Bustest scores for various systems

trixster · 01 December 2021, 17:16

Here are a load of bustest scores for fast ram from a few different system configurations. Some are medium word and some are long word, so not all scores are exactly comparable but should be pretty close.

Code:

1) WarpEngine 060 96Mhz (ram 48Mhz) = 54.3MB/s readm, 34.3MB/s writem
2) WarpEngine 80Mhz (ram 40Mhz) = 45.4MB/s readm, 28.0MB/s writem
3) WarpEngine 50Mhz (ram 25Mhz) = 27.7MB/s read, 17.8MB/s write
4) WarpEngine 040 40Mhz = 51.8MB/s readm, 28.5 writem

5) CSMK2 66Mhz (ram 66Mhz) = 43MB/s read, 30MB/s write
6) CSMK2 100Mhz (ram 50Mhz) = 37MB/s read, 23MB/s write
7) CSMK2 50Mhz (ram 50Mhz) = 33.1 MB/s readm, 23MB/s writem
8) CSMk2 040 40Mhz = 29.6MB/s read, 19.8MB/s write

9) GVP Trexx-II 060 50MHz = 41.3MB/s readm, 22.3MB/s writem
10) Apollo 4060 75Mhz (ram 37.5mhz) = 36.2MB/s readm, 24.3MB/s writem
11) Apollo 4060 80Mhz (ram 40Mhz) = 39.0MB/s readm, 26.0MB/s writem

12) BFG9060 (68060-TK) 100Mhz (ram 100Mhz) = 93.2MB/s readm, 88.1MB/s writem. Chip = 3.5 readw, 7.1 readm, 7.0 writem
12a) BFG9060 (68060-TK) 50Mhz (ram 50Mhz) = 49.8MB/s readm, 46.0MB/s writem. 
12b) BFG9060 040 25mhz = 39.2MB/s readm, 29.6MB/s writem

13) DarrenHD reported his CSMk3 at 50Mhz = 45MB/s read
14) CSMk3 75Mhz = 71.9MB/s readm, 54.8MB/s writem
15) CSPPC 50Mhz (60ns ram 50Mhz) = 54.4MB/s readm, 38.9MB/s writem
16) CSPPC 60Mhz = 46.9MB/s readm, 36.1MB/s writem (faster ram needed!)

17) TF360 50Mhz = 16.5MB/s read, 9.8MB/s write (beta firmware)
18) TF360 75Mhz = 24.8MB/s read, 14.5MB/s write (beta firmware)
19) TF330 50Mhz = 17.0MB/s readl, 23.5MB/s writel

20) Warp1260 50Mhz = 40.4MB/s readm, 30.7MB/s writem
21) Warp1260 80Mhz = 58.7MB/s readm, 44.5MB/s writem
22) Warp1260 100Mhz = 72.1MB/s readm, 53.2MB/s writem
22a) Warp1260 105Mhz = 75.6MB/s readm, 55.6MB/s writem

23) TF1260 50Mhz 7 Sep 21 beta fw = 42.6MB/s read, 41.4MB/s write
24) TF1260 50Mhz 17 Dec 21 beta fw 44.1MB/s readl, 41.1MB/s writel
25) TF1260 80Mhz 7 Sep 21 beta fw = 66.6MB/s read, 67.2MB/s write
26) TF1260 100Mhz 17 Dec 21 beta fw = 83.0MB/s readl, 74.2MB/s writel

27) Blizzard 1260 50Mhz (ram 50Mhz) = 36.2MB/s readm, 24.3MB/s writem
28) Blizzard 1260 80Mhz (ram 80Mhz) = 57.6MB/s read, 38.5MB/s write

29) Apollo 1240 40Mhz = 46.2MB/s readm, 26.2MB/s writem

30) BPPC 060 50Mhz = 38.8MB/s readm, 27.4MB/s writem
31) BPPC 040 25Mhz = 20.9MB/s readm, 10.6MB/s writem

32) TF1230 030 50Mhz = 25.8MB/s readm, 35.3MB/s writem
33) Blizzard 1230 50Mhz = 26.3MB/s readm, 31.0MB/s writem
34) Blizzard 1230 60Mhz = 32.2MB/s readm, 37.6MB/s writem
35) ACA1221 020 42.56Mhz = 26.0MB/s readm, 37.5MB/s writem

36) A3000 ZorRam 256MB Zorro3 ram = 6.8MB/s read, 4.0MB/s write
36a) A4000 ZorRam 256MB Zorro3 ram = 7.8MB/s readm, 4.4MB/s writem
37) PicassoIV Z2 060 50Mhz = 1.8MB/s readm, 3.6MB/s writem
38) PicassoIV Z3 060 50Mhz = 3.9MB/s readm, 9.6MB/s writem
39) Cybervison 64 A3000 060 50Mhz = 4.8MB/s readm, 12.4MB/s writem

40) A4000 64MB mobo fast ram = 11.5MB/s read, 6.1MB/s write
41) A4000 A3640 25Mhz 040 = 11.5MB/s readm, 6.1MB/s writem
42) A4000 A3640 with clk90 & waitstate mod = 13.9MB/s readm, 8.1MB.s writem
43) A3660 50Mhz = 13.9MB/s readm, 8.1MB/s writem
44) A3660SDR (presumed 060 at 50Mhz) = 38.4MB/s readm, 30.0MB/s writem
45) ZZ9000 = 6.5MB/s readm, 7.2MB/s writem

46) TekMagic 2060 50Mhz = 40.9MB/s readm, 34.2MB/s writem
47) TekMagic 2060 64Mhz = 52.8MB/s readm, 44.3MB/s writem

48) A500 ACA500Plus + Blizzard 1260 66Mhz = 46.6MB/s readm, 32.0MB/s

Please post your own results and I'll add them to the table, or to update my figures if they're wrong

https://aminet.net/package/util/moni/bustest

Basic usage is (in a shell): bustest fast
You can also get chip and rom speeds with "bustest chip rom". All three can be strung together.
For specific address ranges (ie you want to test zorro 3 ram rather than your highest priority ram) then consult the documentation, but it's basically bustest addr=07010000

Superman · 01 December 2021, 23:36

I have attached a screenshot of my A1200 with a TF1230 50mhz with 128mb ram

trixster · 02 December 2021, 00:12

Thanks, added

SpeedGeek · 02 December 2021, 00:56

The error problem with Bustest is the data cache is enabled (by Setpatch). Now, most results for the Fast RAM and Kickstart ROM are a mixture (cache + memory). But Chip RAM is correct (Thanks to the cache disable in hardware).

The short solution is - "CPU NODATACACHE" but is anyone interested? Probably not...

P.S. I already posted this reply on A1K.org (using my crappy German).

daxb · 02 December 2021, 12:09

I don't see that as a problem. You get (much) slower values if data cache is disabled. Practically CPU caches are activated and used in "real world" situation. To compare different systems it is good enough IMO and the important bottleneck is chipram what gives correct values. However, why not write a bus speed test program that measures real bus speed?

indigolemon · 02 December 2021, 12:14

Attached results for BPPC 060@50MHz with 256MB RAM

trixster · 02 December 2021, 12:46

Thanks, added.

SpeedGeek · 02 December 2021, 14:08

Quote:

Originally Posted by daxb

I don't see that as a problem. You get (much) slower values if data cache is disabled. Practically CPU caches are activated and used in "real world" situation. To compare different systems it is good enough IMO and the important bottleneck is chipram what gives correct values. However, why not write a bus speed test program that measures real bus speed?

From the Bustest documentation:

"Bustest is a small utility to measure data transfer speed
to and from memory."

The speed of the data cache in many cases is very much faster than physical memory but there is a notable exception for the 040 and 060 with the "Copyback" mode (see the Bustest documentation).

Now, how can you expect realistic results with 3 different CPUs having different data cache sizes and modes, unless those caches are disabled by default?

Damion · 02 December 2021, 18:44

TekMagic 2060 (and others) in this old thread:

https://eab.abime.net/showthread.php?t=48407

daxb · 02 December 2021, 20:49

Quote:

Originally Posted by SpeedGeek

Now, how can you expect realistic results with 3 different CPUs having different data cache sizes and modes, unless those caches are disabled by default?

Just measure chipram for "realistic" results. Or use SysInfo.

nogginthenog · 03 December 2021, 19:16

CSMK2 68060 @ 50mhz

Code:

7.Ram Disk:> bustest fast 
BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $08908000  readw      71.5 ns   normal      28.0 * 10^6 byte/s
fast      $08908000  readl     122.1 ns   normal      32.8 * 10^6 byte/s
fast      $08908000  readm     122.2 ns   normal      32.7 * 10^6 byte/s
fast      $08908000  writew     87.4 ns   normal      22.9 * 10^6 byte/s
fast      $08908000  writel    173.4 ns   normal      23.1 * 10^6 byte/s
fast      $08908000  writem    175.2 ns   normal      22.8 * 10^6 byte/s

trixster · 03 December 2021, 20:35

Quote:

Originally Posted by Damion

TekMagic 2060 (and others) in this old thread:

https://eab.abime.net/showthread.php?t=48407

Great link, many thanks! I had no idea that was on eab. I’ll take a few results from there to add to the op

Wrangler · 03 December 2021, 23:07

Here you are mate, some extra stats for you:

WarpEngine 060@80MHz (mem at 40MHz)

Code:

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $097E0000  readw      52.5 ns   normal      38.1 * 10^6 byte/s
fast      $097E0000  readl      88.9 ns   normal      45.0 * 10^6 byte/s
fast      $097E0000  readm      87.5 ns   normal      45.7 * 10^6 byte/s
fast      $097E0000  writew     70.3 ns   normal      28.5 * 10^6 byte/s
fast      $097E0000  writel    141.3 ns   normal      28.3 * 10^6 byte/s
fast      $097E0000  writem    142.4 ns   normal      28.1 * 10^6 byte/s

CSPPC 060@60MHz

Code:

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $08BF0000  readw      51.0 ns   normal      39.2 * 10^6 byte/s
fast      $08BF0000  readl      85.1 ns   normal      47.0 * 10^6 byte/s
fast      $08BF0000  readm      85.3 ns   normal      46.9 * 10^6 byte/s
fast      $08BF0000  writew     53.8 ns   normal      37.2 * 10^6 byte/s
fast      $08BF0000  writel    106.1 ns   normal      37.7 * 10^6 byte/s
fast      $08BF0000  writem    110.7 ns   normal      36.1 * 10^6 byte/s

CSMKII 060@50MHz

Code:

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $08FB0000  readw      71.2 ns   normal      28.1 * 10^6 byte/s
fast      $08FB0000  readl     121.7 ns   normal      32.9 * 10^6 byte/s
fast      $08FB0000  readm     120.8 ns   normal      33.1 * 10^6 byte/s
fast      $08FB0000  writew     87.2 ns   normal      22.9 * 10^6 byte/s
fast      $08FB0000  writel    173.0 ns   normal      23.1 * 10^6 byte/s
fast      $08FB0000  writem    174.1 ns   normal      23.0 * 10^6 byte/s

Apollo 4060@80MHz

Code:

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $08D70000  readw      58.5 ns   normal      34.2 * 10^6 byte/s
fast      $08D70000  readl     103.1 ns   normal      38.8 * 10^6 byte/s
fast      $08D70000  readm     102.7 ns   normal      39.0 * 10^6 byte/s
fast      $08D70000  writew     77.9 ns   normal      25.7 * 10^6 byte/s
fast      $08D70000  writel    154.4 ns   normal      25.9 * 10^6 byte/s
fast      $08D70000  writem    153.6 ns   normal      26.0 * 10^6 byte/s

Looks like I need to get some faster RAM for the CSPPC, but they are a bit arsey about the internal chip layout. And the Apollo is crappy...

trixster · 03 December 2021, 23:38

Thanks Wrangler, loads of new data added.

Chucky · 04 December 2021, 15:46

a list with chipmemspeeds would also be very interesting

grelbfarlk · 05 December 2021, 02:09

Also a list for accessing RTG RAM
Also Wrangler after you release that EMU10k DSP driver on monday, can you build bustest for WarpOS?

Wrangler · 05 December 2021, 16:50

I bunged the CSPPC into 60ns mode and achieved 60MB/s:

Code:

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $08BB8000  readw      42.8 ns   normal      46.7 * 10^6 byte/s
fast      $08BB8000  readl      68.5 ns   normal      58.4 * 10^6 byte/s
fast      $08BB8000  readm      66.7 ns   normal      60.0 * 10^6 byte/s
fast      $08BB8000  writew     42.2 ns   normal      47.4 * 10^6 byte/s
fast      $08BB8000  writel     85.0 ns   normal      47.1 * 10^6 byte/s
fast      $08BB8000  writem     90.3 ns   normal      44.3 * 10^6 byte/s
chip      $00018000  readw     856.8 ns   normal       2.3 * 10^6 byte/s
chip      $00018000  readl     857.0 ns   normal       4.7 * 10^6 byte/s
chip      $00018000  readm     857.3 ns   normal       4.7 * 10^6 byte/s
chip      $00018000  writew    568.5 ns   normal       3.5 * 10^6 byte/s
chip      $00018000  writel    568.6 ns   normal       7.0 * 10^6 byte/s
chip      $00018000  writem    569.0 ns   normal       7.0 * 10^6 byte/s

@chucky: chip mem speeds just for you
@grelbfarlk: another couple of weeks and it'll be done

. In the meantime, ppcmemtest says the PPC accesses at 78MB/s...

trixster · 18 December 2021, 11:20

TF1260 17 Dec 21 fw results added, for 50 and 100Mhz

trixster · 21 April 2022, 10:59

Added CyberVision64 memory speeds

utri007 · 23 April 2022, 02:03

Apollo 1260 68060 80/40mhz

Quote:

7.WorkBench:> bustest fast
BusSpeedTest 0.19 (mlelstv) Buffer: 262144 Bytes, Alignment: 32768
========================================================================
memtype addr op cycle calib bandwidth
fast $01AA0000 readw 57.7 ns normal 34.7 * 10^6 byte/s
fast $01AA0000 readl 102.4 ns normal 39.1 * 10^6 byte/s
fast $01AA0000 readm 102.6 ns normal 39.0 * 10^6 byte/s
fast $01AA0000 writew 77.6 ns normal 25.8 * 10^6 byte/s
fast $01AA0000 writel 155.1 ns normal 25.8 * 10^6 byte/s
fast $01AA0000 writem 154.3 ns normal 25.9 * 10^6 byte/s

01 December 2021, 23:36	#2
Superman Super Member Join Date: Sep 2014 Location: Wakefield Age: 48 Posts: 1,334	I have attached a screenshot of my A1200 with a TF1230 50mhz with 128mb ram Attached Thumbnails

02 December 2021, 12:14	#6
indigolemon Bit Copying Bard Join Date: Jan 2017 Location: Kelty, Fife, Scotland Age: 41 Posts: 1,293	Attached results for BPPC 060@50MHz with 256MB RAM Attached Thumbnails

05 December 2021, 02:09	#16
grelbfarlk Registered User Join Date: Dec 2015 Location: USA Posts: 2,902	Also a list for accessing RTG RAM Also Wrangler after you release that EMU10k DSP driver on monday, can you build bustest for WarpOS? Last edited by grelbfarlk; 05 December 2021 at 04:40.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Gotek saves hi scores?	ransom1122	Amiga scene	9	13 February 2015 13:11
Scores/Instruments for Sonix	BigB	request.Modules	63	23 August 2012 22:47
High Scores?	adolescent	Retrogaming General Discussion	5	25 March 2007 17:57
Unbeatable high scores	Shoonay	project.WHDLoad	6	04 October 2005 10:32
Pinball Illusions - saving scores???	Medvind	support.Games	2	20 September 2002 16:14

02 December 2021, 00:12	#3
trixster Guru Meditating Join Date: Jun 2014 Location: England Posts: 2,337	Thanks, added

02 December 2021, 00:56	#4
SpeedGeek Moderator Join Date: Dec 2010 Location: Wisconsin USA Age: 60 Posts: 839	The error problem with Bustest is the data cache is enabled (by Setpatch). Now, most results for the Fast RAM and Kickstart ROM are a mixture (cache + memory). But Chip RAM is correct (Thanks to the cache disable in hardware). The short solution is - "CPU NODATACACHE" but is anyone interested? Probably not... P.S. I already posted this reply on A1K.org (using my crappy German).

02 December 2021, 12:09	#5
daxb Registered User Join Date: Oct 2009 Location: Germany Posts: 3,303	I don't see that as a problem. You get (much) slower values if data cache is disabled. Practically CPU caches are activated and used in "real world" situation. To compare different systems it is good enough IMO and the important bottleneck is chipram what gives correct values. However, why not write a bus speed test program that measures real bus speed?

02 December 2021, 12:46	#7
trixster Guru Meditating Join Date: Jun 2014 Location: England Posts: 2,337	Thanks, added.

02 December 2021, 18:44	#9
Damion Registered User Join Date: Mar 2008 Location: US Posts: 315	TekMagic 2060 (and others) in this old thread: https://eab.abime.net/showthread.php?t=48407

03 December 2021, 23:38	#14
trixster Guru Meditating Join Date: Jun 2014 Location: England Posts: 2,337	Thanks Wrangler, loads of new data added.

04 December 2021, 15:46	#15
Chucky Registered User Join Date: Mar 2015 Location: Karlstad / Sweden Age: 52 Posts: 1,210	a list with chipmemspeeds would also be very interesting

18 December 2021, 11:20	#18
trixster Guru Meditating Join Date: Jun 2014 Location: England Posts: 2,337	TF1260 17 Dec 21 fw results added, for 50 and 100Mhz

21 April 2022, 10:59	#19
trixster Guru Meditating Join Date: Jun 2014 Location: England Posts: 2,337	Added CyberVision64 memory speeds

Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)