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!

Can You Beat This Score?page  1 2 3 

Gunnar von Boehn
(Apollo Team Member)
Posts 6207
08 Nov 2016 20:51


Here some speed result of the popular SORTBENCH.

EXTERNAL LINK 
As you see the result of GOLD2 rc is very good.

you can find the source of the benchmark here:
CLICK HERE 



Claudio Guglielmotti
(Apollo Team Member)
Posts 185
08 Nov 2016 21:04


Yes i could, with my old dev board


Gregthe Canuck

Posts 274
08 Nov 2016 21:29


Very nice indeed. :)
 
Close to a Cortex A4 @ 800MHz!


Kolbjørn Barmen
(Needs Verification)
Posts 219/ 2
09 Nov 2016 01:26


raspberry pi zero:
 
  EXTERNAL LINK


Henryk Richter
(Apollo Team Member)
Posts 128/ 1
09 Nov 2016 05:53


Out of curiosity, I asked my venerable home server for some numbers:

EXTERNAL LINK 
But more importantly, said machine just passed the 1000 days uptime mark. 1002 days uptime, for a 15yr old computer. Current models from the rotten fruit company won't even last that long in many cases.


Markus (mfro)

Posts 99
09 Nov 2016 07:01


Although I'd consider this more like a compiler test (where the code appears to be tweaked to suit specific compiler needs already) than a CPU benchmark, here are the FireBee results:
 
 

  firebee:~#./sortbench
  -------------------------------------------------------------
  SORTBENCH 1.1 (Gunnar von Boehn)
  Its a CPU benchmark that stresses CPU, DCache and branch prediction.
  -------------------------------------------------------------
    1 K Element :  399.02 MB/sec
    2 K Element :  400.39 MB/sec
    4 K Element :  400.69 MB/sec
    8 K Element :  399.01 MB/sec
  16 K Element :  160.13 MB/sec
  32 K Element :  109.73 MB/sec
  firebee:~#cat /kern/cpuinfo
  CPU:  ColdFire V4e
  MMU:  ColdFire V4e
  FPU:  ColdFire V4e
  Clocking: 263.7MHz
  BogoMIPS: 263.78
  Calibration: 131891200 loops
  firebee:~#
 

 
  Note that I took the freedom to add one single word to the sources to suit my compiler's needs as well (and no, it's not "inline").


Mo Retro

Posts 241
09 Nov 2016 07:27


Kolbjørn Barmen wrote:

The 5 dollar raspberry pi zero:
 
  EXTERNAL LINK 

Wow these RPI zero scores are factor 4 compared to the Apollo with Gold core 2. What speed does the Rpi0 run?


Mr-Z EdgeOfPanic

Posts 189
09 Nov 2016 07:51


Mo Retro wrote:

Kolbjørn Barmen wrote:

  The 5 dollar raspberry pi zero:
 
  EXTERNAL LINK 

  Wow these RPI zero scores are factor 4 compared to the Apollo with Gold core 2. What speed does the Rpi0 run?

1 Ghz single core proc
So if Apollo was running on the same clock freq, it would beat it hands down.


Markus (mfro)

Posts 99
09 Nov 2016 07:55


An interesting oddity: on x86, with a modern compiler (gcc 6.2), -O1 gives way better results than -O2:
 
  -O2:
 

  -------------------------------------------------------------
  SORTBENCH 1.1 (Gunnar von Boehn)
  Its a CPU benchmark that stresses CPU, DCache and branch prediction.
  -------------------------------------------------------------
  1 K Element :  9448.20 MB/sec
  2 K Element :  12285.07 MB/sec
  4 K Element :  12441.59 MB/sec
  8 K Element :  12459.49 MB/sec
  16 K Element :  12378.39 MB/sec
  32 K Element :  12474.50 MB/sec
 

 
  -O1:
 
 

  -------------------------------------------------------------
  SORTBENCH 1.1 (Gunnar von Boehn)
  Its a CPU benchmark that stresses CPU, DCache and branch prediction.
  -------------------------------------------------------------
  1 K Element :  15019.83 MB/sec
  2 K Element :  16179.41 MB/sec
  4 K Element :  16305.13 MB/sec
  8 K Element :  16517.26 MB/sec
  16 K Element :  16526.07 MB/sec
  32 K Element :  16556.53 MB/sec
 

It would be interesting to see if this oddity is also present on ARM and PPC platforms (which I assume to have modern "smartass" compilers as well).


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
09 Nov 2016 09:54


Markus (mfro) wrote:

Note that I took the freedom to add one single word to the sources to suit my compiler's needs as well (and no, it's not "inline").

What did you change?

Can you post the ASM of the function that your compiler created?

Cheers


Markus (mfro)

Posts 99
09 Nov 2016 10:02


Gunnar von Boehn wrote:

 
Markus (mfro) wrote:

  Note that I took the freedom to add one single word to the sources to suit my compiler's needs as well (and no, it's not "inline").
 

 
  What did you change?
 

 
  Just made sort() static.
 


John Heritage

Posts 111
09 Nov 2016 13:02


Mo Retro wrote:

Kolbjørn Barmen wrote:

  The 5 dollar raspberry pi zero:
 
  EXTERNAL LINK 

  Wow these RPI zero scores are factor 4 compared to the Apollo with Gold core 2. What speed does the Rpi0 run?

We should keep this forum focused on Apollo/Vampire.

(But note that the Rpi0 would be slower by a factor of 10-20x if it were emulating 68K code).  Rpi 0 is a ARMv11 (slower than Cortex-A8) @ 1 GHz. 


Michal Warzecha

Posts 209
09 Nov 2016 13:53


Mayby in future. But CPU must be tested and stable as hell. And, need money, of course :)


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
09 Nov 2016 13:54


Markus (mfro) wrote:

        Note that I took the freedom to add one single word to the sources to suit my compiler's needs as well (and no, it's not "inline").
       

       
Can you post the ASM of the function that your compiler created?
 
Would be nice to see how smart your GCC was.


Markus (mfro)

Posts 99
09 Nov 2016 14:30


Maybe tonight.

What compiler are you using on the Amiga?


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
09 Nov 2016 14:32


Markus (mfro) wrote:

  Maybe tonight.
 
  What compiler are you using on the Amiga?
 

 
GCC 2.95

Not necessary to post the whole output only the inner loop of the SORT() is interesting to see.



Markus (mfro)

Posts 99
09 Nov 2016 14:45


Gunnar von Boehn wrote:

Markus (mfro) wrote:

  Maybe tonight.
   
    What compiler are you using on the Amiga?
 

 
  GCC 2.95

That's a very old version. Expect way better scores with something more current.



Markus (mfro)

Posts 99
10 Nov 2016 12:23



0000023c <.L6>:
      23c: 2052            moveal %a2@,%a0
      23e: 264a            moveal %a2,%a3
      240: 224c            moveal %a4,%a1
      242: 2205            movel %d5,%d1

00000244 <.L7>:
      244: 5381            subql #1,%d1
      246: 2019            movel %a1@+,%d0
      248: b1c0            cmpal %d0,%a0
      24a: 6e06            bgts 252 <.L8>
      24c: 2800            movel %d0,%d4
      24e: 2008            movel %a0,%d0
      250: 2044            moveal %d4,%a0

00000252 <.L8>:
      252: 26c0            movel %d0,%a3@+
      254: 4a81            tstl %d1
      256: 6cec            bges 244 <.L7>
      258: 2005            movel %d5,%d0
      25a: 4680            notl %d0
      25c: d080            addl %d0,%d0
      25e: 9180            subxl %d0,%d0
      260: c085            andl %d5,%d0
      262: 5385            subql #1,%d5
      264: 2988 0c00      movel %a0,%a4@(00000000,%d0:l:4)
      268: 4a85            tstl %d5
      26a: 6cd0            bges 23c <.L6>

Not sure I've got everything, but this should be it, basically.


Wawa T

Posts 695
10 Nov 2016 16:47


Markus (mfro) wrote:

 
Gunnar von Boehn wrote:

   
Markus (mfro) wrote:

      Maybe tonight.
     
      What compiler are you using on the Amiga?
     

     
    GCC 2.95
   

    That's a very old version. Expect way better scores with something more current.
   
 

 
  apparently this is being used on amiga for a reason. i have not examined the quality of code it produces (neither im able to), but others did. gcc 68k backend along with appropriate platform extensions isnt maintained properly anymore as it seems, therefore the results 3.x.x and upwards are not being considered superior.
 
  nevertheless worth to mention perhaps, that there are gcc 3.x.x and 4.x.x toolchains for amiga and up to gcc 6.1.0 for aros 68k, which should be able to produce regular amiga 68k binaries.
   

 
 


Markus (mfro)

Posts 99
10 Nov 2016 20:46


We still have gcc 2.59 on the Atari platform as well.

It's main advantage is it's size and memory requirements that are moderate enough to run on the limited, old machines themselves.

Other than that, it's hopelessly aged.

Can't judge on the Amiga platform (don't know anything about it), but on Atari, I can say that newer compilers (although it's correct the m68k backend is not maintained anymore as actively as desireable) nevertheless still benefit from active development. Not only that optimization improves (gcc mainly does this on RTL level, so the backend doesn't matter much) and new features (like e.g. LTO) also improve code, the syntax analyzer has improved much. I.e. it's much more difficult to create bugs if you listen to the compiler warnings.
The drawback - as mentioned already - is that its performance and memory requirements allow it to run only on high end Atari clones (and even there, it's slow) or as cross compiler.

posts 48page  1 2 3