Overview Features Coding ApolloOS Performance Forum Downloads Products Order Contact

Welcome to the Apollo Forum

This forum is for people interested in the APOLLO CPU.
Please read the forum usage manual.
Please visit our Apollo-Discord Server for support.



All TopicsNewsPerformanceGamesDemosApolloVampireAROSWorkbenchATARIReleases
Performance and Benchmark Results!

Vampire2 80MHz Outclasses the AMIGAONE 800MHzpage  1 2 3 4 

Daytona 675x

Posts 32
11 Sep 2017 10:12


@Gunnar

Please.

I did not handwrite / handtune anything (if I did then the other parts of the benchmark would look quite different too...).

As I said I recompiled the benchmark using a recent gcc, that's it.
If you used an outdated gcc for your Vampire test, then I suggest you use a new one and post the results.

Why? Because it's about measuring copy-speed. Obviously your old build doesn't do that. It measures bad compiler speed and not the capabilities of the machines.

Your complain is therefore not valid.

What about the other issues? Like the missing / wrong sortbench stats?



Saladriel Amrael

Posts 166
11 Sep 2017 10:12


Of course comparing benchmarks with the same build compiled with the same compiler version is important.

Of course benchmarks can give us a guideline of the "horsepower" of a given system and allow us to compare it to results obtained on other systems to let us understand what this could mean.

In this reguards I've learned to not take benchmarks too seriously, and always with a pinch of salt. All I know now is that Vampire is more or less on the same scale with mid/range PPC Amigas and that's gret and far more than I did expect.

How well it will perform in real life, I will be able to tell when I'll have the board in my hands.

Keep on going!




Gunnar von Boehn
(Apollo Team Member)
Posts 6207
11 Sep 2017 10:18


Daytona 675x wrote:

Your complain is therefore not valid.

 
The bad memory performance of the AMIGA-ONE is not an invention of us.
Its sad but the bad memory performance of the AMIGA-ONE is a technical fact.
 
Please try your "new" compiles on an AMIGA-ONE with G3-PowerPC or with G4-PowerPC, yourself.
And please come back with your scores after doing this.
 
You will see yourself, that the memory performance of the AMIGA-ONE will always be lower than the VAMPIRE.
 
Its a technical fact that the AMIGA-ONE, can by hardware design, not reach the memory speed of the VAMPIRE.
The AMIGA-ONE is very limited by its north-bridge.



Vojin Vidanovic

Posts 770
11 Sep 2017 10:28


Gunnar von Boehn wrote:

  As its a technical fact that the AMIGA-ONE can by hardware design not reach the memory speed of the VAMPIRE.

Short story: Great archievement, keep improving other stuff. Currently SAGA over DIGITAL-VIDEO and full AGA implementation with sound chip and SDBoot seems like "wanted feature".

Long story: I am glad about Vampire memory performance, and an aspect where it
outclassess SAM440/460 and G3 and G4. Seems to me G5, PA Semi and similar designs (PS3 CPU ...) have improved memory controllers you are perforance wise looking towards. Tabor,  and even 5020/5040 does not seem to shine there even using DDR3, but what kills PA Semi is poor MIPS/Mhz, not memory, SATA or even FPU speed - G4 CPU is faster there and gives overally nicer taste and feel. And I am glad Vamps are good there.


Daytona 675x

Posts 32
11 Sep 2017 10:29


@Gunnar

Again: I suggest you use fresh benchmark compiles to give ALL systems a fair chance to really show us their TRUE memory performance.
As I pointed out this old stream build you used simply does not tell us correct data.

So you based your claims earlier in this thread on FALSE memory performance numbers of stream, period.

And again: what about the other issues? Like the missing / wrong sortbench stats?

It's not my intention to piss on your parade, as I said: I am a Vampire fan and love to see it progressing.
It's my intention that you make fair benchmarks and post ALL result though, not just the ones in Vampire's favor.

Is that too much to ask for or hard to understand?


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
11 Sep 2017 10:37


Daytona 675x wrote:

@Gunnar
 
Again: I suggest you use fresh benchmark compiles

Yes, please go ahead and do this. :-)
Run your compile on an AMIGA-ONE.

And maybe you can also run BUSTEST on the PPC-AMIGAs for comparison.
BUSTEST on AMINET EXTERNAL LINK  The Vampire reaches 600MB/sec in ">BUSTEST FAST"

Lets take this with some sporting spirit and have a little fun bet.
I'll bet that you will not be able to beat
-  the STREAM memcopy speed of the Vampire - with any G3 or G4 PowerPC amiga.
- and not be able to beat the BUSTEST result of the Vampire with any
  G3 or G4 PowerPC amiga.

Are you accepting the challenge?  ;-)


Daytona 675x

Posts 32
11 Sep 2017 10:40


As I said: I have no such AmigaOne.
  But you apparently have, I mean, after all you ran the test on it, right? In my initial post you find a link to the fresh build.
  Please run it and tell us.
 
  This is no competition btw.
  You see, I think this is the problem: YOU seem to think here's some sort of competition running. It is not.
  I only want correct and complete numbers, not some sort of childish challenge.

And again: what about the other issues? Like the missing / wrong sortbench stats?


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
11 Sep 2017 10:44


Daytona 675x wrote:

  As I said: I have no such AmigaOne.
    But you apparently have, I mean, after all you ran the test on it, right?
 

 
Nope, I did not run this - we quoted the scores by the OS4 devs.
As clearly stated before: the AMIGA-ONE scores are the official scores from the OS4 guys.
 
If you think the OS4 devs are not able to benchmark their own machines - talk to them directly. ;-)
 
But in anyway the AMIGA-ONE can by hardware design not beat the VAMPIRE.
This is obvious. Ask the OS4 devs if you not believe me.


Daytona 675x

Posts 32
11 Sep 2017 10:48


"The AMIGA-ONE scores are the official scores from the OS4 guys."
 
That's wrong.
Apparently you simply posted the results of the guy who compiled that stream-exe back in 2005 then. Those are no "official scores".

But anyway: it doesn't make scores calculated by an exe that doesn't  truely exploit the hardware any better...
 
So, can anyone here with such an AmigaOne 800MHz please download my zip and please run the tests?
 
  Btw.: What about the other issues? Like the missing / wrong sortbench stats?


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
11 Sep 2017 10:57


Daytona 675x wrote:

But anyway: it doesn't make scores calculated by an exe that doesn't  truely exploit the hardware any better...

 
You are funny. :-)
 
Rethink the facts.
The 80 MHZ 68080 does this benchmark 2 times faster than the 800 MHz PowerPC.
 
The PPC can have 5 Gigaherz - it does not matter.
The AMIGA-ONE will always loose again the VAMPIRE.
The reason should be well known, the AMIGA ONE is crippled by its memory controller.
 
The AMIGA ONE can by hardware design never match the memory speed of the VAMPIRE. You can write as many DCBT as you want - they will not change this.
 

But the SAM does not have this crippling Northbridge.
So lets widen the challenge to be more fun - shall we :-)
Can you show us the BUSTEST scores the SAM can reach?
 


Daytona 675x

Posts 32
11 Sep 2017 11:02


Ah, I actually got an G4, a PowerMac 3,5 G4 733 MHz with MorphOS to be exakt.
  Here's the results:
 
  -------------------------------------------------------------
  Function      Rate (MB/s)  Avg time    Min time    Max time
  Copy:        407.5924      0.0790      0.0785      0.0798
  Scale:        413.6547      0.0780      0.0774      0.0801
  Add:          472.6132      0.1021      0.1016      0.1031
  Triad:        473.2230      0.1020      0.1014      0.1024
  -------------------------------------------------------------
  Solution Validates
  -------------------------------------------------------------
 
Again: I have no such machine and so far you only presented numbers using a compile that we definitely know to create massively false numbers.
I don't see what's particularly funny here, maybe your sense of humor is different to mine.

I suggest to simply wait for somebody with such an AONE 800MHz to do proper test?

Until then you can finally answer why you only publish partial / wrong sortbench stats.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
11 Sep 2017 11:04


Daytona 675x wrote:

  Ah, I actually got an G4, a PowerMac 3,5 G4 733 MHz with MorphOS to be exakt.
  Here's the results:
  -------------------------------------------------------------
  Function      Rate (MB/s)  Avg time    Min time    Max time
  Copy:        407.5924      0.0790      0.0785      0.0798
 

 
And this shows that old MAC Hardware is better than AMIGA ONE hardware.
This is No news.
 


Daytona 675x

Posts 32
11 Sep 2017 11:11


Correct, and it is also no news that the AONE has a bad mem-controller.

But the question is: is it as bad as your numbers suggest?

And I simply say: it may well be that its performance is actually much better because of the wrong measurement you made (and I asume even you agree that the old exe did not measure real copy speed, do you?)

Your post overlapped an edit of mine, I'll repeat the added part here:

I suggest to simply wait for somebody with such an AONE 800MHz to do a proper test?
Until then you can finally answer why you only publish partial / wrong sortbench stats.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
11 Sep 2017 11:14


Daytona 675x wrote:

Correct, and it is also no news that the AONE has a bad mem-controller.
 
But the question is: is it as bad as your numbers suggest?

I know the PPC memory interface by heart.
You seem to be not sure about it?
Do you dare to take a bet? ;-)

Can you run BUSTEST on your SAM?
EXTERNAL LINK


Daytona 675x

Posts 32
11 Sep 2017 11:17


Apparently you ARE kind of childish.
Again: I don't bet or play games here.
I just want to get correct tests from you.
(and why I should run an 68k bustest exe on a ppc and expect real meaningful results is beyond me since it'll first of all measure the 68k-emu speed :P )
 
  One bet I have for you though:
  I bet that you will never answer my question about why you post wrong / partial sortbench results, namely only those that are in favor or Vampire :)


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
11 Sep 2017 11:36


Daytona 675x wrote:

  I just want to get correct tests from you.
 

Then post correct scores.
  * Please post BUSTEST
  * Please post stream on AMIGA-ONE 750-PPC

You are twisting the facts.
You came here to claim that the scores posted by the OS4 guys are not correct.
Don't make false claim but prove this!
 

Also the SORT score were done correctly.
The scores were achieved, with the correct code being compiled and ran and the correct score was printed.
 
If you changed the code, changed the program, and then got slightly different scores than this does not invalidate the correct scores got before.


Daytona 675x

Posts 32
11 Sep 2017 11:42


Gunnar von Boehn wrote:

      The post correct scores.
      * Please post BUSTEST
   

   
    Please provide a PPC build of that, then I'll do so.
   
   
Gunnar von Boehn wrote:

    * Please post stream on AMIGA-ONE 750-PPC
   

   
    Once again: I have no such machine.
     
   
Gunnar von Boehn wrote:

    The SORT score were done correctly.
   

    Once again: I did just what you asked for in a1k and delivered correct results for the 440ep which you apparently ignored (and we talk about roughly 20% difference in the 32k test, not some "minor" amount...).
   
   
Gunnar von Boehn wrote:

    The scores were achieved with code was compiled and ran and the correct score was printed.
   

  My correctly produced numbers tell a different story. They tell pretty much the same story as with the stream-test:
  if you let a shitty ppc-compiler produce bad code then you get bad results. Such values are not reflecting the true power of a system, simple as that (needless to say I compiled it exactly as you told us to do in a1k, -O2).
I wonder why you apparently ignore better results and prefer the bad ones here.
   
   
Gunnar von Boehn wrote:

    If you changed the code, changed the program, and then got slightly different scores than this does not invalidate the correct scores got before.
   

    I did not change any code.
   
    And still no answer why the 1k-to-8k results of sortbench didn't make it to your performance comparison charts...
   
    So much about twisting the facts...
    All I delivered are valid numbers, period.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
11 Sep 2017 11:50



Daytona 675x wrote:

  And still no answer why the 1k-to-8k results of sortbench didn't make it to your performance comparison charts...

This is not a clever question.
Sortbench can provide thousands of different results from 1 to size of your memory.
People can easily read a chart with 1 line.
Drawing a chart with tens, or hundreds, or thousand lines is unreadable. So it makes sense to "pick" one value and choose it.

Daytona 675x wrote:

     
Gunnar von Boehn wrote:

      * Please post stream on AMIGA-ONE 750-PPC
     

     
      Once again: I have no such machine.
   

   
You came here and post claims that the scores of the OS4 guys about their own machine are wrong. ;-)
Why did you do this if you do not know it?
   
I think it makes good sense that you first verify your claim, don't you agree?
Please verify it, and please come back after doing so.


Daytona 675x

Posts 32
11 Sep 2017 12:10


Gunnar von Boehn wrote:

  So WHY do you come here and post claims that the scores of the OS4 guys about their own machine are wrong?

Sorry, but what's your problem?
Carefully read my initial post again...

I said that there are very strong indications that the ppc-test you ran (well, actually you didn't as it turned out) and on which you based your ppc-performance theories delivers FALSE numbers.

To show that FACT I delivered numbers of the machines I got using the old and freshly compiled exe (only modification: a decent compiler).

Since the results differ MASSIVELY "I suggest you make sure you actually measure somewhat correctly" before .

"You should probably remeasure the AONE results you posted before making toooooo strong claims about the speed of certain PPC systems here..."

Why do I do so?
Actually I also wrote so in my very first post:

"I'm a Vampire fan. I got one myself and I love it.
But I am no fan of "beautified", imcomplete or simply wrong benchmarks."

YOU are the one posting claims in the first place.
And your claims are based on an exe that is known to produce much too low numbers on similar systems. That's all there's to say actually.

All I ask for is that YOU (or somebody with such hardware) do some real tests with a more valid exe (which I provided) before claiming stuff.
And I ask for TRUE and complete benchmark results - and not some beautified stuff like the 32k stats you present.

Again: is that too much to ask for? Is that too hard to understand?

I understand that the Vampire is "your baby". And I also understand that you apparently feel easily offended or believe that you're being attacked if somebody post something that questions some data you provided.

Well, I understand that to a certain degree only though.

I don't understand why you react as you do and react as hostile as you do, even questioning my data and saying that I said this and that, which I didn't.

I wrote the following in my initial post too:

"The Vampire is such a great thing, you really shouldn't taint its reputation by "beautifying" benchmark-results or by trying to "bash" PPC (at least that's the impression one might get here very easily)."

Unfortunately you are doing exactly nothing to allay those concerns.


Hbarra 2Pi

Posts 8
11 Sep 2017 12:12


Hi Gunnar...

Pelease... Don't feed the 'Vampire fan' Troll!

All the fans love your competitive and protective spirit of the Apollo Team.

The talent and the good know how always provoke the envy in which it lacks of these gifts.

Mediocrity and ignorance are brave!

Do not enter the rag. Do not waste your time...

Excuse my poor English ;p

posts 70page  1 2 3 4