Apollo 1260 bustest results

[Damion]

I had some spare time today, so I thought I'd run some "bustest" benchmarks at various clockspeeds.

All tests were performed in WB, 640x512, 32 colors. (Pal High Res Laced.)

Hardware is an A1200, Escom 1D4 motherboard (timing fixes already done by Escom), Apollo 1260 with Rev 6 68060 (from PG) .

Bustest result at 50MHz:

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $01460000  readw      86.9 ns   normal      23.0 * 10^6 byte/s
fast      $01460000  readl     154.2 ns   normal      25.9 * 10^6 byte/s
fast      $01460000  readm     153.5 ns   normal      26.1 * 10^6 byte/s
fast      $01460000  writew    113.2 ns   normal      17.7 * 10^6 byte/s
fast      $01460000  writel    227.5 ns   normal      17.6 * 10^6 byte/s
fast      $01460000  writem    225.1 ns   normal      17.8 * 10^6 byte/s
chip      $00060000  readw    1006.4 ns   normal       2.0 * 10^6 byte/s
chip      $00060000  readl    1007.7 ns   normal       4.0 * 10^6 byte/s
chip      $00060000  readm    1006.9 ns   normal       4.0 * 10^6 byte/s
chip      $00060000  writew    916.7 ns   normal       2.2 * 10^6 byte/s
chip      $00060000  writel    917.5 ns   normal       4.4 * 10^6 byte/s
chip      $00060000  writem    914.9 ns   normal       4.4 * 10^6 byte/s



@ 64MHz:

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $01440000  readw      67.8 ns   normal      29.5 * 10^6 byte/s
fast      $01440000  readl     119.9 ns   normal      33.4 * 10^6 byte/s
fast      $01440000  readm     118.8 ns   normal      33.7 * 10^6 byte/s
fast      $01440000  writew     88.3 ns   normal      22.7 * 10^6 byte/s
fast      $01440000  writel    176.1 ns   normal      22.7 * 10^6 byte/s
fast      $01440000  writem    175.8 ns   normal      22.8 * 10^6 byte/s
chip      $00058000  readw    1005.1 ns   normal       2.0 * 10^6 byte/s
chip      $00058000  readl    1006.4 ns   normal       4.0 * 10^6 byte/s
chip      $00058000  readm    1004.0 ns   normal       4.0 * 10^6 byte/s
chip      $00058000  writew    613.4 ns   normal       3.3 * 10^6 byte/s
chip      $00058000  writel    613.7 ns   normal       6.5 * 10^6 byte/s
chip      $00058000  writem    614.2 ns   normal       6.5 * 10^6 byte/s



@ 66.6MHz

would not boot -- tried two separate oscillators



@ 70 MHz:

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $01408000  readw      61.8 ns   normal      32.4 * 10^6 byte/s
fast      $01408000  readl     109.4 ns   normal      36.6 * 10^6 byte/s
fast      $01408000  readm     108.6 ns   normal      36.8 * 10^6 byte/s
fast      $01408000  writew     80.9 ns   normal      24.7 * 10^6 byte/s
fast      $01408000  writel    161.3 ns   normal      24.8 * 10^6 byte/s
fast      $01408000  writem    161.7 ns   normal      24.7 * 10^6 byte/s
chip      $00058000  readw    1006.1 ns   normal       2.0 * 10^6 byte/s
chip      $00058000  readl    1005.6 ns   normal       4.0 * 10^6 byte/s
chip      $00058000  readm    1005.6 ns   normal       4.0 * 10^6 byte/s
chip      $00058000  writew    613.0 ns   normal       3.3 * 10^6 byte/s
chip      $00058000  writel    613.3 ns   normal       6.5 * 10^6 byte/s
chip      $00058000  writem    613.6 ns   normal       6.5 * 10^6 byte/s



@ 72MHz -- would not boot, may be a faulty/poorly seated oscillator


@ 73MHz

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $01460000  readw      58.9 ns   normal      33.9 * 10^6 byte/s
fast      $01460000  readl     104.8 ns   normal      38.2 * 10^6 byte/s
fast      $01460000  readm     105.3 ns   normal      38.0 * 10^6 byte/s
fast      $01460000  writew     77.4 ns   normal      25.8 * 10^6 byte/s
fast      $01460000  writel    153.5 ns   normal      26.1 * 10^6 byte/s
fast      $01460000  writem    154.1 ns   normal      26.0 * 10^6 byte/s
chip      $00060000  readw    1003.3 ns   normal       2.0 * 10^6 byte/s
chip      $00060000  readl    1004.4 ns   normal       4.0 * 10^6 byte/s
chip      $00060000  readm    1005.8 ns   normal       4.0 * 10^6 byte/s
chip      $00060000  writew    613.6 ns   normal       3.3 * 10^6 byte/s
chip      $00060000  writel    615.1 ns   normal       6.5 * 10^6 byte/s
chip      $00060000  writem    612.9 ns   normal       6.5 * 10^6 byte/s


@ 80MHz

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $01458000  readw      53.7 ns   normal      37.2 * 10^6 byte/s
fast      $01458000  readl      95.5 ns   normal      41.9 * 10^6 byte/s
fast      $01458000  readm      95.6 ns   normal      41.8 * 10^6 byte/s
fast      $01458000  writew     70.5 ns   normal      28.4 * 10^6 byte/s
fast      $01458000  writel    140.9 ns   normal      28.4 * 10^6 byte/s
fast      $01458000  writem    141.1 ns   normal      28.3 * 10^6 byte/s
chip      $00060000  readw     968.1 ns   normal       2.1 * 10^6 byte/s
chip      $00060000  readl     970.0 ns   normal       4.1 * 10^6 byte/s
chip      $00060000  readm     970.3 ns   normal       4.1 * 10^6 byte/s
chip      $00060000  writew    613.1 ns   normal       3.3 * 10^6 byte/s
chip      $00060000  writel    614.0 ns   normal       6.5 * 10^6 byte/s
chip      $00060000  writem    614.1 ns   normal       6.5 * 10^6 byte/s


Some PCMCIA tests (performed with nothing in the slot, after a reboot or after running "cardreset"):

@ 50MHz

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
user      $00628000  readw     702.2 ns   normal       2.8 * 10^6 byte/s
user      $00628000  readl    1264.5 ns   normal       3.2 * 10^6 byte/s
user      $00628000  readm    1263.6 ns   normal       3.2 * 10^6 byte/s
user      $00628000  writew    702.2 ns   normal       2.8 * 10^6 byte/s
user      $00628000  writel   1263.9 ns   normal       3.2 * 10^6 byte/s
user      $00628000  writem   1262.8 ns   normal       3.2 * 10^6 byte/s


@ 80MHz

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
user      $00628000  readw     692.9 ns   normal       2.9 * 10^6 byte/s
user      $00628000  readl    1259.8 ns   normal       3.2 * 10^6 byte/s
user      $00628000  readm    1259.7 ns   normal       3.2 * 10^6 byte/s
user      $00628000  writew    692.8 ns   normal       2.9 * 10^6 byte/s
user      $00628000  writel   1259.6 ns   normal       3.2 * 10^6 byte/s
user      $00628000  writem   1259.5 ns   normal       3.2 * 10^6 byte/s


When I have time again, I'll post similar results from an '030 card I have. IIRC the "chip" results are comparative (or better) due to faster bus access. I'd also like to try my unfixed C= 2B board (which seems to work with fine with this 1260 card), and check for any notable differences. If this CPU can be clocked further, I will post findings as soon as I acquire more oscillators.

I think it would be great to see some results from other hardware! Anything is welcome, though a 68060 equipped A4000 would be of particular interest to me. Screenmode alters the results somewhat... as do certain sw patches. ;-) Simply download the archive, make sure you're in 640x512x32 screenmode, and run bustest with "chip" and "fast" tags. ("rom" test is kind of worthless, since it is usually mirrored anyway.)

A few predictions:

Blizz 1260 cards (at least) will return better "fast" benchmarks at the same clock, since memory and CPU are synchronous. (My Apollo card runs at 1/2 divider to achieve <50MHz overclock.)

Some "big-box" accelerators take advantage of EDO, this would be a cool benchmark to compare.

A4000 machines will have generally faster chipmem access?

Maybe this particular benchmark is worthless??? Comments welcome.

[Damion]

Just for kicks, a few SysInfo results:

Disk read with an "IDE Fix Express":

50MHz: 3.38 MB/s
64MHz: 3.81 MB/s
70MHz: 4.13 MB/s
73MHz: 4.29 MB/s
80MHz: 4.39 MB/s

Mips @ 50 MHz 39.06, 62.42 @ 80MHz.

(I actually like SysSpeed best for more "in depth" CPU/disk benchmarks, but don't have it installed yet. Maybe I'll start another silly thread about it someday. :lol:)

[ChaosLord]

I have a bunch of timing tests that Team Chaos ppl did for me years ago on various hardware.  But they are all 640x512x256 colors.

[adolescent]

A4000 w/CSMK3 68060/50MHz.  I'll see if I can hook up the A1200 with Blizzard or my A3000 with CSMK2 68060/66MHz in a week or so.

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $08718000  readw      64.2 ns   normal      31.2 * 10^6 byte/s
fast      $08718000  readl     107.8 ns   normal      37.1 * 10^6 byte/s
fast      $08718000  readm     108.9 ns   normal      36.7 * 10^6 byte/s
fast      $08718000  writew     72.3 ns   normal      27.7 * 10^6 byte/s
fast      $08718000  writel    144.9 ns   normal      27.6 * 10^6 byte/s
fast      $08718000  writem    150.2 ns   normal      26.6 * 10^6 byte/s
chip      $00130000  readw    1005.6 ns   normal       2.0 * 10^6 byte/s
chip      $00130000  readl    1006.3 ns   normal       4.0 * 10^6 byte/s
chip      $00130000  readm    1005.7 ns   normal       4.0 * 10^6 byte/s
chip      $00130000  writew    612.7 ns   normal       3.3 * 10^6 byte/s
chip      $00130000  writel    612.9 ns   normal       6.5 * 10^6 byte/s
chip      $00130000  writem    614.4 ns   normal       6.5 * 10^6 byte/s

[fiat1100d]

This is the output of BusSpeed on my A4000T with Apollo 4060 @ 100 MHz, I have a Picasso IV but set the screen to 640x512 32 colors for comparison with the other tests.

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $08348000  readw      47.0 ns   normal      42.5 * 10^6 byte/s
fast      $08348000  readl      81.0 ns   normal      49.4 * 10^6 byte/s
fast      $08348000  readm      82.7 ns   normal      48.4 * 10^6 byte/s
fast      $08348000  writew     61.5 ns   normal      32.5 * 10^6 byte/s
fast      $08348000  writel    122.5 ns   normal      32.7 * 10^6 byte/s
fast      $08348000  writem    125.0 ns   normal      32.0 * 10^6 byte/s
chip      $00050000  readw     633.8 ns   normal       3.2 * 10^6 byte/s
chip      $00050000  readl     632.4 ns   normal       6.3 * 10^6 byte/s
chip      $00050000  readm     632.5 ns   normal       6.3 * 10^6 byte/s
chip      $00050000  writew    596.1 ns   normal       3.4 * 10^6 byte/s
chip      $00050000  writel    596.1 ns   normal       6.7 * 10^6 byte/s
chip      $00050000  writem    596.3 ns   normal       6.7 * 10^6 byte/s

[Damion]

Thanks for the responses so far! Initially, I was attempting to discover why Starstruck ran with such jerky motion and relatively low framerate during its opening sequence (though every other demo runs just fine here), and figured it might be fun to compare benchmarks with other hardware configurations, looking for some kind of obvious disparity on my end.

A4000T with Apollo 4060 @ 100 MHz

:-o

Nice. Is it relatively stable at that speed?

[Damion]

-D- wrote:
Just for kicks, a few SysInfo results:

Disk read with an "IDE Fix Express":

50MHz: 3.38 MB/s
64MHz: 3.81 MB/s
70MHz: 4.13 MB/s
73MHz: 4.29 MB/s
80MHz: 4.39 MB/s

Mips @ 50 MHz 39.06, 62.42 @ 80MHz.

(I actually like SysSpeed best for more "in depth" CPU/disk benchmarks, but don't have it installed yet. Maybe I'll start another silly thread about it someday. :lol:)

WOOT! I just tried Piru's "SpeedyIDE" module with RemAPollo, and gained almost 1 MB/s increase in read speed, from 4.39 to 5.2 MB/s!

After the previous tweaking I expected maybe a small improvement (if it even worked at all with RemAPollo), but this certainly exceeded my expectations. Quite a huge increase (considering the hardware) if you ask me -- plus hit the theoretical maximum of the IDE-Fix Express with an 060, which is pretty sweet.

YOU ROCK PIRU! :pint:

[Bamiga2002]

SpeedyIDE was good up 'til OS 3.1 but it didn't work with OS 3.9 anymore. I think it has to do with the new scsi.device. Don't need it though, BlizzPPC scsi is faster anyway :-).

[buzz]

The framerate is not amazing in this demo on 060 amiga. The demo compo entry was made on winuae, and it is much faster. I doubt you can get similar speeds on a real amiga.

[Damion]

Bamiga2002 wrote:
SpeedyIDE was good up 'til OS 3.1 but it didn't work with OS 3.9 anymore. I think it has to do with the new scsi.device. Don't need it though, BlizzPPC scsi is faster anyway :-).

Interestingly, scsi.device from 3.9 slows things down quite a bit on my system, and idefix immediately kills device access. (Seems to be something with my CF cards.) I had been using scsi.device from 3.1 ROM, which worked great. I was pretty shocked to see such a large gain with SpeedyIDE, especially since the theoretical maximum of the IDE-Fix Express is said to only be reachable on certain 030 cards.

I agree Blizz SCSI is pretty cool, but all I need this thing for is demos and a little experimentation, so fast 68k is ideal for me.

Buzz wrote:
The framerate is not amazing in this demo on 060 amiga. The demo compo entry was made on winuae, and it is much faster. I doubt you can get similar speeds on a real amiga.

Yeah, I thought that might be the case. The same "jerkiness" and low framerate I'm seeing in the opening sequence is even slightly visible on a fast WinUAE machine, so must be the way it is. :shrug: I guess not everything can run like the second half of Hostyle Takeover. ;-)

 

[fiat1100d]

-D- wrote:

:-o

Nice. Is it relatively stable at that speed?

Well, the system can work for hours before freezing, or it can work for few seconds... I wouldn't define it "stable" but I think it depends on the Apollo hardware being not so good... An idea I have is to build another voltage regulator using a single IC which delivers more current and has higher efficiency (the Apollo one is pretty much copied from a schematic on the 68060 User's Manual and it is rated for 3.3V @ 2 A, while @ 100 MHz the 68060 may need more current).