Math CoProcessors

[DamienC]

I have a few FPU questions.

First, after reading a post here, I know that a FPU that's clocked higher will work in a system with a slower CPU, i.e. a 387SX-25 will work fine in a 386SX-16 system. Does this work the other way? For example, would a 387SX-16 work in a 386SX-25 system? Logic tells me that it won't work, but I wanted to make absolutely sure.

Also, would a Math CoProcessor provide any significant performance boosts to an AMD 386DX-40 system? I know programs that specifically utilize them (like spreadsheets and whatnot) will benefit, but will games, for example, run noticeably faster?

 

[mbbrutman]

Before the 486DX era, a co-processor was an unusual option to find in a machine unless you had a piece of software that explicitly called for it. The 486DX changed all of that by integrating an FPU onto the chip, and the 486SX confused everybody by taking it out.

You can expect that almost none of your games or DOS software from before the Pentium era will use floating point. And even for a few years after that it would be unusual because integer arithmetic is still far faster.

 

[Dave Farquhar]

I know that a FPU that's clocked higher will work in a system with a slower CPU, i.e. a 387SX-25 will work fine in a 386SX-16 system. Does this work the other way? For example, would a 387SX-16 work in a 386SX-25 system? Logic tells me that it won't work, but I wanted to make absolutely sure.

It depends on the motherboard. Some motherboards did allow you to clock the CPU at one speed and the FPU at another speed.

Also, would a Math CoProcessor provide any significant performance boosts to an AMD 386DX-40 system? I know programs that specifically utilize them (like spreadsheets and whatnot) will benefit, but will games, for example, run noticeably faster?

I recall very few games utilizing the FPU. In the 486 era, I remember the people with DXs teasing the people who had SXs, but as one of those guys with an SX, I never had DX envy. There were some graphics utilities that benefitted from the math coprocessor, but today, if I'm re-sizing images or anything like that, I'm going to do it on a modern PC anyway.

 

[Unknown_K]

A few old flight SIMs will use the FPU but not much else in games of the era. I purchased a FPU back when the 386/40 was new and that was for doing spreadsheets and scientific calculations, not for games. These days I collect and install FPUs just to do it (and I run old scientific apps for kicks once in a while).

 

[Captain Midnight]

In general, the answer is yes. The concept of a faster CPU and slower FPU isn't really that uncommon. For example, take the Compaq Portable III, which uses a 12.5 Mhz 80286, but only uses an 8 Mhz 80287 co-processor by design, since the co-pro doesn't take advantage of the clock multiplication that the main CPU does. Having tested the co-pro with AutoCAD, I can tell you that it works; however, it does cause some minor processing delays because of the processing difference. However, given the overall system configuration, the delay really isn't too noticeable or that bad.

As another example, consider the 1986/1987 series of 386DX computers, which used 80287s as co-processors because Intel hadn't finalized the 80387's design prior to the release of the main processor. I've not used a machine with that configuration, but I've been told you can really "feel" the co-processor's lag as it does calculation in, say, AutoCAD or 123. Plus fundamentally, the 80287 is an inferior design to the 80387, but that's another ball of wax.

So in other words, provided you are using a co-processor that's compatible with mainboard pin-wise, you should be able to run it slower than the main processor. Note that I make this statement because I'm in the 386 and higher forum. If we were talking 8088s, that's a different ballgame.

Hope that helps.

 

[deadcrickets]

Before the 486DX era, a co-processor was an unusual option to find in a machine unless you had a piece of software that explicitly called for it. The 486DX changed all of that by integrating an FPU onto the chip, and the 486SX confused everybody by taking it out.

You can expect that almost none of your games or DOS software from before the Pentium era will use floating point. And even for a few years after that it would be unusual because integer arithmetic is still far faster.

Actually it was only disabled due to error testing. Intel didn't want to just trash the chips so they disabled and re-binned them. By the time FPU became common-place games were already switching direct to MMX.

 

[Chuck(G)]

What most people fail to understand is the complexity of the 8087 in relation to the 8086/8088. The 8087 was far more complicated, with wider data paths. Because of the peculiar way Intel implemented the coprocessor linkage, the 8087 had to match the 8086/8088 in the way instructions were decoded (remember the instruction cache?). So even when the 8087 wasn't computing, it was mirroring the 8086/8088 operation when it came to interpreting instructions. And, the 8087 could be switched between 8 and 16-bit data buses, something the 8086 could never do.

I seem to remember that the 8087 used a ROM with ternary values to conserve space.

Technically, it was a real wonder.