Talk:Coprocessor

coprocessor may not be a general-purpose processor in its own right. Coprocessors cannot fetch instructions from memory

I'm not sure about this statement. i've heard the term coprocessor used for many units that *can* fetch instructions, it's just they're not optimal for (or capable of) general purpose processing — Preceding unsigned comment added by 81.153.147.173 (talk) 21:05, 17 May 2016 (UTC)

How CO Processor Works.. Whether it read All instructiuons given to CPU or CPU will instruct this if any FPU operations Come...

--I added some information about a new unit called the PhysX, which I believe functions like a co-processor in that it offloads physics computations from the CPU. - fredtorrey

Cart before horse

Somehow I doubt that there was no demand for floating point calculations before the coprocessor came along - but rather once integration got cheap enough, there was no more money to be made by having separate FP-less and FP versions of processors. The article hadsdss a bit of cart before horse feeling about it now. --Wtshymanski 23:26, 9 August 2006 (UTC)

Minor discrepancy

The 8088 actually can't trap FPU opcodes when no coprocessor is present (in fact, it can't trap invalid opcodes at all) - this is only possible with 286+. This necessitated building the floating point routines directly into the application (or even having separate binaries), as no interrupt trap emulation could be done. - bigbadbrad at gmail —Preceding unsigned comment added by 99.165.105.105 (talk) 03:54, 5 February 2008 (UTC)

CPU is not a processor

I think that the article assumes that CPU = processor, but that isn't true. Citation: "Sometimes an entire computer is made out of coprocessors instead of a single CPU." I think this is a whole nonsense, as CPU is a term of von Neumann architecture - a piece of a computer which is often implemented by a microprocessor. I assume if the computer consists of two so called coprocessors, so they are CPU or even can be called processors. (I don't get it: How something can be a "co"processor if there's nothing in there to what this "co" item would assist in operation.) AleyCZ (talk) 18:04, 6 March 2008 (UTC)

I removed the offending statement. In the case of the Emotion Engine, the VPUs (coprocessors) receives instructions from the MIPS core. It is impossible for the Emotion Engine to run on the VPUs alone. All coprocessors cannot operate independently without a host - this includes GPUs and FPGA-based accelerators like the SGI RASC. If anyone can prove that an Intel 8087 can operate without an 8086 (or any other coprocessor), they get a cookie! Rilak (talk) 08:42, 8 June 2008 (UTC)

Well you could do something like the did to make a calculator IC into a CPU... They used feedback, but that's not exactly a practical CPU in the sense that even a PIC's Harvard architecture would be. Even to just do something as essential as branching would require quite a hack. Needless to say, a standard 8086/8087 assembler would be just about useless since you would really need to go ahead and invent a new macro-instruction-set that uses the 8087 for the microcodes. Again, why bother? JWhiteheadcc (talk) 06:59, 16 May 2009 (UTC)

Who did what to which calculator IC? I don't know of any calculator that uses a coprocessor but not a CPU, since coprocessors tend to be more expensive. Rilak (talk) 07:21, 16 May 2009 (UTC)

Not all coprocessors are expensive. Of course, I don't think an 8087 would cost any more than an 8088 anymore, assuming you could find a pair with the board they came on.  ;) Of course the calculator chip wasn't strictly a coprocessor, but at least some of the coprocessors are more than capable of being forced to operate as a more generic state machine (at great reduction in efficiency or with a ton of external components=cheating). I believe the way you would implement branching, is to do something like copying a value from one register to the "address" buss based on multiplication by a result added to some base address. Hackish but likely doable!

JWhiteheadcc (talk) 19:48, 21 December 2009 (UTC)

Overlaps with Asymetric Multiprocessing

Some sections seem to overlap with the Asymmetric multiprocessing article, as the scope goes beyond math coprocessors. I'd suggest removing/merging these sections into that article and adding a link to it. Mantique (talk) 22:23, 28 May 2009 (UTC)

Pro-Intel segment

I removed the following from the article, which adds nothing to the page whatsoever, "While consumers may have resented paying for a processor that was essentially wasted when the upgrade was installed, this marketing strategy did allow increased volume of sales of 80486 family processors, thereby accelerating the eventual price reductions." This seems like a terrible explanation, which could be used to justify any attempt to rip off consumers. 65.29.31.3 (talk) 15:16, 15 August 2010 (UTC)

Pro-Bigfoot segment

I have removed:

"Tests have shown it indeed does take an appreciable load off the processor which would otherwise have to run the XP or Vista Network Stack. However its high price tag (over 250 USD) has kept it out of reach for all but the most performance-minded users."

A quick internet search reveals that many reviews show inconclusive benefits to using the card. Thus this paragraph shouldn't be included unless a citation is explicitly provided- and in any case, I don't think that promoting the performance benefits of this particular product serves its purpose in this article (providing an example of a coprocessor)

I don't even know if this truly counts as a coprocessor. It's got its own memory space and instruction stream. Surely a coprocessor has to be more tightly integrated with the CPU than an I/O device? We don't call a UART a coprocessor even though the bit-banging approach would burn up lots of CPU time. ( One chuckles at the thought of a modern multicore Pentium being forced to do the stuff we made old-time Z80's and 6502's do - scan keyboards, step disk drives, read the joystick position, squeak the speakers; a modern processor is loafing compared to the 8-bit days.) --Wtshymanski (talk) 18:19, 31 March 2011 (UTC)