wbmw,
Simply having the hardware there doesn't mean you can extract the ILP to make full use of it. Otherwise, it would have been a "no-brainer" for Intel to already add it.
I think it is much easier to extract parallelism on the FP side than on integer side, and gains should be substantial when you go from 1 to 2, unlike, say going from, say, 3 to 4, or 4 to 6 on integer side.
The designers of CPUs need to strike some balance between what can be done, and what has the best benefit / cost. Looking at the majority of PC apps out there, the cost benefit analysis for adding another FP unit probably came with "no" at the time.
As far as AMD is concerned, K8 was due for a reworked FP subsystem, but it was rejected because of time to market, and because of lack of availability of human and monetary resources. You can bet that the next major code revision will take up this project.
On Intel side, adding second FP unit to P4 would blow Itanium out of the water, eliminating its most notable advantage over competing chips. Holding back Xeon to make Itanium look better was a philosophy that governed Intel, until it came back to haunt Intel (in form of Opteron).
Furthermore, new execution engines are the second most power hungry units for a processor, right after the decode engine. I really don't think the tradeoffs are as favorable as you may think.
I agree fully. The tradeoffs were not as favorable, but may be in the near future.
Joe
News 
Market Data 
Discover 
AI
