Ive not run any of these tests myself, so dont know what they are, but dont os4 and mos use ppc native code sometimes (libs, etc.) even when running 68k binaries ? This, plus the overhead of running (win)UAE and its OS make it tricky to compare the speed of the 68k cpu emulation technologies (plus the fact that UAE emulates entire system which causes the 68k side of things to sometimes wait because of chipset timing).
Just my 2 cents.

@fishy_fis

>Ive not run any of these tests myself, so dont know what they are, but dont os4 and >mos use ppc native code sometimes (libs, etc.) even when running 68k binaries ?

in real world that code is used for a audio denoiser in hd-rec.but to make speedtests easy Thilo have add in the lib timer code so there is no need to dowload hd-rec and do the test here.

But its a real world test.

Sure its possible to do it native, also it can do winuae/amithlon native but nobody translate the code in C and compile for PPC, so thats what we have to compare.

I have done btw other fft code for winuae native and test if its worth the work to do native code for winuae.only the float Version is usefull, the integer version have worse quality, because fft is a iterative algorithm the errors by unacuraty of the integer Version are heavly sum up.but the integer version is usefull to see, if it is on modern CPU usefull to do the extra work and code for integer.because image and audio processing use floats and use floats on program make coding more easy.

And because on PPC the float values are lower as integer performance let not hope for a speedup with native code, because the asm code the JIT produce have fewer instructions.

was only 60% faster even if the X86 code need no byteswap.but i use no ISSE code.fft can only 2* faster because its stereo, so when use ISSE can get double speed.

but when take a look at the values the code is fast enough on winuae Systems.

i find out some intresting.PPC is slow when use 32 bit float, faster on 64 bit when data is in cache

64 bit float use more cache mem and more mem transfer.so its possible that there is more slowdown then.

see here for some values of SAM aone and X86

http://utilitybase.com/forum/index.php?action=vthread&forum=16&topic=1876&page=1

I think the weak point wy PPC is so slow, is that 32 bit float data load is extrem slow on PPC

maybe a PPC 970 is here lots faster.

Last edited by bernd_afa on 14-Dec-2009 at 03:00 PM.
Last edited by bernd_afa on 14-Dec-2009 at 09:28 AM.
Last edited by bernd_afa on 14-Dec-2009 at 09:24 AM.
Last edited by bernd_afa on 14-Dec-2009 at 09:22 AM.
Last edited by bernd_afa on 14-Dec-2009 at 09:21 AM.
Last edited by bernd_afa on 14-Dec-2009 at 09:20 AM.
Last edited by bernd_afa on 14-Dec-2009 at 09:19 AM.

Dont forget WinUae emulates the custom chips not just the 68k - If its possible to have such fast 68k emulation on PPC why cant this be combined with UAE?

@pixie: >So you don't think that a cheap yet powerful machine is important into this kind of >reasoning?

Sure it´s important, too. But in fact we are talking about AmigaOS4 on x86 again and this is discussed to dead here already. As you can see with AROS (or ZETA or BeOS), x86-hardware alone is no guarantee to attract more users or developers (or companys) for a niche-OS.

And Bernd seems to live in a dreamworld, when he think, that there is so much more money for AOS4. AROS has a good Bounty-system and as much "marketing" as AOS4, when you check the latest news on Amiga-related sites (ICAROS, ARES One etc.). And I can´t see, that there is more bashing against AROS-developers. In fact there is much more bashing against Hyperion and AmigaOS4 (thanks people like Bernd).

Last edited by AP on 14-Dec-2009 at 11:00 AM.

_________________
AmigaOne X5000/40, 2.2 Ghz, 4 GB RAM, Radeon R9 280X, M-Audio Revolution 5.1, 240 GB SSD

@AP
>As you can see with AROS (or ZETA or BeOS), x86-hardware alone is no guarantee >to attract more users or developers (or companys) for a niche-OS.

you can just see that there are much less users that use PPC with this OS.Or do you think when Linux, Mac OS, Zeta, run only on SAM Hardware and cost so much as PPC AOS they get more users than currently ?

>But in fact we are talking about AmigaOS4 on x86 again

No, we are talking about that a OS to attract more users need also comparable Hardware in power and price to attract developers to do something for this system.
Or in other words wy a developer should develop for a Hardware that run so much slower than other comparable systems ?.when sombody develop a image processor or videocut system or other modern software, there is CPU Power need.
also compiling a big program need much CPU power.

When they maybe can sell complete SAM Computer system(Case) for 100 Eur thats a fair price and new users , or the childs of users that need computer only for internet skool. and mail need not invest much money and can maybe get happy with SAM for 100 Eur.

When somebody invest money for a OS he should look that he can build a attractive Hardware.

>In fact there is much more bashing against Hyperion and AmigaOS4 (thanks people >like Bernd).

bash is tell something bad without facts
and 99.99% what i say i can show with links.So i really do not bash.


>And Bernd seems to live in a dreamworld, when he think, that there is so much >more money for AOS4.

lets say OS4 and 4.1 and cybppc Version is sell 1000 times incl cybppc Version.this give with 100 Eur per OS4 100 000 Eur.We all know there are several users that buy OS4/4.1 for more than 1 OS4 system.

So what do you think, how many buy OS4.0/OS4.1 and OS4 cybppc ?

Now count the money for the AROS bounty, thats far not so much.

then you must also add the money for OS4 bounties etc.

>And I can´t see, that there is more bashing against AROS-developers

you should read AROS news threads then.should i really post links ?

Last edited by bernd_afa on 14-Dec-2009 at 03:00 PM.
Last edited by bernd_afa on 14-Dec-2009 at 02:59 PM.
Last edited by bernd_afa on 14-Dec-2009 at 02:58 PM.
Last edited by bernd_afa on 14-Dec-2009 at 02:53 PM.

@bernd_afa

Yeah, Bernd, give us some useless links, start a fruitless discussion, why something over there is AROS bashing in your honest opinion and transform it into a OS4 bashing discussion. Nothin´ new in 2009

Regards

@whose

ok,after your post, i just catch 1 from the next actual AROS news today, from a popular MOS user.You can click on profile and find his Homepage and what he do.

In short last sentence

"""
http://www.amiga-news.de/de/news/comments/243102.html
not yet stable enough.
""

His sentence help not to make AROS better, because he tell no reason wy AROS is unstable.Maybe he have a not good support Hardware, or he do something wrong.nobody can tell him if its really AROS or can tell him a trick how can AROS get better working.

So question is wy he invest time to write this ?

If this tell maybe a annonymous then more people think thats not true, but because a better known registered guy tell this, this maybe prevent some users to try out AROS and report Bugs to make AROS better.Should this the reason to write that ?

when i write about Bugs or unstable i tell always whats unstable or must enhance and post if possible links, so the Devs see what can it better.So i do not bash.

I think btw you and AP bash here against me without any Facts, because you name me offtopic as a basher and i think your posts are a valid abuse report

Last edited by bernd_afa on 22-Dec-2009 at 08:48 AM.
Last edited by bernd_afa on 22-Dec-2009 at 08:47 AM.
Last edited by bernd_afa on 22-Dec-2009 at 08:45 AM.

@bernd_afa
On the minimig it sadly just crashes with a ramlib program failed (error #80000003).

Probably as it only has a MC68SEC000 cpu and no fpu, but I thought it would be fun to run the test on my 4MB Minimig with ARM addon board as it the only non emulated 68k CPU Amiga system still in production today.

If the tests had run to completion in 28MHz turbo mode I would then have repeated them in the normal 7MHz non turbo mode just to let us see how far the performance of Amiga 68K Systems has increased since the first Amiga A1000s were the computer to have

Oh well I guess it a good thing the test do not run as it would probably have temp one of my Minimigs out of action for playing games for days

_________________
Sam440ep 667mhz, 512MB, 120GB 2.5" HD, OS4.1FE
WinUae 3.0.0, OS 3.9, BB3, Catweasel MkIV
Amiga 1200, Blizzard 040/40 (BlizzardPPC 060/200 with SCSI removed at present), mediatorSX pci, Voodoo3, PCI network card os 3.9BB2
4MB Minimig with ARM addon boar

@Dwyloc

see how fast minmig is, is intresting.I switch FPU usage in amiblitz compile off so all should work in software and use ffp single float math of AOS.but i am not sure if it really do work without FPU, because non of my installs boot when i switch off the FPU in winuae.

I upload it here

http://www.daten-transport.de/?id=UzWy6tpqM8td

SO i run the test on my winuae with switched on FPU.The first test is extrem slow, so it seem use ffp. you need wait very long.

Speed test for FFT + iFFT: (float)
time needed 14976ms for 413696 samples, => .3131964x speed @44100Hz/stereo, RTF=3.192883
Speed test for FFT + iFFT: (integer)
time needed 1056ms for 413696 samples, => 4.441695x speed @44100Hz/stereo, RTF=.2251392
Speed test for FFT + iFFT: (integer handoptimized 68K ASM)
time needed 530ms for 413696 samples, => 8.849868x speed @44100Hz/stereo, RTF=.1129961

15.h0:wbstartup>

Edit

Here is Version that is compile without 68020 (mul.l) instructions too

http://www.daten-transport.de/?id=Wf27nHJfrHvF

its still slower.maybe on the asm optimized version it crash, when use mul.l

Speed test for FFT + iFFT: (float)
time needed 20012ms for 413696 samples, => .2343809x speed @44100Hz/stereo, RTF=4.266559
Speed test for FFT + iFFT: (integer)
time needed 8304ms for 413696 samples, => .5648399x speed @44100Hz/stereo, RTF=1.770413
Speed test for FFT + iFFT: (integer handoptimized 68K ASM)
time needed 1182ms for 413696 samples, => 3.968215x speed @44100Hz/stereo, RTF=.2520024
1

Last edited by bernd_afa on 22-Dec-2009 at 07:11 PM.
Last edited by bernd_afa on 22-Dec-2009 at 07:10 PM.
Last edited by bernd_afa on 22-Dec-2009 at 07:10 PM.

@bernd_afa

...and as always, you justify the expectations on you.

Just keep your very own opinion about OS4 vs. AROS vs. MOS vs. anything out of it and make benchmark tests. That is all I wish for christmas. Thank you!

@bernd_afa
using AmiKit/winuae 2.0:
Speed test for FFT + iFFT: (float)
time needed 575ms for 413696 samples, => 8.15727138519287x speed @44100Hz/stereo
Speed test for FFT + iFFT: (integer)
time needed 609ms for 413696 samples, => 7.70185708999633x speed @44100Hz/stereo
Speed test for FFT + iFFT: (integer handoptimized 68K ASM)
time needed 284ms for 413696 samples, => 16.5156021118164x speed @44100Hz/stereo

@alanwall
>using AmiKit/winuae 2.0:
>Speed test for FFT + iFFT: (float)
>time needed 575ms for 413696 samples, => 8.15727138519287x speed >@44100Hz/stereo
>Speed test for FFT + iFFT: (integer)
>time needed 609ms for 413696 samples, => 7.70185708999633x speed >@44100Hz/stereo
>Speed test for FFT + iFFT: (integer handoptimized 68K ASM)
>time needed 284ms for 413696 samples, => 16.5156021118164x speed >@44100Hz/stereo

What hardware is this ?

@whose
>Just keep your very own opinion about OS4 vs. AROS vs. MOS vs. anything out of it >and make benchmark tests.

I dont accept that guys write wrong, when AP and you not write wrong, then i need not post some Posts.
I bring the facts so everybody can verify it and can see the reality and can get more information.

i only answer on your posts because i dont want that you get the happiness feeling that you win the attack.

My wish for christmas and new year is that the red versus blue war and the other attacks end and all can work together same as on Linux World, and nobody make announces that arent reality or are far a away from get reality, and not suggest the system have better future with announcements

Last edited by bernd_afa on 23-Dec-2009 at 08:20 AM.
Last edited by bernd_afa on 23-Dec-2009 at 08:19 AM.
Last edited by bernd_afa on 23-Dec-2009 at 08:16 AM.

@bernd_afa

I am sorry to say that both of your new executables still crash in the same way, so as of yet no speed results for the 68000 chip in the minimig.

_________________
Sam440ep 667mhz, 512MB, 120GB 2.5" HD, OS4.1FE
WinUae 3.0.0, OS 3.9, BB3, Catweasel MkIV
Amiga 1200, Blizzard 040/40 (BlizzardPPC 060/200 with SCSI removed at present), mediatorSX pci, Voodoo3, PCI network card os 3.9BB2
4MB Minimig with ARM addon boar

@bernd_afa

here is results from my improved Blizzard PPC. i did say im not happy with the results and im still not happy it needs to inprove even if it has set a new world record for CPU & bus speed on classic amiga.

Blizzard PPC 322Mhz 80.666Mhz bus OS4.0


Speed test for FFT + iFFT: (float)
time needed 17588ms for 413696 samples, => .26668357849121x speed @44100Hz/stereo
Speed test for FFT + iFFT: (integer)
time needed 15123ms for 413696 samples, => .310152143239975x speed @44100Hz/stereo
Speed test for FFT + iFFT: (integer handoptimized 68K ASM)
time needed 12575ms for 413696 samples, => .372996479272842x speed @44100Hz/stereo

Last edited by delshay on 27-Aug-2010 at 05:44 PM.
Last edited by delshay on 11-Jan-2010 at 01:08 PM.
Last edited by delshay on 11-Jan-2010 at 12:52 PM.
Last edited by delshay on 11-Jan-2010 at 12:51 PM.
Last edited by delshay on 11-Jan-2010 at 12:40 PM.

_________________
The Machine: Bride Of The Pin•Bot by Williams Electronics

@delshay

your values are only 2% faster.thats maybe measurement accuracy.I think this test is only exact about 5% because the Test do not disable multitasking.

I get when repeat the test several times around 1-2% diffrent values too.

Last edited by bernd_afa on 11-Jan-2010 at 02:07 PM.

@bernd_afa

NEW UPDATE

Blizzard PPC OS4.O 334Mhz


Speed test for FFT + iFFT: (float)
time needed 15359ms for 413696 samples, => .305386483669281x speed @44100Hz/stereo
Speed test for FFT + iFFT: (integer)
time needed 12910ms for 413696 samples, => .363317638635635x speed @44100Hz/stereo
Speed test for FFT + iFFT: (integer handoptimized 68K ASM)
time needed 10423ms for 413696 samples, => .450007766485214x speed @44100Hz/stereo

Last edited by delshay on 26-Aug-2010 at 07:26 PM.
Last edited by delshay on 26-Aug-2010 at 07:17 PM.
Last edited by delshay on 06-Feb-2010 at 12:36 PM.

_________________
The Machine: Bride Of The Pin•Bot by Williams Electronics

@thread

Cyberstorm PPC @ 366MHz, 66.7MHz bus speed, 70ns ram, OS4:

FFTDemo
Speed test for FFT + iFFT: (float)
time needed 8362ms for 413696 samples, => .560922145843505x speed @44100Hz/stereo
Speed test for FFT + iFFT: (integer)
time needed 7154ms for 413696 samples, => .655637502670288x speed @44100Hz/stereo
Speed test for FFT + iFFT: (integer handoptimized 68K ASM)
time needed 3357ms for 413696 samples, => 1.39720904827117x speed @44100Hz/stereo

_________________
AmigaOS 4.x Beta Tester - Classic Amiga enthusiast - http://www.hd-zone.com is my Amiga Blog, check it out!

@Thread

Sam 440-EP Flex 800Mhz OS4.1U2:

FFTDemo
Speed test for FFT + iFFT: (float)
time needed 5614ms for 413696 samples, => .835488200187683x speed @44100Hz/stereo
Speed test for FFT + iFFT: (integer)
time needed 4205ms for 413696 samples, => 1.11544132232666x speed @44100Hz/stereo
Speed test for FFT + iFFT: (integer handoptimized 68K ASM)
time needed 2308ms for 413696 samples, => 2.03224897384643x speed @44100Hz/stereo

Last edited by Spectre660 on 25-Aug-2010 at 11:13 PM.

_________________
Sam460ex : Radeon Rx550 Single slot Video Card : SIL3112 SATA card

oh a 366 MHZ cybppc.this is only 30% slower as a SAM 667 MHZ in test

intresting that cyberstorm PPC can run with that high clockrate.

thanks for the results, i merge your resuls in the intro Post

@bernd_afa

Quote:

oh a 366 MHZ cybppc.this is only 30% slower as a SAM 667 MHZ in test

That says something about this test...

(and Spectre660´s SAM has 800 MHz)