Color Classic won't boot with FPU installed

[johnklos]

Later m68882 do exist, but 2019? Also, 16 MHz doesn't seem plausible at all. Anything made after 1990 is going to be of a mask that'll run at 40 MHz minimum, so who would want or sell 16 MHz FPUs?

Hopefully you find a good one in the UK.

 

[jimbojones]

Just to close this out, I got a 40Mhz one from a seller in the UK and it works perfectly. Seems there are lots of fakes from China on ebay. Seems mad considering how little they must make!

 

[cheesestraws]

Good to hear

 

[dan.dem]

@jimbojones Are you saying that a MC68882 labeled as 40 MHz works with 16? Interesting.

16 MHz doesn't seem plausible at all. Anything made after 1990 is going to be of a mask that'll run at 40 MHz minimum, so who would want or sell 16 MHz FPUs?

I bought and installed a FPU advertised as 16 MHz in 1992 (roughly ~ 100 US$ then) into my original LC (very similar to a CC). I don't know about other manufacturers at that time but speaking about Macs the majority (if not all?) that could be retrofitted with a FPU had 16 to 25 MHz clock speed. Actually I would be happy if my 68882 would work with 20 MHz, I could transplant it into a IIsi.

About speed gains: If memory serves me, there are relatively few ones.

My main goal then was to make a relatively cheap 68020 machine more Mac II-compatible. I mostly succeeded since many Mac II-applications demanded a FPU, and were happy when they found one on the LC.

Some scientific software came both in FPU- and SANE-version (SANE is the set of software calls or functions - never sure what is what - for floating point operations). The FPU versions turned out to be about twice as fast as the SANE-versions. This was still a bit of disappointment for me, since a FPU call should be many multiple times faster than going the long way through a SANE call. And I remember how much faster a 8087 equipped IBM PC was compared to a plain one with a similar application type.

Many other software showed no speed gain at all, probably - I guess - as floating point arithmetic is usually avoided as much as possible because it is so slow. I'm, quite sure JPG compression is solely integer based.

 

[johnklos]

Are you saying that a MC68882 labeled as 40 MHz works with 16? Interesting.

Faster marked CPUs will always run at lower speeds.

The performance difference between an 8088 / 8086 doing floating point and floating point on an 8087 is significant in part because floating point performance on the 8086 / 8088 is really bad. It's basically a souped up 8 bit processor that's register starved. Doing floating point in software on an m68000 is much faster, and faster still on an m68020, so the speedup of an m68881 is not nearly as significant as the speedup of an 8087.

 

[dan.dem]

@johnklos ad 1: I am aware this is used in modern processors tweaking their clock speed, but did not know this works with old processors too. I always thought they are phase optimized for a frequency on a deep hardware level.

But what makes me wonder is why this isn't true for many RAM modules. At least modern (DDR3) faster labeled DIMMs "usually"/"often"/(or just) "in my case" don't work with slower boards. But there may be different reasons why it didn't work (at least voltage was ok), and I don't have any deeper understanding of RAM-technology (unlike some brilliant folk here who are designing SIMM sticks). But we probably stop here for avoiding to become totally off topic.

ad 2: Yes, this should be the main reason for the disappointing speed gain. And I can imagine that the 68020's little instructions cache makes a noticable difference as long as the data to be processed is already in one of its registers. Thank you for pointing this out.

 

[johnklos]

Modern RAM speed is specific because accesses can't happen too soon and shouldn't happen too long after data is ready. The speeds have to do with clocking of the data in and out of memory sequentially after the first location is accessed. Old RAM used to just care about overall speed  - if your computer requires 80 nanosecond or faster, then that's the only measurement that matters, for the most part (this is before EDO).

Completely static chip designs can run at any speed, or can even be held in a stopped state. Chip designs with dynamic components (registers, caches, whatever, implemented using capacitors) can only be clocked as slow as the safe minimum speed, but since all Motorola m68020 / m68030 / m68881 / m68882 chips sold at speeds from 16 MHz (or 12 MHz, for the early '020) to 50 MHz, all chips will run safely at 16 MHz without issues.

 

[dan.dem]

I'm late replying, sorry @johnklos.

Translating this to my very limited understanding of RAM: The specific speed of modern DDRn RAM is determined by the RAM-internal time management necessary to bring the data to the in-out-pins [and additionally handling refresh cycles?]. Is this a reasonable interpretation?

About my 68882 I bought in 1992: So it may even be rated for higher speeds. I'll look that up when I'm burying it out of the storage boxes pile.

 

[johnklos]

Modern DDR memory runs communication to the computer's bus at many gigahertz (whatever the memory is rated for), but in reality dynamic RAM is still not all that much faster than it ever has been. Memory access time in the late 1980s was typically in the range of 60 to 100 nanoseconds. The fastest DDR4 memory in the world can only manage access in around 8 nanoseconds.

That represents a speed increase of eight to ten times faster, while density has increased ten thousand fold. This is why DDR exists - you pipeline reads and writes so that you're waiting that full 8 nanoseconds as infrequently as possible. At 4 GHz, your cycle time is .25 nanoseconds, so waiting for memory could take up to 32 clock cycles!

To answer your question, the specific speed is the speed that the memory controller and memory on the DIMM can pipeline continuous operations.

 

[dan.dem]

To answer your question, the specific speed is the speed that the memory controller and memory on the DIMM can pipeline continuous operations.

Thank you @johnklos for making this more clear to me.

And yes, I still remember buying 70ns RAM for Macs.