I am afraid my question might be taken as a stupid one, but I did my best to find an answer on all search engines and find nothing relevant.
My question is about FPU (floating point unit). Since early 90's FPU (math coprocessor) is integrated in the CPU.
Although new APUs (A4 ~ A10) have integrated GPU into the CPU, but the integrated GPUs aren't as strong as are the Graphic Cards (GPUs). I suppose the same for FPU.
Is there any FPU enhancement available for today's CPUs (similar to PCI-e Graphic Cards e.g. Nvidia/Radeon) ??
Since I work mostly with numeric computations, I am fed up of the CPU time for the calculations, (I have FX 6300 at home, and Xeon 5450 at workplace).
I have seen the specification of Xeon Phi, and Xeon E series, but they are extremely expensive, I am looking for cheaper alternatives (e.g. FPU card that can be simply plugged in a PCI-e).
There is the AVX instruction set (Advanced Vector Extensions) that are supported by current Intel CPU and AMD CPUs. It is used for floating point calculations.
Intel's up coming Haswell CPUs will have AVX2 which basically has 256-bit integer vectors and FMA (Fused Multiply-Add). Haswell is capable of up to 32 single-precision and 16 double-precision floating point operations per core; which double what Sandy Bridge and Ivy Bridge can achieve.
While the gigaflops is not even close to what graphic cards are capable of, it is going in the right direction for CPU; up. People which just need to wait and see what the actual performance level of Haswell CPUs will be.