I don't understand, are you two advocating Itanium instead of AMD64 or just general ramblings?

I wrote down some ideas of mine, including that I'm disappointed EPIC architecture not only did not break through, but it's still unknown to most computer fans. Me too I don't want to spend money and I don't want to wander without support, moreover Itanium is still a bit far from the ideas I have in mind; but it's much better than what we have now and I didn't like at all the way it went unnoticed.
High price could have been a must for a new technology, maybe Intel and friends thought to introduce it to high end markets first, create interest on it and then spread it on low end markets. By the way, this is how PC came to us.
So this shouldn't have kept EPIC from breaking through, nor should have a slow/complicated emulation: it should have been a temporary "bridge", not necessarily well done. Consider also that AMD introduced its idea well after seeing that EPIC was not flying. My point is: the problem is that the market (that is, us) is very different from what it was in the 80's. And not in a better way.
In general I agree with f0dder, but I don't understand the hardware/software emulation problem: nowadays the edges are blending, modern Pentiums are RISC machines spending their time interpreting x86 language with micro-code. I can say they are "software emulators", and Transmeta processors show this very well. I didn't know first Itaniums emulated x86 on hardware, if they did it more slowly than on software, maybe Intel should invent better designs...

To return to the thread topic, I agree that paging and segmentation are redundant; choosing segmentation as the issue to throw away seems to me straightforward. If we have to stick with x86 assembly, the changes they made for long mode look to me very reasonable. They don't take away anything from multitasking or OS construction in general, they just simplify a little bit the big mess we have come to.

I don't have any Itanium experience, unfortunately, so I can't advocate the processor... I can, however, advocate the idea of doing something clean instead of dragging along the burden of legacy.


The problem with hardware emulation is that it takes up transistors, either for hardcoded logic or for storing microcode (and I'm sure a hardware guy like Maverick can phrase it more precisely ), and when (in the case of Iatnium-1) it can be beaten with software emulation, then it's even worse.

And yes, for quite a while you could say that x86 has been "emulated" as well, since the complex x86 instructions are broken down by the x86 frontend... I wonder how much more efficient the current x86 chips would be if we could program the execution engines directly? Ie., no out-of-order, direct access to the RISCy instructions, all of the hardware registers instead of shadowed/register-renaming, et cetera...

When we are here at the toppic "X86-64 is bad", than I want to add this question:

Why did they add the PML4-table? Is this nessesary???
My computations came to the following result:

PML4 has a size of 4kByte.
All PDP-tables would use 512*4kB = 2MB
All PDirectories would use 512*512*4kB = 1GB
All PTables would use 512*512*512*4kB = 512GB

To use all possible Paging entrys you would need a physical and virtual Memory of ~ 513 GB only for the tables. And with the PML4-extension u can use an amount of 2^36 pages, that is:
68719476740 pages.
You could manage:
256 TB of memory

Will we ever reach this state or will there be any borders by hardware?

Correct me in the case I'm wrong