Please help me, and run this test on 68k Amiga!

[biggun]

Many thanks in advance for helping me ...

As you might know I part of the Natami fans (www.natami.net) and I'm right now evaluating the usability of using a Coldfire for a new inexpensive classic AMIGA.

I believe that "whatever CPU" we are going to use, it should provide a minimum performance level for satisfying todays demands.
To be able to better understand where the Coldfire stands compared to classic 68k systems or compared to UAE, OS4,
I've put together a small benchsuit.

 
You would help me a lot if you could run this below program on your AMIGA and post the results.
Please be prepared that it will run several minutes.
Please be so kind and state your machine details (CPU etc)

AMIGA-OS EXE (50k)
bench68k

For those interested here is the source:
bench68k.c
bench68k_test.s


Many thanks in advance

Gunnar

[biggun]

Hi,

What is your biggest free block? (output of avail)

Cheers

[biggun]

1728216  :-D

If you have a free block of 2.2 MB it should work.
Maybe you could close some aps?
But be aware that on the 020 it will runs for a while.
The test should give you time for a coffee or two.

[biggun]

Framiga wrote:
here under AmigaOS3.9-68060 @ 60 Mhz (CSPPC)

and here under AmigaOS4 on the same machine - 604e @ 200 Mhz (just for the sake of)

Cool, many Thanks!

[biggun]

DJBase wrote:

biggun wrote:

1728216  :-D

If you have a free block of 2.2 MB it should work.
Maybe you could close some aps?
But be aware that on the 020 it will runs for a while.
The test should give you time for a coffee or two.

Will 2530608 be enough? OS3.9 eats all my memory.

I've just uploaded a new version for you.
The new version will go down to lower memory requirements if it can't get the big buffers.
It should work on your 020 system with 2MB free.

Can you please redownload the test and try again?

many thanks

[biggun]

DJBase wrote:
The test will run some minutes, please be patient.
EMT trap
and then it quits.  :-?

Do you have a FPU?

[biggun]

First: Many thanks for your help!!!

Its real great to have a community of helpful people!!

To answer some questions:
The test needs 2MB free memory and it needs ixemul.library.
At the moment I have no idea why the test does not run on DJbase 020.

Thanks for your testing. I spotted an error in the test,
corrected it and added a some more CPU subtest. I have uploaded a new test file version (new version 4.2).

It would be cool to have some 030, 040 and 060 results with the V4.2 version of the test.
Framiga, maybe you can run the V4 test agin on your 060?


I'll post some CF results

Many thanks !!

[biggun]

wawrzon wrote:
@biggun:
ok i will run the tests later today. by the way: i cant get your 194x working on the mediator/voodoo3 setup. "Unable to init SDL: Found no sufficiently capable CGX visuals". of course voodoo has an p96 driver but this should not be an issue, right? on picasso4 its working fine. any clues?

I'll make some native CyberGFX port when I have time.
This will run faster than SDL. :-)

To run it today, please tell me which screen modes have you got configured?

If I recall it correctly the game wants a 15 or 16 bit screen mode with 320x200 resolution.
If you do not have such screen mode configured then the game can not start.

Cheers

[biggun]

wawrzon wrote:
lets start the orgy then.
i still would like to see how an a4k and a1k2 030 behaves, but it occures all the a4ks have already 040 built in and i am too tired now to swap the cpu-board again. maybe tomorrow.

cheers


a1k2_040-50 (blizzard?):

PROCESSOR:      CPU 68040/68882fpu
CUSTOM CHIPS:   AA PAL Alice (id=$0023), AA Lisa (id=$00F8)
VERS:   Kickstart version 39.106, Exec version 39.47, Disk version 39.29
RAM:    Node type $A, Attributes $5 (FAST), at $68000000-$68FFFFFF (16.0 meg)
        Node type $A, Attributes $703 (CHIP), at $1000-$1FFFFF (~2.0 meg)
    BOARDS:
     Board + ROM (HD?) (unidentified):   Prod=8512/17($2140/$11) (@$EA0000 128K)

-------------------------------------------
Processor & Memory Performance Bench v4.20
-------------------------------------------
Stop all program before the test.
Do not use the computer during the test.
The test will run some minutes, please be patient.
Total memory required = 4.2 MB.
Calibration loops: 3
-------------------------------------------
Comparing different CPU functions:
Results are in million instructions per sec.
Higher value is faster.
CPU-Benchmark      2MB   16KB    4KB    1KB
-------------------------------------------
addi              90.7   90.6   89.3   84.8
shift             92.0   91.2   90.5   85.6
mix               91.7   91.1   89.9   84.9
mul               24.9   24.8   24.8   24.4
bra-un             8.5    8.4    8.5    8.4
bra-pre           31.6   31.5   31.4   30.9
bsr                3.9    3.9    3.9    3.9
nop                5.7    5.7    5.6    5.6
-------------------------------------------
Measuring memory latency:
Result is Million random accesses per sec.
Higher value is faster.
Memory Latency     2MB   16KB    4KB    1KB
-------------------------------------------
random read        2.7
-------------------------------------------
Measuring memory throughput:
Results are in MB/sec. Higher value is faster.
Memory 2 Memory
Alignment 0-0      2MB   16KB    4KB    1KB
-------------------------------------------
glibc memcpy      31.8   31.6   31.4   30.5
read 8            19.9   19.9   19.9   19.8
read 16           31.2   31.2   31.2   30.9
read 32           37.0   36.9   36.8   36.5
read 32x4         33.9   33.9   33.8   33.4
read 32x4B        37.0   36.9   36.8   36.4
write 8           23.9   23.9   23.9   23.6
write 16          25.3   25.3   25.2   25.1
write 32          25.3   25.3   25.3   25.1
write 32x4        25.3   25.3   25.3   25.1
write 32x4B       25.3   25.3   25.3   25.2
copy 8            22.6   22.3   21.5   21.8
copy 32           31.7   31.5   30.6   30.3
copy 32x4         30.1   30.0   29.8   28.7
copy 32x4B        29.5   29.4   29.3   28.2
-------------------------------------------
Cache 2 Cache
Alignment 0-0      2MB   16KB    4KB    1KB
-------------------------------------------
glibc memcpy      31.8   30.6  173.7  184.8
read 8            19.9   20.6   25.4   25.5
read 16           31.2   34.8   97.1   99.9
read 32           37.0   42.0  184.3  197.1
read 32x4         33.9   38.0  125.6  131.3
read 32x4B        37.0   41.7  177.3  195.5
write 8           23.9   25.8   49.6   50.5
write 16          25.3   27.4   50.3   50.8
write 32          25.3   28.5   98.4  100.2
write 32x4        25.4   29.1  181.0  194.3
write 32x4B       25.3   28.9  150.0  158.5
copy 8            22.6   23.2   43.9   44.8
copy 32           31.7   33.6  177.0  192.9
copy 32x4         30.1   30.3  174.2  190.4
copy 32x4B        29.5   29.6  145.6  153.5
-------------------------------------------

Hi,

Are you sure that this is a 68040 system?

Your result looks 100% like a 50MHz 68060.
A 68060 can execute 2 instructions per clock.
Which means that a 50 MHz 060 will reach 90-100 in the first three CPU tests (addi,shift,mix)
A 68060 only needs 2 clocks for a multiplication, which means it will reach a score of 25 in mul test.

A 68040 would be much slower at the same clockspeed.

Cheers

[biggun]

Here is the result of the Coldfire 54455@266MHz
The 54455 is an embedded CF with 16bit bus.

I think the 266MHz V4m is CPU wise roughly comparable
to a 68060 with 130 MHz
to a 68040 with 280 MHz
to a 68030 with 600 MHz

-------------------------------------------                                    
Processor & Memory Performance Bench v4.20                                      
-------------------------------------------                                    
Stop all program before the test.                                              
Do not use the computer during the test.                                        
The test will run some minutes, please be patient.                              
Total memory required = 4.2 MB.                                                
Calibration loops: 8                                                            
-------------------------------------------                                    
Comparing different CPU functions:                                              
Results are in million instructions per sec.                                    
Higher value is faster.                                                        
CPU-Benchmark      2MB   16KB    4KB    1KB                                    
-------------------------------------------                                    
addi             251.3  251.3  251.3  251.3                                    
shift            293.2  293.2  251.3  251.3                                    
mix              439.8  439.8  439.8  439.8                                    
mul               67.7   67.7   67.7   65.2                                    
bra-un            41.9   42.9   40.9   41.9                                    
bra-pre          117.3  117.3  117.3  109.9                                    
bsr               13.3   13.4   13.3   13.2                                    
nop               45.1   44.0   44.0   44.0                                    
-------------------------------------------                                    
Measuring memory latency:                                                      
Result is Million random accesses per sec.                                      
Higher value is faster.                                                        
Memory Latency     2MB   16KB    4KB    1KB                                    
-------------------------------------------                                    
random read        1.0                                                          
-------------------------------------------                                    
Measuring memory throughput:                                                    
Results are in MB/sec. Higher value is faster.                                  
Memory 2 Memory                                                                
Alignment 0-0      2MB   16KB    4KB    1KB                                    
-------------------------------------------                                    
glibc memcpy      67.7   67.7   66.4   56.7                                    
read 8            81.8   74.9   74.9   69.0                                    
read 16           95.1   95.1   95.1   88.0                                    
read 32          121.3  121.3  117.3  109.9                                    
read 32x4        121.3  121.3  121.3  106.6                                    
read 32x4B       121.3  121.3  117.3  106.6                                    
write 8           13.1   13.2   13.2   13.0                                    
write 16          26.1   26.1   26.1   25.5                                    
write 32          51.0   51.0   51.0   48.9                                    
write 32x4        51.7   51.0   51.0   48.9                                    
write 32x4B      185.2  185.2  175.9  153.0                                    
copy 8            23.3   23.3   23.3   22.0                                    
copy 32           67.7   67.7   67.7   57.7                                    
copy 32x4         65.2   65.2   65.2   55.8                                    
copy 32x4B       117.3  121.3  117.3   90.2                                    
-------------------------------------------                                    
Cache 2 Cache                                                                  
Alignment 0-0      2MB   16KB    4KB    1KB                                    
-------------------------------------------                                    
glibc memcpy      67.7  106.6  106.6  103.5                                    
read 8            74.9   92.6  140.7  135.3                                    
read 16           92.6  125.7  219.9  219.9                                    
read 32          121.3  175.9  586.4  502.6                                    
read 32x4        121.3  185.2  703.7  703.7                                    
read 32x4B       121.3  185.2  879.6  879.6                                    
write 8           13.2   13.3   13.3   13.3                                    
write 16          26.1   26.7   26.7   26.7                                    
write 32          51.0   53.3   53.3   52.5                                    
write 32x4        51.0   53.3   53.3   53.3                                    
write 32x4B      185.2  207.0  207.0  207.0                                    
copy 8            23.1   23.8   26.7   26.7                                    
copy 32           67.7  100.5  106.6  103.5                                    
copy 32x4         65.2  103.5  106.6  103.5                                    
copy 32x4B       117.3  439.8  390.9  390.9                                    
-------------------------------------------

[biggun]

wawrzon wrote:

@biggun: "Are you sure that this is a 68040 system?"
my mistake! sure its 060 (corrected), ive only turned it yesterday for the first time since a year and didnt realize the old showconfig desnt recognize the cpu properly.

do you still need an a4k 030 benchmark? (i dont think so)

the results of the test are not really in favour of the coldfire. am i right that it doesnt contain a fpu unit? (which would render it still little less usable)

A 030@25Mhz would be interesting but only if its no hassle.

>the results of the test are not really in favour of the coldfire.
> am i right that it doesnt contain a fpu unit

The 54455 is the tiniest of the V4 Coldfires.
It has no FPU and only a 16 Bit bus.

There are other V4 Coldfires available that have more on chip cache, that have a FPU and that have 32Bit memory bus.

And the used V4m is a bit handicapped by the memory on dev-board. The memory is 16-bit and its DD2 memory.
Using DD2 memory makes sense if you have a very high clocked CPU. When you use it with a rather low clocked CPU the result will only be that you increase the memory latency.

A V4 with better memory (32bit) will have noticeable higher memory performance.

I leaning to use the V5 Coldfire.
From what I have read about the V5 Coldfire, I understand that the CPU performance should be about 3 times faster than the tested V4m.

Cheers

[biggun]

Thanks for all your help.

The information is very useful to me.
I made some stats out of the results I feel they provide a good overview. I'm not sure how to best visualize this here.

In short:
The 68K emul on MOS-PPC seems to more efficient than on x86-UAE (clock by clock).
A 1000 MHz G4 is about as effective in 68k emulation as a Core2Duo with 1500 Mhz.

The 16bit-Coldfire-V4m is performance wise roughly
in the middle between original 68060@50 and AONE@1000Mhz running 68k-Jit.


Cheers
Gunnar

[biggun]

a1200 wrote:
@biggun

Forget all this benchmarking stuff, I have a question... is that woman in the pic your other half?

LOL.
Yes this is Anja my wife. :-)
BTW I've been through this already, we have two kids now.

[biggun]

adonay wrote:
Seems there are sertin tests dealing with cache 2 cache that proves to be faster on a real 060 what is the reason for this ? V5 write 8 13.2 13.3 13.3 13.3 my 060 write 8 26.2 28.9 75.9 78.

To prevent any misunderstandings,
the CF that I've tested was not the V5 but the V4m.
(I'm still a waiting to get the my V5 board)

You are absolutely right with your observation:
"write 8 13.2 13.3 13.3 13.3"
This looks very suspicious.

I think the memory of the V4 is setup differently to the other machine.
I looks to me is the V4m is set up to "write through" instead of "copyback mode". I'm probably measuring something complete wrong here.

I'll check this