A530 and WHDLOAD

[CrashMidnick]

The picture :

http://img340.imageshack.us/i/photolt.jpg

Some values are changing...

[Lockon]

Here goes mine...
Your A530 is still clocked at 50Mhz ?

[CrashMidnick]

Yes still clocked @50Mhz but I have made tests @40Mhz as it is socketed, same results anyway.

[Wepl]

Originally Posted by CrashMidnick

The picture :

http://img340.imageshack.us/i/photolt.jpg

Some values are changing...

press left mouse button if screen is filled, and attach the benchmark.ilbm created in the working directory, directly here...

[CrashMidnick]

Ok will do it tomorrow evening

[CrashMidnick]

Here are my files (one @40Mhz the other @50Mhz)

[Wepl]

at least compared to a a2000 with gforce30 the chip memory access is significant slower (half speed on writes and 35% slower on reads)

but better would be to compare it to a plain a500/a2000 without any accelerator, but I have no chart for this.

if the gvp530 really slows down the chipmem access this would explain the problems encountered

[JHawk]

Originally Posted by Wepl

but better would be to compare it to a plain a500/a2000 without any accelerator, but I have no chart for this.

I will try this later on the HD8+ with a 68000.

[CrashMidnick]

Thanks a lot for your help mates ! As far as I know there is no slowdown with HD+8 or stock A2000 with only fastram... Lockon is right, there is something wrong in the gvp A530 architecture and we unfortunately do not know what are the main functions of the service jumpers...

[JHawk]

Not sure that the speed slave runs on a 68000.

I got:

Exception "Illegal instruction" ($E010) at $9F9208 (Slave $68A0) occurred.

[CrashMidnick]

Here is a bustest picture. Look at the chipram access... So f****** slow :

[Wepl]

Originally Posted by JHawk

Not sure that the speed slave runs on a 68000.

I got:

Exception "Illegal instruction" ($E010) at $9F9208 (Slave $68A0) occurred.

sorry yes, it uses bitfield instructions only available on 68020+ and therefore won't run on a 68000

[Damion]

Originally Posted by CrashMidnick

Here is a bustest picture. Look at the chipram access... So f****** slow :

That's terrible. It's been a while, but I don't recall mine being that slow (around 2.3 - 2.5 MB/s IIRC). I think I'm going to drag the 500/530 out of the closet and join the party.

[Lockon]

Originally Posted by CrashMidnick

Here is a bustest picture. Look at the chipram access... So f****** slow :

You tested running @ 50Mhz ?
Your FASTRAM performs some 27% better than my system which is clocked at 40MHz, nothing strange here really.

But then, CHIPRAM tests really puzzles...

I took some shots showing full A530 setup (Bustest_68030)and stripped down A500 (Bustest_68000) which is essentially a A500+ with 2 Megs of CHIP only.

Without jumping into final conclusions, this clearly starts backing up previous Wepl's clue on A530 hogging CHIPRAM access.

[CrashMidnick]

Yes, 50 mhz... Will try at 40 mhz to see what happens.

I will check the GVP ram utility as well to see what it shows.


@Lockon : could you tell me what kind of chipRAM expansion you have : Megachip/mini megi chip, expansion slot card (A502?), Chip Ram hack (1mb chipram soldered over the other 1MB soldered on the Mother board) ?

Mine is a A502 so it is slow chipRam... It could explains better results for you

@Damion : You are more than welcome to join us

[Lockon]

It's a trapdoor expansion, too. Kinda no name, I guess...no TM or brand silkscreen usually associated with it. Not sure about slow Chip RAM
If you have A500 and A502, without motherboard hacks, then it's perhaps Pseudo FAST(or Ranger) RAM.

Try another batch of tests without that piece of hardware.

[rkauer]

Only the A500+ can use trapdoor RAM as chip RAM without any mobo hack. If this is the case, don't worry. In fact, the CHIP RAM is the most valuable RAM on the system.

Without it you'll not have video nor audio, nor applications that call this specific kind of RAM.

[Lockon]

Well said.

I just removed my additional 1Meg CHIP RAM form trapdoor and run 'bustest' again.
Access times and bandwidth figures are still consistent, even in case of test being executed using 'upper' quarter of ChipRAM ($180000).

IMHO, CHIP memory performance for plain A500 expressed by 'bustest' application should read something like 1.7 - 2.3 - 2.9 - 1.7 - 2.3 - 3.0. At least that's something verified with cycle exact emulated A500 in WinUAE.

[CrashMidnick]

Same results here with or without 1MB expansion board.

I tried at 40MHZ and even at 25MHZ, it is worse on my system. Lockon, maybe you have faster chipram which could explain better results for you. I know that A500+ were delivered with 60ns and 70ns chipram.

Unfortunately I think there is no solution, I do not fear to say that it is not what I have expected for this accelerator. However in 90% of cases, it is better than the floppy drive and well, it is good enough.

Thanks everybody for the time you spent on this issue (which is not really an issue )

[Damion]

Finally got my A500 up and running again - what fun! The 3.1 EPROM probably wasn't burned correctly and went bad just sitting there. Anyhow, the first thing I did was pull out my old floppy collection and fired up Desert Dream - wow, it's still fantastic!

Anyhow, my results so far have been exactly consistent with Lockon's. (And, the bustest without the A530 activated is precisely the same as what's reported in WinUAE "cycle exact" mode.) Unless we're all missing something obvious, It does seem the hardware wasn't optimized, as my TekMagic in the A2K has much better chipram access (better than the A500 on its own, in fact). I also found a few threads about this in the old usenet archive, going all the way back to '92/'93 when the hardware was first sold.

Maybe a minor frequency adjustment will help the hardware sync and provide better results. Going to try with a 36 and 43MHz oscillator (if I can find them) and see what happens.