The right CoPro for me...

[chalackd]

Hi guys, I've got a couple of systems around here that I'm tempted to add the *87 co-processors to. I don't really have any experience with these chips though, so maybe you guys can help me out with some really basic questions.

One of the systems is an old Compaq 386, not entirely sure if I want to drop one into this one as the cases were badged differently for the systems with/without copro. This one seems easy enough tome though as any version of an 80387 should work fine, correct?

The one that has me wondering though is a Grid laptop that I have. It's sporting an P80C86A-2. Now, a quick epay search has provided me a list of various 8087 revisions, C8087-1, C8087-2, and C8087-3... Now, it would seem that logically I should go for the -2 revision to match the CPU, but I'm wondering if there's a real difference that's going to matter. Like if I can find a 8087-1 for super cheap would it perform properly, or will I need at least the 8087-2 or higher?

Also, I seem to remember seeing some document that implied that the 8087 is also the proper copro for an 80186 CPU? I don't have anything around currently with a '186, but I'm curious anyways.

Lastly, are the 8087's rarer in the black plastic form, rather than the purple/gold ceramic package? Personally I think that the purple chips look neater, but I'd like a "matching" chip to the CPU. No basis for this last question other than there seems to be way more of the purple chips out ere on epay...

Thanks,
Dan

 

[Chuck(G)]

Much depends on how the original motherboard was designed. For -86 and -88 systems, it's safest to select an NDP that matches the CPU speed. Same for the -286 system. After that, it's a guessing game (for example, there are 386 systems that will run just fine with an 80287 at a very different clock speed).

But most important, exactly what are you going to use the NDP for? I never found much use for them, except when a customer requested that one be used.

 

[chalackd]

Interesting to know, I do remember trying to put an 80287 into the 386 machine once, I can't remember what program it was that wanted it at the time, I think it was some game... Anyways, all that happened in that instance was that the machine would refuse to boot until it was removed again.

I don't really have a specific purpose for wanting to put the NDP into these systems other than to "complete" the motherboard. Really I never really ran into much actual use for them either.

So, in what way would I go about identifying the speed of a particulate -87 chip? Is that where the revision number (-1, -2, etc) comes into play, or it it a matter of trial and error?

 

[Raven]

The numbers at the end correlate to the speeds, I believe - I don't know by memory which numbers mean what speeds, however. Wikipedia may have the answer to that in the form of a chart.

 

[retrohimpi]

But most important, exactly what are you going to use the NDP for? I never found much use for them, except when a customer requested that one be used.

In the mid-80s, we relied on a 286 with co-processor at work for scientific number crunching (mapping human disease genes to chromosomes). The software had to be specially compiled to use it, but it made the difference between getting the results tomorrow or next week (or sometimes not at all). It probably would have helped with spreadsheet recalculation speed, too, but we didn't do any of that as it was a lot slower than compiled purpose-designed software.

 

[Unknown_K]

I used a IIt 387-40 back in the day on my 386DX/40 system because I liked how fast it crunched excell spreadsheets. It would also help out in Autocadd (as would a weitek). Outside of a few apps that were big in number crunching I don't see the need for one, especially the 287 versions (which I believe can be clocked at a different speed then the CPU while the 387 needed to be the same speed or faster then the 386).

F-16 Fighting Falcon 3.0 needed a FPU back in 1991.

 

[Chuck(G)]

While the 8087/287/387 were coprocessors, the Intel nomenclature was Numeric Data Processor. I think that someone at Intel had a plan for other types of coprocessors to be used, as the mechanism used was pretty generic, but the NDP and IOP are the only Intel co-processors I know of for the x86 family. The coprocessor had to mimic the CPU instruction cache and bus cycles until an ESC instruction came along, after which the coprocessor would interpret whatever came next.

Weitek also had coprocessors for the 286, 386, etc. (the 1067, 1167, etc.) but were not instruction compatible with the x87 chips. The 1167 (386 platform) actually came out before the 80387.

NDPs were pretty mandatory if you were doing big spreadsheet work. If you've ever used SuperCalc (either for x80 or x86), that's my code in there twiddling the numbers.

I've seen a few mentions of NDPs for the high-end Pentiums to handle high-precision decimal math (very important for financial calculations), all done in an FPGA.

 

[Raven]

In the mid-80s, we relied on a 286 with co-processor at work for scientific number crunching (mapping human disease genes to chromosomes). The software had to be specially compiled to use it, but it made the difference between getting the results tomorrow or next week (or sometimes not at all). It probably would have helped with spreadsheet recalculation speed, too, but we didn't do any of that as it was a lot slower than compiled purpose-designed software.

Are you saying you worked on the Human Genome Project? That's pretty damn cool. :D

 

[Neon_WA]

It's sporting an P80C86A-2.

a CMOS low-power 8086 running at 8MHz
(low power function >> runs at 0MHz in standby mode)

the speed designation on 8086, 80C86 & 8087 is as follow
no speed marking >> 5MHz
speed marking -1 >> 10MHz
speed marking -2 >> 8MHz
speed marking -4 >> 4MHz

I'm not sure if a HMOS 8087 will work with a CMOS 8086

 

[Neon_WA]

as to your other questions

the 8087 can be used for 8086, 8088, 80186 & 80188 processors
the 80C187 can be only used for 80C186
no FPU available for 80C188

intel never made plastic 8087s as they ran so hot. They were only made in purple/gold or brown ceramic packages

 

[Chuck(G)]

I'm not sure if a HMOS 8087 will work with a CMOS 8086

It will. Remember that the V20/V30 are CMOS and work just fine with the 8087. I've certainly used one with an 80C88.

Back in the rear of my skull, I seem to remember that it's even possible to use an 8087 with an 80186, with some amount of "glue".

 

[retrohimpi]

Are you saying you worked on the Human Genome Project? That's pretty damn cool. :D

Yes, but not until 1998 when I moved to the Sanger Centre, which sequenced over 1/3 of the human genome. I designed and wrote the networking software that drove their DNA sequencers, which ran 24/7, and collected up the results to be assembled into the finished sequence. It was an exciting time, and great to contribute to such a large-scale project (over 200 sequencers, monitored/controlled by a dedicated 6-CPU Unix cluster, with other clusters to do the downstream assembly). My software, written in Perl and Java, handled billions of short sequences over 10 years, keeping track of them as the barcoded plates of DNA went from the labs to the sequencers and then into the main assembly pipeline. I still have the T-shirts

Before that, I worked in a small university lab, helping the scientists to make maps of the genes on individual human chromosomes, initially using an IBM PC-AT and later a small network of Sun Unix boxes. I didn't write the actual mapping software but developed databases to hold the raw data, along with conversion scripts to extract and reformat it for use by the different mapping programs of the day and then draw out the resulting maps for publication.

And all because I bought an 8Kb home computer in 1979 and taught myself to program in BASIC and 6502 assembly language, not knowing at the time where this obscure new hobby would lead me.

 

[Chuck(G)]

In the mid-80s, we relied on a 286 with co-processor at work for scientific number crunching (mapping human disease genes to chromosomes). The software had to be specially compiled to use it, but it made the difference between getting the results tomorrow or next week (or sometimes not at all). It probably would have helped with spreadsheet recalculation speed, too, but we didn't do any of that as it was a lot slower than compiled purpose-designed software.

That's pretty curious--there were certainly faster and better choices then (e.g. some of the MicroWay or Alpha Micro products). Did FPS have an add-on box for the 286 systems back then?

 

[retrohimpi]

That's pretty curious--there were certainly faster and better choices then (e.g. some of the MicroWay or Alpha Micro products). Did FPS have an add-on box for the 286 systems back then?

It was partly money (we didn't have any grants to support the work when we started out) and partly software compatiblilty as we were part of an international consortium (CEPH) and had to buy hardware that could run the software they supplied. It was written in Pascal and compilers of the day were pretty idiosynchratic so it wasn't easy to port software to different hardware platforms.

My memory is that the 287 was part of the system (a genuine IBM PC AT) but it came ready installed and I never opened up the case, so I couldn't be sure. Wikipedia says "the PC/AT supported an optional math co-processor chip, the Intel 80287, for faster execution of floating point operations." I think it was just plugged into an existing socket, right beside the 286, on the motherboard. When we outgrew this, we got time on another lab's VAX mini, but I had to use it out of hours as the number-crunching slowed their system to a crawl, then we got a grant to buy a Sun 4 of our own and moved on to analysis software that was automatic, rather than interactive, and could be left to run unattended for days or weeks to come up with the best map.

 

[Raven]

Very cool, retrohimpi. That's something that you can be proud to put on a resume, I'm sure. :D

You guys who are a generation or two older than me really don't realize how lucky you are - both being able to see home computers develop and being able to find work as those with computer knowledge were less numerous. I suppose my generation has it's benefits too - more numerous computers and cheaper, and obviously the newer technology is available, but I'd trade that stuff to be there as it began if I had a choice.