I think if Itanium were decent, x86 could be killed pretty easily. I disagree that x86-64 made any real difference. You got more registers, but lower code density to go with it. None of the real problems with x86 were corrected; you still have mixed length instructions, overly complex instructions, too few registers (helped somewhat though) and the necessity for a decoupled architecture. This is very inefficient, particularly now with power being such a factor in processor design. The decoders kill speed because they lengthen the pipeline, and add size and power use because they take a lot of transistors to implement. On top of this, the RISC core has to be built to work with what comes out of the horrible x86 instructions, so is less efficient than a design built as a stand-alone design.

x86-64 currently does not even demonstrate any serious speed advantage over 32-bit applications, except in very limited situations. Some run a little faster, and some run a little slower. It's not a big improvement.

If Intel had just done a well thought out RISC instruction set instead of the weird EPIC based Itanium, they would have had no problems beating x86. Instead, they tried to be too smart and created a processor that currently isn't demonstrating the type of performance, in integer, that has any chance to compromise x86 sales. Why would a company with their position take such a terrible gamble anyway? Everyone would have been happy if a nice clean RISC processor replaced the vile x86 processors once and for all. Instead, Intel screwed up and we are stuck with this neurotic Itanium line and the degenerate x86 processors. What a choice.

Incidentally, Intel does not license the x86 instruction set to companies that make processors based on it (although, after a lot of litigation against AMD, which they largely lost, they did). Companies like Cyrix, Transmeta, Rise, Centaur (now VIA) all made processors without licensing the instruction set from Intel. So, it is probably safe to assume they would regardless of what instruction set was dominant. The cost of entering the business is really high though; this is why almost all the companies have abandoned their internal designs and there are only a handful of microprocessor makers left. Even AMD as the number 2 maker of processors is always in danger of losing money, and they have a pretty decent market share. So, I think the cost of entrance inhibit companies more than the instruction set does, and maintaining start of the art manufacturing as well as design to keep up with Intel is extremely difficult. Even AMD with IBM's help can't approach Intel's superlative manufacturing technology, and those two are the best after Intel. It's not an easy business to be in, and especially if Intel produces a good processor in the Conroe/Merom.

StarFox,

I agree with most of what you say, but not necessarily in the details.

x86 is truly horrible, and the only reason no one has passed it is because of volumes and the fact everyone exited the business because of costs. A true RISC implementation shouldn't need decoding, and POWER did not when it was created. It still is a lot simpler than x86, but admittedly isn't a great implementation.

However, x86 got raped by Alpha and even PA-RISC when they were being backed aggressively, as well as MIPS. Intel announced the Itanium and pretty much everyone stopped development, except for IBM. SPARC always sucked, and was never competitive, and Sun has pretty much exited the mainstream processor business and is using Opteron for that instead, and going with their threaded monster. So, there isn't much competition anymore, with really only IBM, Intel and AMD making performance processors.

When everyone was in the market though x86 was not their equal.

Forgive me if I am putting words in your mouth, but I think what you are saying is that implementation is more important than instruction set, and if it is what you are saying, I would agree with you. It really is the most important aspect, but instruction set is still part of it.

Even looking at the x86 processors, how much better performance would we get if the instructions received by the processor were the ones that are actually executed? You would lose all those stages, and all the nasty decoder logic. So, you'd get considerably better performance and a smaller die with lower usage. So, while AMD in particular has done a great job in compensating for the terrible instruction set, nothing will ever remove the penalty associated with it. What I mean is, if AMD were able to choose their instruction set, and it sold in big volumes (allowing for the same degree of design commitment), you can be sure it would easily outperform the Athlon with it's lousy instruction set. I mention AMD only because Intel had that opportunity and completely screwed it up. It's not the first horrible processor from Intel either, I am sure people remember the awful 432 and 860. Makes you wonder what they are thinking sometimes.