Performance and Benchmark Results!