Overview Features Instructions Performance Forum Downloads Products Reseller Contact

Welcome to the Apollo Forum

This forum is for people interested in the APOLLO CPU.
Please read the forum usage manual.
VISIT APOLLO IRC CHANNEL



All TopicsNewsPerformanceGamesDemosApolloVampireReleases
Performance and Benchmark Results!

World Record ATARI FreeMiNT Kronos 68080-FPUpage  1 2 3 4 5 6 

Markus (mfro)

Posts 83
23 Dec 2017 10:22


Gunnar von Boehn wrote:

    "Bogo" means literally "nonsense".
   

   
    Wouldn't agree its "nonsense" in general (and literally, it rather means "fake" than "nonsense"). It's just that most people didn't understand its meaning when it first appeared on the Linux console (and later on in MiNT and elsewhere as well).
   
    It is (and was always) meant for calibration of sub-timer-resolution delay loops in operating systems and drivers *only*. There, it absolutely makes sense and is indispensable (as there is no alternative). Basically , it numbers the minimum time for a certain µC platform "to do nothing".
   
    Not really suitable as a benchmark number, I agree ;). 
   


Peter Slegg

Posts 13
01 Jan 2018 12:43


It's more primitive than I imagined.

The OpenGL value on the Milan60 is 187.6


Gunnar von Boehn
(Apollo Team Member)
Posts 3089
01 Jan 2018 13:45


Peter Slegg wrote:

  It's more primitive than I imagined.
 
  The OpenGL value on the Milan60 is 187.6
 

 
The screnshots shown so far were compared to 100Mhz 68060
- with 100Mhz 68060 as basis of 100.

What is MILAN compared to 100MHz 68060?


Olivier Landemarre

Posts 39
01 Jan 2018 19:03


Hello

Reference value

CT60 68060 at 100Mhz with radeon card (most common configuration I have several results) reference value is around 1100
Result Appolo is around 960 (for the moment)
This results can be compared it is not video dependant.
Milan 040 68040 at 25Mhz give a result of around 210
A quadra 650 68040 at 33Mhz running MagicMac : 157

Looks like CT63 68060 at 100Mhz with Supervidel video results presented here are far lower than CT60 + radeon, why I don't know, perhaps link to supervidel or configuration

With Appolo results we should pay attention to timer 200Hz, if this timer is wrong results are wrong, I have found some issue for exemple under Aranym with JIT under supervisor mode the timer is twice slower as it should so some results are wrong such like VDI

Gunnar von Boehn wrote:

Peter Slegg wrote:

  It's more primitive than I imagined.
   
    The OpenGL value on the Milan60 is 187.6
 

 
  The screnshots shown so far were compared to 100Mhz 68060
  - with 100Mhz 68060 as basis of 100.
 
  What is MILAN compared to 100MHz 68060?




OneSTone O2o

Posts 119
01 Jan 2018 19:53


Olivier Landemarre wrote:

  Looks like CT63 68060 at 100Mhz with Supervidel video results presented here are far lower than CT60 + radeon, why I don't know, perhaps link to supervidel or configuration
 

 
  Maybe some parts of the benchmark runs in ST-RAM. On a ct60 based system that RAM is still on the Falcon mainboard and clocks at just 16 MHz. 68060 with it's cache relativates that a bit.
 
  For another reason I don't know if it is of relevance: Videl core in unaccelerated Falcon slows down original CPU when using higher resolutions. This is similar to original Amiga design, also there Denise slows down the CPU with more graphics activity. As far as I know SuperVidel is also a shared memory graphics architecture. CT60 with CTPCI and Radeon PCI card is not.
 


Gunnar von Boehn
(Apollo Team Member)
Posts 3089
01 Jan 2018 19:57


Olivier Landemarre wrote:

Hello
   
Reference value
   
CT60 68060 at 100Mhz with radeon card (most common configuration I have several results) reference value is around 1100
Result Appolo is around 960 (for the moment)

 
The screenshot above shows 1.76 times 68060@100
So the score is ~ 1920, right?
 
You are joking right, this can not be, then it would be 10 times more than the MILAN 060?


Olivier Landemarre

Posts 39
02 Jan 2018 13:26


Hello Gunnar

This result is video dependant, so from one configuration to the other result can be far different. So if you want compare apple with apple you should not look at it. The result is not an aritmetic average but geometric, if only one test give far lower result than other the weight in the result is more important.
This 68060@100 looks give low result for a 68060@100. I have often seen such issue on Atari 68060 configuration, for exemple for compatiblity reason system deactivate cache after load program for several seconds (need for auto modification code of some very old software). Kronos use plugins and this plugins are fully managed by system as simple programs, it is loaded just before the test and this tests are quite small in time so the test can be partially or fully done without cache. There is several possible setup provide bien system to fix this. So I said result are low in this configuration, it is not for me representative to what we are able to do with 68060@100 and I think it is what is interesting you.
As today Vampire results are probably not the best result we can have and there is not yet dedicated 68080 tinygl optimized module.

If you want compare best CT60 possible results as reference take this file as reference (rename it as default.abh) : EXTERNAL LINK 

I suppose if I use only vbcc try to compil C code will not help to have better result and I should add inline 68080 instruction in source code? I think it could be interesting to do it.

Gunnar von Boehn wrote:

Olivier Landemarre wrote:

  Hello
   
  Reference value
   
  CT60 68060 at 100Mhz with radeon card (most common configuration I have several results) reference value is around 1100
  Result Appolo is around 960 (for the moment)
 

 
   
  The screenshot above shows 1.76 times 68060@100
  So the score is ~ 1920, right?
 
  You are joking right, this can not be, then it would be 10 times more than the MILAN 060?




Olivier Landemarre

Posts 39
02 Jan 2018 13:30


If there enough memory all test is done  in fast ram only possible slow done is the transfer to video on slow bus. I don't think with CT60 + supervidel there is this issue, Supervidel use it's own video memory as Radeon.

For me slow down is to fix in configuration.

oneSTone o2o wrote:

Olivier Landemarre wrote:

    Looks like CT63 68060 at 100Mhz with Supervidel video results presented here are far lower than CT60 + radeon, why I don't know, perhaps link to supervidel or configuration
 

 
  Maybe some parts of the benchmark runs in ST-RAM. On a ct60 based system that RAM is still on the Falcon mainboard and clocks at just 16 MHz. 68060 with it's cache relativates that a bit.
 
  For another reason I don't know if it is of relevance: Videl core in unaccelerated Falcon slows down original CPU when using higher resolutions. This is similar to original Amiga design, also there Denise slows down the CPU with more graphics activity. As far as I know SuperVidel is also a shared memory graphics architecture. CT60 with CTPCI and Radeon PCI card is not.
 




Peter Slegg

Posts 13
02 Jan 2018 20:31


Using the CT60 abh file as my default the Milan give an OpenGL figure of 56.4  The other tests show it to be about 45% the speed of the CT60@100 which makes sense for a Milan60@50

posts 109page  1 2 3 4 5 6