Can't confirm it, but it makes a lot of sense!

Just like you could put an AMD 386DX-40 into a board taht is max. 33 MHz.

In these days the bus speed = cpu speed.

I'm pretty sure that's right, based on my knowledge of how clock speeds work with processors, but I will admit I have virtually no experience with co-processors. I did some reading before replying and it seems that most processors have equivalent co-processors to go with them so it's best to find the exact match of co-processor for the processor you're upgrading.

Also, you may be aware that a lot of people tended to have 486/DX2 processors right around when Pentiums came about. The DX is the key here because there was also an SX which lacked the FPU and thus there were 487SX co-processors to add on.

Gemini000 wrote:

I'm pretty sure that's right, based on my knowledge of how clock speeds work with processors, but I will admit I have virtually no experience with co-processors. I did some reading before replying and it seems that most processors have equivalent co-processors to go with them so it's best to find the exact match of co-processor for the processor you're upgrading.

Also, you may be aware that a lot of people tended to have 486/DX2 processors right around when Pentiums came about. The DX is the key here because there was also an SX which lacked the FPU and thus there were 487SX co-processors to add on.

Now let's introduce the following: You have a 16MHz processor, and you adjust the RAM refresh rate cycles (lower them) in order to give an effective 19MHz processor speed. Should the Math Co be a 20MHz native speed, or should it ideally be a 16MHz native speed?

[I have a "guess" as to the answer, but I'd like a second opinion before I blab my thoughts]

The system (and thus CPU) clock speed will remain 16 MHz, no matter what RAM timings you set.

So I don't see the need for a 20 MHz COPRO.

Mau1wurf1977 wrote:

The system (and thus CPU) clock speed will remain 16 MHz, no matter what RAM timings you set.

So I don't see the need for a 20 MHz COPRO.

Well, by adjusting the RAM refresh rate on an 8 MHz 286 PC, I have achieved an effective (benchmarked) processing speed of over 9 MHz.

But, my thought was, that I would imagine the MathCo would achieve similar percentage gains from tinkering with RAM refresh, as the main processor, thus (as you concluded) making any faster MathCo superfluous and not beneficial. [EDIT: on second thought, I don't think the MathCo would see a speed benefit, as I believe its functions do not include refreshing the RAM to begin with. Thus, no load is reduced from lessening the RAM refresh rate]

I believe I read somewhere, that you can put a faster speed (than the PC's speed) math coprocessor in a PC and it works fine, but you don't want to put a slower math coprocessor in, as it slows the apps from the speed they could be running.

Of course you can put in a faster coprocessor, but it'll just run at a reduced speed. Unless you're talking about actually running it faster than the main CPU, which I've never heard of.

It used to be pretty standard for the copro to run slower than the CPU. Most 286 systems ran the 287 at 2/3 CPU speed. (It was so common, in fact, that there was a version of the 287 made with a 3/2 internal divider, to allow it to run at the full CPU speed.) A lot of 386DX/40 systems were also set up by default to run the 387 at 33mhz.

...Most 286 systems ran the 287 at 2/3 CPU speed.

Does that mean that in such systems, one could possibly benefit from installing a copro a bit faster than the main processor? (so that 2/3 of the copro came close to equalling the speed of the main processor? Or, is that ineffective without an integrated 3/2 internal divider?)

Regardless of the rated speed printed on the coprocessor, it will only run at the speed set by the motherboard. If your CPU is running at 12mhz, and the board is set to run the coprocessor at 2/3 CPU speed, then the coprocessor will only run at 8mhz. If it's a 10mhz chip, it'll run at 8mhz. If it's a 20mhz chip, it'll run at 8mhz. If its a 570ghz chip, it'll run at 8mhz.

The only way around that is to either change the divider on the motherboard (many 386 boards allow it by either a jumper, or changing an oscillator crystal), or use one of the copros with the internal divider (which was only the one series of 287, AFAIK).

On my IBM PS1/pro with a 386sx/20mhz there is installed a 25mhz fpu and it worked since many many years fine... although it is somewhat useless 🤣

keropi wrote:

On my IBM PS1/pro with a 386sx/20mhz there is installed a 25mhz fpu and it worked since many many years fine... although it is somewhat useless 🤣

If you ever try Falcon 3.0 on that machine, you'll appreciate your fpu.
😀

It's kind of funny, though. There's a few very early games that don't seem very "demanding" of resources, that nevertheless are listed at Moby Games as utilizing Math Coprocessors (e.g., Pub Pool, and Combination Lock). Pub Pool is a CGA pool game, and Combination Lock is a trivia/quiz/gameshow type game.

I read somewhat recently that Falcon 3 only uses a FPU for the most "realistic" flight model and that it is pretty buggy. Some fans of the game preferred to use one of the other flight model modes.

I always thought that if you tried putting a 16mhz coprocessor in a 20mhz motherboard you would be overclocking the coprocessor.

I would be very surprised if all of the i287 and i387 processors are not the same, and are just printed up at different clock speeds for marketing requirements. This was the case with the Motorola 68881 and 68882 floating point processors - a 16MHz labelled 68882 will quite happily run at 50MHz as they're all identical cores.

From the dim and dusty corners of my memory, I seem to recall that TFX used the fpu - at least I think the setup utility measured fpu performance. Although trying to play TFX would be ridiculous on a 386 anyway 😀

On 8088/86 systems, the FPU always runs at the same speed as the CPU.

On early 80286 systems, the FPU usually runs at either the 2/3 the speed of the CPU, or in sync. Apparently that is a built-in function of the 80287 rather than the motherboard. On later 286 boards, it was more of a function of the chipset that determined the speed of the FPU. Sometimes an extra OSC socket was provided to run the FPU at whatever speed the user desired, and sometimes there were 2 or 3 jumper selectable settings.

Early 386 designs typically followed the same design as later 286 systems, where an extra OSC socket was provided. Jumper settings would allow you to set the FPU as SYNC or ASYNC. However, one thing that is commonly not taken into consideration is that not all 387 FPUs allow ASYNC operation. To the best of my knowledge that is only possible with intel FPUs. Later 386 boards typically run the FPU in sync with the CPU with no provisions for ASYNC operation. It is under synchronous operation that you want to avoid using an FPU rated slower than your CPU.

megatron-uk wrote:

Although trying to play TFX would be ridiculous on a 386 anyway 😀

"Minimum" requirements for TFX with sound: 386/33 with 4MB RAM.

I think it would be fun seeing how well you could run TFX with the minimum system specs...but with a MathCo, of course.

But then again, I kinda like to do goofy, masochistic stuff like that. 🤐

Here's a final question from me, on this topic:

Tandy released an Intel 286-based PC, that ran at 10MHz.

However, the Tandy Support literature says to specifically use an 8MHz math coprocessor.

Anybody got an idea why they wouldn't suggest using a matching 10MHz math coprocessor?