68040 FPU malfunction

[Iznougoud]

Having messed around with Amithlon some (which, by the way, was great fun, the best emulator yet in my opinion), I fired up my old A2000 - and still, I don't only feel more comfortable with it, it actually feels faster than Amithlon - which of course it isn't.

This raised an old issue though; using Xopa's speed calculation I always get error messages concerning the FPU. I've found no other way to confirm this, or prove Xopa wrong, and really would like to diagnose wether there's a hardware failure or not. Does anyone here know of a certain way to either confirm or rule this out?

The setup used being:

A2000 (B2000, rev 6)
Phase 5 2040 (68040/40Mhz)
Picasso II+ (2Mb VRAM)
Digitizer (VLab, Macrosystems - and yes, I'm showing off here. I rarely use it)
1Mb ChipRAM, 128Mb FastRAM
4.2Gb HDD
CD-ROM
ROM 3.1 and AmigaOS3.9

[Charlie]

Not direct answers, but may help:
Didn't most of these use recycled Mac CPU's?
-Maybe your's is a 68040LC..? or at least a full '040 'binned' as an LC for having a defective FPU. What's the code on the chip?
If memory serves isn't the '040 FPU the most marginal part with regard to heat?
-Maybe better airflow will reduce the errors? An occ problem with that card I believe.

Unreleated question:
You're running 3.9?
My '060 A1200 runs like a huge pile of poo under 3.9 ( slow ) Thinking of going to 3.1... Is this your experience or am I doing something wrong?
( yes, using P5's latests libs )
Ran great with 2.1 so I don't think it's me unfairly comparing it to Amithlon.

[Iznougoud]

Originally Posted by Charlie

Not direct answers, but may help:
Didn't most of these use recycled Mac CPU's?

Yes. I'm quite confident it's a recycled CPU. Haven't heard of any of these boards with new CPUs. It seems to do the job in Imagine and ImageFX both though, so there definitely is an FPU present. And should be, according to the documentation.

Unreleated question:
You're running 3.9?
My '060 A1200 runs like a huge pile of poo under 3.9 ( slow ) Thinking of going to 3.1... Is this your experience or am I doing something wrong?
( yes, using P5's latests libs )
Ran great with 2.1 so I don't think it's me unfairly comparing it to Amithlon.

I honestly wouldn't know about that. The A2000 was equipped with 3.1 ROMs when I got it some years back, and I've experienced no problems whatsoever with AOS3.9 - on the contrary, I've come to like it better all the time (and I'm conservative by nature, so I'm not an easy audience in that respect). Then again, running AOS3.9 with 3.0 ROMs would require softkicking 3.1 before starting up 3.9, which might be one (remotely) plausible explanation. I, for one, would go for 3.1 ROMs if running 3.9, regardless. But that's just me. (And I assume you mean 3.0, not 2.1 - or perhaps 3.1).

[Charlie]

Thanks - FYI my A1200 has 3.1 rom's but had WB2.1 ( no really! ) installed.
Ran damn fast!
Being an eye-candy junkie I used WinUAE to image the drive & installed 3.9.

[Iznougoud]

Originally Posted by Charlie

Thanks - FYI my A1200 has 3.1 rom's but had WB2.1 ( no really! ) installed.
Ran damn fast!
Being an eye-candy junkie I used WinUAE to image the drive & installed 3.9.

My mistake entirely. It was your mentioning of "thinking of going to 3.1" that threw me off. I don't know much about the rest of your hardware setup, but given it's packing a 68060, I take it you have decent amounts of RAM and probably a GFX-board as well. So there's really no reason for it lagging behind in any respect. Perhaps trying a clean install would be an option? I'm suspicious about mixing emulators such as WinUAE with the "real thing".

[Zetr0]

@Charlie,

I think you might prefer OS 3.5

if you have RTG then maybe OS3.9 but if you dont then i would sware os3.5 its nice / clean and works quite well with RTG cards too

os3.9 is bloatware in comparrison.

[Charlie]

Originally Posted by Zetr0

@Charlie,

I think you might prefer OS 3.5

if you have RTG then maybe OS3.9 but if you dont then i would sware os3.5 its nice / clean and works quite well with RTG cards too

os3.9 is bloatware in comparrison.

Thanks - I'm rapidly coming to that opinion too.
No RTG for wedgie miggy... but as you know I have a plan.
( fingers crossed that my eBay-sniping proggie does it's job )

@Iznougoud:
Sorry - I think I just hijacked your thread. I hope you get a helpful answer!

[Iznougoud]

Originally Posted by Charlie

@Iznougoud:
Sorry - I think I just hijacked your thread. I hope you get a helpful answer!

Well. You've given me quite a few helpful answers in the past, so.. My thread is your thread.. Anytime

[Zetr0]

@Iznougoud,

I am probly grasping at straws here, but have you changed the maths libraries ?

there maybe an error with them so might try replacing them....

heres a nice little site with some good info

http://www.youngmonkey.ca/nose/artic...ngPerformance/

I hope it helps sir.

[Iznougoud]

Originally Posted by Zetr0

@Iznougoud,

I am probly grasping at straws here, but have you changed the maths libraries ?

Although sound advice, I find it a somewhat unlikely cause in this particular case. Still. It can't hurt. So I might as well have a go at it.. Then again, the workings of libs has always to some extent been a mystery to me. Isn't the 68040.library supposed to take care of any incompatibilites in that respect? Or are you simply referring to corrupt library-files?

[Zetr0]

The original 68040.library from commodore (os 1.3 - 3.1) is flawed and doesn`t work and ANY 040 accelerator i know of *and this includes their own!* I am unaware if OS3.9 has this 040.library patched, its concerning as most 040/060 accelerators come with an install disk, and 68040.library is one of the few librarys it replaces.. thusly making it proprietary.... hmmmm

Without the library changes, taking this on an A1200 where i have lot more experience, most common errors are inconsistent system halts or hangs...

heres a linkie for some drivers for your board.. I might be mistaken, so check the manual... but the 060 libs etc should work on the 040 version...

[Iznougoud]

Originally Posted by Zetr0

heres a linkie for some drivers for your board.. I might be mistaken, so check the manual... but the 060 libs etc should work on the 040 version...

The link must have slipped away into cyberheaven somehow, but my guess is that it pointed to the "unofficial support page of Phase 5". I downloaded the 68060 install disk (taking my chances that it'll work with 68040 as well), and from the looks of it, matters have gotten no worse. Xopa still reports errors on integer and floating both though. So I guess my quest for some kind of software that can provide me with a second opinion is still on..

Anyway. I do appreciate your efforts.

(By the way, Xopa reports the same errors in WinUAE when using a similar configuration, but I suppose that doesn't necessarily mean anything)

[Zetr0]

hmmm try this one

http://amiga.resource.cx/exp/search....&base=dec&pid=

[Iznougoud]

Originally Posted by Zetr0

hmmm try this one

http://amiga.resource.cx/exp/search....&base=dec&pid=

Thanks. But I think I've installed whatever "new" libraries avilable now..

Guess I'll have to keep on looking for that illusive piece of software that can do a thorough checkup. Whatever its name may. Wherever it is. Provided, of course, it is at all.

[Zetr0]

you could just try imagine 3.1 fp version... this will only work with a present *and functional* FPU... you will notice errors when you try to render a scene..

kinda of a visual way of checking the fpu... but its all i can think of... surely there must be some proggy on aminet that tests the fpu instructions...

[Iznougoud]

Originally Posted by Zetr0

surely there must be some proggy on aminet that tests the fpu instructions...

My thoughts exactly. The only version I have of Imagine however, is 2.0. Pity really, since my present setup (provided it's fully operational) would have been considered a killer system at the release of Imagine 3.1.

Heh. Turns out the A2000 won't boot into Workbench after I replaced the libraries (never did a cold reboot yesterday). Guess I was wrong in assuming the 68060 installation script couldn't do any harm. Ah well, YARL (Yet Another Replacement of Libraries)

Edit: Here's a thought, anyone using an Amiga with a 68881 or 68882 (preferrably a 68040 with a 68882), I'd very much appreciate if he/she could run Xopa on it (speed test; integer and floating) to see if the same errors occur. If they do, there's a good chance I've got a hardware failure on my hands.

[Iznougoud]

Things are getting more odd all the time. Have found a few CPU-benchmarking and configuration programs, and quite a few has identified my MMU as "alien". Now, is it only me or doesn't that sound a bit strange?

[Zetr0]

intrueging..... very....

seen any ufo's lately.....

hmmm what mask and serial is the 040 on your board ? is it an xx68ECxxxx or xx6800xxxx

some EC's dont have a fully working MMU or FPU.... motorolla decided in their wisodom that chips that DONT have a fully working or damaged FPU/MMU they would sell off as EC based packages... however when the demand for EC based chips whent up (since they were a lot cheaper than the other Moto-offerings) they ended up just producing chips and branding them as EC without testing them... clever huh?

i have a theory that you know this already...

[Iznougoud]

Originally Posted by Zetr0

i have a theory that you know this already...

Not really, no. But I can't say I'm surprised. The Phase 5 2060 boards, however, was to my knowledge all equipped with recycled Mac-CPUs (which in itself should be cause for suspicion), so I guess anything is possible.

I havent looked at the mask of the CPU, since it would require removal of the heatsink, and I don't have access to whatever kind of "glueish" thermal paste that keeps it in place. But from the looks of it, I better get me some.

Perhaps the simplest solution would be to just replace the blasted thing with a 68060/50 - but then again, I find it hard to argue spending roughly £80 on a machine that old - particulary considering I'd have to spend another £50 - £60 on a NIC - provided I even manage to find one.

Another possible solution would be replacing it with another 68040, which is far cheaper and require no soldering, but the only 68040s I can find are 33Mhz ones, which would mean a speed decrease AND probably a change of the crystal, unless I'm prepared to run them overclocked. And I can't imagine 68040s overclocking very well..

(The exact information was, by the way, "MMU status = Alien").

[Zetr0]

@Iznougoud

see thats presumption for you... lurks arround a corner just to smack me in the chops lol!

It does sound cool though "MMU = Alien" but there hardly any Workbench programs that actulay can use the MMU infact if i am not mistaken workbench cannot use the MMU... i think linux debian 3 can though....

truth is, it would be nice to have it working properly though.

040's can clock if you get a later revision die / mask of, Alexh - Jope and a few others would know better in this area to be honest

but as long as you add decent active cooling then all should be good after all its only 7mhz over specification i think most 40mhz were infact overclocked 33's

I have recently been discussing 060's with Binox, I am getting some of these, but also a few 040's too... I am sure i can help you in sourcing an 040 (not to sure what these are but i have been assured that there are a couple of 33mhz ones in the bag)

so whan i get them if you are still interested i will mention it