AMD to revive Math co-processors?

[Artic_Kid]

DailyTech says "AMD is interested in reviving the Math Co-processor" for handling floating-point operations. Math co-processors had been used back in the early days of the Intel 8086 micro-processors, but were later integrated onto the main processor chip. It looks like AMD is interested in bringing the idea back, specifically for multi-core processors. My guess is that they want to move the math processing off the CPU chip, so it can be shared by, say, four CPUs. That way each CPU core doesn't need its own math unit. That makes it easier to cram more cores onto the main CPU chip.

It sounds like a good idea to me. Though sometimes these special-purpose co-processors need the software to be written for them (or optimized for them) in order to work best.

I suppose this approach also has the advantage that if OEMs want more math power, then can add more math-co-processor chips. Sounds flexible.

Sounds like AMD wants to hang the math co-processor onto their hyper-transport bus, for greatest speed. There's a lot to like about the idea.

[sawyen]

Would be wonderful for the work im doing..

[DavidM]

Either that, or is it going to be a component on the cpu - rather like a dedicated core?

[Couger]

sounds interesting im gonna read into this more

[aidanjt]

I thought the FPU superceeded the math co-processor to reduce latency and still have a dedicated unit for crunching math. Also our general purpose processors have a heavily underused SIMD instruction sets that can work on massively complex math while the CPU goes off doing other things..

It might not be a bad thing either, but I don't see a purpose for it yet for anything other than scientific applications and running out physics equasions in games, however there's talk of physics cards which would be more specialised for that kind of thing.