ACA500 tested

[Photon]

So the game is very tight at filling the DMA and CPU slots. I haven't seen it in any other game, but few games really pushed the limits like Turrican did. I guess the only way to fix it is to update the slave, but I'm not sure at all there's some unnecessary wait or such to trim off.

ps. Thanks for making me play Turrican again :P

[BarryB]

Being one the few games I've played over and over you notice when things like this happen, but it doesn't make the game unplayable though. Glad you enjoyed playing it again

[CrashMidnick]

@BarryB : Try to add the tooltype CACHE to see if it helps for the slowdown

[BarryB]

Originally Posted by CrashMidnick

@BarryB : Try to add the tooltype CACHE to see if it helps for the slowdown

Didn't seem to make much difference, as didn't adding an Echo line to my s-s so I could CTRL-D and load Turrican without anything loaded. I also enabled MMU in whdload.prefs as my 030 has an MMU. Pulling out the ACA1232 actually seemed to be a bit better, but maybe placebo

The slowdown also happens when you use the beam to show a hidden block and focus the beam on the block and rocks are falling at the same time, plus when many of those beetle things are on the screen and you're zapping them with the beam it slows down too.

Sorry for picking fault with this one game, need to test more games on my A1200 and then on my the ACA500+ACA1232 and see if they do a similar thing.

[Photon]

Well, if there are few to none DMA/CPU slots left, a faster CPU can't do much to change things. And ACA500 can't change ECS DMA speed, so it might be better for a 'WHDLoad on ECS' thread.

While waiting for the eminent WHDLoad slave coders to clear the foggy crystal ball, it tells me the only way to get rid of the slowdowns is a new slave with some luvin' on the interrupts, if it's at all possible.

[Schoenfeld]

Originally Posted by Photon

While waiting for the eminent WHDLoad slave coders to clear the foggy crystal ball, it tells me the only way to get rid of the slowdowns is a new slave with some luvin' on the interrupts, if it's at all possible.

Not sure if you can do much on the "slave programming" side. The ACA1232 is an async design, so it does not get the full 3.5MBytes/sec on chipram. If the game uses almost all DMA slots , then it may be impossible to reach the same speed that a sync 7MHz 68000 can achieve.

Jens

[Psygore]

The main difference between the whd and floppy version is that a lot of blitter waits have been inserted into the whd game. That's why it can have some slowdowns on some machines.
Putting the game stack in the fastmem could help it. Wepl had done a new slave some years ago but it was never released. You should ask him.

[CrashMidnick]

Can someone do a bustest with ACA500 + ACA 123* and posts his results ? I would like to see the chipram access speed.

Thanks.

[BarryB]

@Jens: Just tried the ACA1232 in my A1200 and Turrican takes over twice as long to load, no noticeable slowdowns in game, but noticeably longer 'pauses' in game when I assume it's loading stuff so I guess that's the slow chipram access? What is the actual access speed to chipram on the ACA1232? Will this problem be addressed in future accelerators or is this 'as good as it gets' ?

@CrashMidnick: What program will do this? Sysinfo? I'll test mine when I know the program to use!

[CrashMidnick]

BUSTEST

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

[BarryB]

@Photon/Psygore: I've emailed Wepl about that Turrican slave, so will wait to see if he replies.

@CrashMidnick: Any specific parameters?

I used the chip option 'bustest chip' that does 256k of chip and these are the results:

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
chip      $00090000  readw    2194.5 ns   normal       0.9 * 10^6 byte/s
chip      $00090000  readl    3743.3 ns   normal       1.1 * 10^6 byte/s
chip      $00090000  readm    2907.2 ns   normal       1.4 * 10^6 byte/s
chip      $00090000  writew   1485.4 ns   normal       1.3 * 10^6 byte/s
chip      $00090000  writel   2959.9 ns   normal       1.4 * 10^6 byte/s
chip      $00090000  writem   2967.9 ns   normal       1.3 * 10^6 byte/s

[CrashMidnick]

Thanks BarryB !

As you can see, chipram accesses are quite slow, this (maybe) explains your slowdowns...

Here is the bustest of a blizzard 1230 on A1200 to compare :

http://hschid.files.wordpress.com/20...-blizzard1.jpg

[BarryB]

Result using FAST CHIP ROM options from my ACA500+ACA1232:

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $08248000  readw     126.5 ns   normal      15.8 * 10^6 byte/s
fast      $08248000  readl     154.7 ns   normal      25.9 * 10^6 byte/s
fast      $08248000  readm     138.9 ns   normal      28.8 * 10^6 byte/s
fast      $08248000  writew    100.8 ns   normal      19.8 * 10^6 byte/s
fast      $08248000  writel    101.1 ns   normal      39.6 * 10^6 byte/s
fast      $08248000  writem     90.1 ns   normal      44.4 * 10^6 byte/s
chip      $00040000  readw    2188.9 ns   normal       0.9 * 10^6 byte/s
chip      $00040000  readl    3747.8 ns   normal       1.1 * 10^6 byte/s
chip      $00040000  readm    3032.1 ns   normal       1.3 * 10^6 byte/s
chip      $00040000  writew   1495.4 ns   normal       1.3 * 10^6 byte/s
chip      $00040000  writel   2993.0 ns   normal       1.3 * 10^6 byte/s
chip      $00040000  writem   2982.3 ns   normal       1.3 * 10^6 byte/s
rom       $00F80000  readw     126.2 ns   normal      15.8 * 10^6 byte/s
rom       $00F80000  readl     152.7 ns   normal      26.2 * 10^6 byte/s
rom       $00F80000  readm     139.4 ns   normal      28.7 * 10^6 byte/s

Result using FAST CHIP ROM options from my A1200+MX1230 (No MMU):

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $08338000  readw     159.8 ns   normal      12.5 * 10^6 byte/s
fast      $08338000  readl     204.2 ns   normal      19.6 * 10^6 byte/s
fast      $08338000  readm     193.5 ns   normal      20.7 * 10^6 byte/s
fast      $08338000  writew    155.6 ns   normal      12.9 * 10^6 byte/s
fast      $08338000  writel    156.3 ns   normal      25.6 * 10^6 byte/s
fast      $08338000  writem    142.3 ns   normal      28.1 * 10^6 byte/s
chip      $00068000  readw    1208.2 ns   normal       1.7 * 10^6 byte/s
chip      $00068000  readl    1209.0 ns   normal       3.3 * 10^6 byte/s
chip      $00068000  readm     881.8 ns   normal       4.5 * 10^6 byte/s
chip      $00068000  writew    849.3 ns   normal       2.4 * 10^6 byte/s
chip      $00068000  writel    856.5 ns   normal       4.7 * 10^6 byte/s
chip      $00068000  writem    843.8 ns   normal       4.7 * 10^6 byte/s
rom       $00F80000  readw     356.7 ns   normal       5.6 * 10^6 byte/s
rom       $00F80000  readl     356.2 ns   normal      11.2 * 10^6 byte/s
rom       $00F80000  readm     322.1 ns   normal      12.4 * 10^6 byte/s

Result using FAST CHIP ROM options from my A1200+ACA1232 (MMU):

BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $083E8000  readw     126.8 ns   normal      15.8 * 10^6 byte/s
fast      $083E8000  readl     153.3 ns   normal      26.1 * 10^6 byte/s
fast      $083E8000  readm     138.9 ns   normal      28.8 * 10^6 byte/s
fast      $083E8000  writew    101.4 ns   normal      19.7 * 10^6 byte/s
fast      $083E8000  writel    101.4 ns   normal      39.4 * 10^6 byte/s
fast      $083E8000  writem     90.2 ns   normal      44.3 * 10^6 byte/s
chip      $00060000  readw    1200.4 ns   normal       1.7 * 10^6 byte/s
chip      $00060000  readl    1199.2 ns   normal       3.3 * 10^6 byte/s
chip      $00060000  readm     875.4 ns   normal       4.6 * 10^6 byte/s
chip      $00060000  writew    839.0 ns   normal       2.4 * 10^6 byte/s
chip      $00060000  writel    842.6 ns   normal       4.7 * 10^6 byte/s
chip      $00060000  writem    842.9 ns   normal       4.7 * 10^6 byte/s
rom       $00F80000  readw     126.9 ns   normal      15.8 * 10^6 byte/s
rom       $00F80000  readl     152.5 ns   normal      26.2 * 10^6 byte/s
rom       $00F80000  readm     139.4 ns   normal      28.7 * 10^6 byte/s

The chip speed issue has put me off getting an ACA1232 for my A1200, what's puzzling is that bustest shows the same access speed to chip as my MX1230 yet it clearly takes longer to load Turrican?

[CrashMidnick]

Maybe because it uses Fastram as well and the ACA is faster as per your results above.

Interesting to see the ACA1232 results on A1200 and A500. The A500's bus is a real bottleneck unfortunately due to its 24bit architecture.

[BarryB]

Don't know if I put my previous post clearly enough, I meant the ACA1232 is slower at loading Turrican than my MX1230 on the A1200!!

[BarryB]

I received a reply from wepl along with the new slave, it did run better and the slowdown with the rocks is still there but didn't seem as bad. Have attached it to this post as he gave permission for me to post it here.

Think as everyone says it's a combination of chip access speed on the ACA1232 and the stuff going on in the slave on an A500 to get it running.

[esel]

Originally Posted by amiga1260

I tested the ACA 1230 card with an A500 Plus Rev. 8A1 motherboard. I use a 135 Watt PSU.

I have the same problem i only get a black screen.
testet on rev,6 and rev.8

card firmware is 0.81.

[amiga1260]

Update with the ACA 1220@25 MHz. I tested this card in my ACA 500 for a whole day at a Commodore meeting in The Netherlands and didn't crash.

When I set the 1 MB Chip on both A500 rev.6 and rev.8 with the ACA 500, I get an error message that I haven't enough Chip memory in WHDLoad 17.2. I tried to run Mortal Kombat and Sleepwalker.

Do I need to modify those motherboards?

[FOL]

Originally Posted by amiga1260

Update with the ACA 1220@25 MHz. I tested this card in my ACA 500 for a whole day at a Commodore meeting in The Netherlands and didn't crash.

When I set the 1 MB Chip on both A500 rev.6 and rev.8 with the ACA 500, I get an error message that I haven't enough Chip memory in WHDLoad 17.2. I tried to run Mortal Kombat and Sleepwalker.

Do I need to modify those motherboards?

You tried the new beta version of WHDLoad 17.3?

[BarryB]

WHDLoad 17.2 has worked for me with 1mb chip on my ACA500+ACA1232 on a Rev6 A500. Just tried Mortal Kombat and Sleepwalker from WHDownload and both run fine for me with just over 800kb chip free.

Has the ACA500 got the 0.81 f/w with chipmap option? What does WB report for Graphics Mem?