66mhz 060's

[shoggoth]

I've seen rev. 6 (mask 71E41J) '060:s running as fast as 100Mhz in the Atari Falcon using a normal fan.

My 060 (full version, but labelled as EC, mask 01F43G) runs at to 72Mhz without problems. 80Mhz caused cache problems. I use a normal 486-type fan.

Seems like some 060s can run at twice their rated clockspeed, and some can't.

-- PeP

[shoggoth]

Forgot to mention; I use thermal tape (the white kind) to attach the heatsink to my 060.

-- PeP

[shoggoth]

Tape can be used to attach the heatsink to the CPU.
Paste can't, but is probably better.

-- PeP (well, a.k.a. Shoggoth apparently)

[shoggoth]

I don't really get your post here. I hope I am not offending you with my comments, I can assure you that this is not my intention.

The FPU in the 060 is (insanly much) faster than the 68881/68882. A few FPU instructions are missing, but this can easily be handled with very little slowdown.

> But I was planning on using a simple "LC040" + the 882.
> Something like 40 mhz 040 and a 50mhz 68882.

You can't use a 68881/68882 with a 040 or a 060. Well, you could use it in memorymapped mode, but that's not much point since it is very very slow. Even if you could use in co-processor mode (which you can't), it is so much slower than the built in FPU.

>Most of the 060's made, all the 50mhz versions seem to have
>the FPU, while more 66mhz versions and 75mhz versions DONT
>seem to have the FPU. At least the "available" ones dont.

Rev 6 (mask 1E41J) is marked as 50Mhz, but runs quite happily at 80,90 or even 100Mhz. It has a built in FPU.

> Remake your own 060 card, and use either a PPC's FPU,
> or a coldfire's internal FPU.

You can't use a PPC or coldfire as an FPU for the 060. You can use them as separate CPU's, but they won't execute FPU instructions for the 060.

> Remake the card, put a socket for a 68882 @50mhz in and
> try and find one of the latest revisions. Overclock that
> thing to 75 mhz or even 100mhz. Supposedly, and this is a
> BIG SUPPOSEDLY, they really did make 100mhz 882's and
> stamped them with "50"

... or use a 060, because it will usually be faster even if you run it at a lower frequency.

If you're serious about this, I suggest you do something similar to the CT60 accellerator for the Atari Falcon. It uses SDRAM, allowing it to operate at 100Mhz (provided that you have a Rev. 6 060).

Or, use a Coldfire, but there are quite a few differences between the 68k and the coldfire, which might cause a lot of compatibility problems (=the software effort needed is quite big).

Sorry for comparing your post to an old hoax on the Atariscene. I'm a bit triggerhappy, I guess.

Also, sorry for posting info about Ataristuff in this forum, but I thought it was relevant in this case. I am planning to get my first Amiga soon, since I want to see this side of life too.

-- PeP

[shoggoth]

Hi there,

> I don't want to get into an argument on this, so I'll
> simply say this. I was "brainstorming out some ideas"
> which I had based on a pair of articles I had read either
> online or from a book.

Well, I understand that you're brainstorming. My intention is to contribute to your project, whatever it may be, by giving you some hints & tips.

>One of the articles exclusively explains.... well it's >titled: 68040+68882 apparently some early form of >accelerators actually used those two chips together.
>But as I tried in vain to find the article and came up >short.....

As I said, you can use a 68882 in memory mapped mode, but not in coprocessor mode. If you code in assembly language, which I suspect you don't (judging by your previous posts, my appologies if I'm wrong), you know that accessing an FPU in memorymapped mode is very slow.

>Many of the software I have come across personally has 882 >code, and thus going by another written article, I'm told >the 040 has to work harder to process this 882 code. (it >reverts to software libraries to process the code. the slow >down hence is what I was focusing on)

Unimplemented instructions has to be emulated using other instructions. While this adds some overhead, this is not necessarily slower.

When designing the 040, Motorola studied how FPU instructions where used & choose to implement any instruction that made up more than 1% of the FPU code base.
This means that the missing instructions are used (very) rarely, and thus the emulation of those instructions causes very little overhead.

>An aquaintance of mine from New Zealand had this to say
>upon my asking in relation to your comment....
you would just patch the OS to use the extern FPU ...
any hardcoded direct usage within a program would vector to the illegal intruction TrapCall and be vectored to the FPU as well
so technically patching any FPU usage from the CPU at all would revector to the extern FPU

Yeah, of course you can do that. This is how missing FPU instructions are handled on a 040/060, except that in your case you will have to do it on _every_ FPU instruction (which would definitely slow things down to a crawl).

and as the 882 runs Asynch from the CPU you would be able to run a triple-pipeline of operations as long as the FPU ops dont get out of sync to the MCU ops
actually the 040 and 060 keep the cp protocol
they simply dont go external for the internal mapped logic IDs
you CAN use upto 4 external FPUs without problem if you really want
you have to map them as "extra" cps

This is not entirely true. For example, FPU 2, 3, 4 has no idea of the register context of FPU 1 (and vice versa). This is not a problem if the code takes this into account, but unfortunately this is not the case since noone expects more than 1 FPU in the system. There is no workaround for that.

Also, I think the dude is referring to the coprocessor protocol of the 020 & 030, rather than the 040 & 060, since what he describes simply isn't possible on these CPUs (unless the FPUs are memory mapped, which is ... well... stupid, in that case).

Furthermore, the FPU in the 040 & 060 runs in parallel with the integer unit. This means that while the FPU is busy calculating stuff, the CPU can carry on with other stuff (provided that you organise your code properly - which people do).

>In any event, I was only testing the waters of some ideas
>out, and don't claim to be an 060 expert by a long shot.

Yep, and as I said I am just helping you to get on the right track. Wouldn't want you spending lots of time designing stuff that won't work!

regards & good luck

-- PeP