Not having the "Performance Crown" hurting Intel?

[joe emo]

I think they need to raise the production level soon. It'd be good to have a second Dresden but financially it'd kill them.

Fabs are an extremely costly venture, and keeping them up to date is nearly as financially challanging. Intel's budget for Fab R&D is astronomical...

 

[Akumajou]

Although to many it sounds good that on paper the AMD-64 is backwards compatible with X86-32 everything, would'nt it have made better sense to have introduced a completely new CPU architecture that would have been 64 bit only and represent a new standard while letting other software developers make "virtual 32bit" program emulators?

Didn't Intel try that?

I was under the impression from Intel that Itanium series was just being aimed at a server, enterprise and workstation market, not the desktop market because the Pentium 4 would only get enhanced to the point that the netburst architecture would have its multi stage pipelines fed constantly to provide performance and increase clock speeds later leaving something like a P5 as a cpu that would probably show up much much later.

Now this is just me asking but why would AMD think that Itanium series would be eventually slapped into the desktop market and therefore need to pre-emtively strike with its version of a X86-64. Then again Bush did assume Sadam had nukes...

What I mean is wouldn't AMD have been able to make a core to be compatible with whatever, if ever Intel went for an Itanium 4 desktop edition?

 

[Fox5]

Although to many it sounds good that on paper the AMD-64 is backwards compatible with X86-32 everything, would'nt it have made better sense to have introduced a completely new CPU architecture that would have been 64 bit only and represent a new standard while letting other software developers make "virtual 32bit" program emulators?

Didn't Intel try that?

I was under the impression from Intel that Itanium series was just being aimed at a server, enterprise and workstation market, not the desktop market because the Pentium 4 would only get enhanced to the point that the netburst architecture would have its multi stage pipelines fed constantly to provide performance and increase clock speeds later leaving something like a P5 as a cpu that would probably show up much much later.

Now this is just me asking but why would AMD think that Itanium series would be eventually slapped into the desktop market and therefore need to pre-emtively strike with its version of a X86-64. Then again Bush did assume Sadam had nukes...

What I mean is wouldn't AMD have been able to make a core to be compatible with whatever, if ever Intel went for an Itanium 4 desktop edition?

Maybe AMD was more in competition with non x86 designs than with Intel.
The memory limit of 32 bit processors was quickly approaching. If that 2GB limit(or 4GB?) then what happens? Does everything start switching to Mac? Do they make some kind of bridge chip to get around the problem? Some people may consider x86-64 a bit hacky, but a bridge chip would be even more so.
64 bit makes AMD much more attractive in the server market, and was something to hold over intel.
And I'd assume it allowed amd to mold the ISA a little to how they want their future processors to be, rather than whatever intel's vision is.

And as far as keeping x86 instead of going with a new design...
Well, it keeps them from sacrificing a lot of performance suddenly, and allows companies/people to use existing programs. X86 is supposed to be really hard to emulate too, yet having hardware support for it costs practically nothing, except for whatever headaches that occur from having to program for x86 or design around its limitations.(transistor budget to support it is small though)

 

[pcchen]

Maybe AMD was more in competition with non x86 designs than with Intel.
The memory limit of 32 bit processors was quickly approaching. If that 2GB limit(or 4GB?) then what happens? Does everything start switching to Mac? Do they make some kind of bridge chip to get around the problem? Some people may consider x86-64 a bit hacky, but a bridge chip would be even more so.

There is an extension called PAE introduced in P6 (Pentium Pro) which allows 36 bits memory addressing. However, it's not a flat memory space (of course) and it is cumbersome. Some versions of Windows support it and there are Win32 API for it. So, if x86-64 does not exist, we can still use more than 4GB memory, but that will not be pretty.

]64 bit makes AMD much more attractive in the server market, and was something to hold over intel.
And I'd assume it allowed amd to mold the ISA a little to how they want their future processors to be, rather than whatever intel's vision is.

The problem of Intel is probably about timing. Intel wants Itanium to eat server markets. Now they performs fairly well at high-end server, but low end servers was also the primary target. Intel probably delayed their own 64 bits extension to x86 to avoid 64 bits Xeons eroding Itanium's market. And when x86-64 starting to gain momentum, it's too late for Intel to introduce their own extension. Using IA-64 (ISA of Itanium) is simply not a viable option.

]And as far as keeping x86 instead of going with a new design...
Well, it keeps them from sacrificing a lot of performance suddenly, and allows companies/people to use existing programs. X86 is supposed to be really hard to emulate too, yet having hardware support for it costs practically nothing, except for whatever headaches that occur from having to program for x86 or design around its limitations.(transistor budget to support it is small though)

Actually, x86-64 is a different ISA from x86. They are similar in many ways, but they are different. You can't run a x86 program directly in long mode (64 bits mode). They are not compatible. Therefore, in theory, x86-64 can be a completely different ISA, as all programs need to use long mode still require recompliation. Of course, it costs more to support two completely different ISAs on one CPU, while a x86-64 decoder can share a lot with a x86 decoder.

On the other hand, it was rumored that AMD wanted to use a completely new FP instruction set for x86-64 to replace the very old x87. It would be RISC-like three operand ISA. However, later AMD decided on using SSE and SSE2 as the standard FP ISA for x86-64.

 

[Fox5]

On the other hand, it was rumored that AMD wanted to use a completely new FP instruction set for x86-64 to replace the very old x87. It would be RISC-like three operand ISA. However, later AMD decided on using SSE and SSE2 as the standard FP ISA for x86-64.

x87 was removed from the architecture though, wasn't it?
I had heard that was the one thing athlon xp had over athlon 64.

 

[pcchen]

Although you can't use x87 in long mode, but x87 is still supported in 32 bits mode. Otherwise many 32 bits programs won't work anymore.

 

[CI]

I think they need to raise the production level soon. It'd be good to have a second Dresden but financially it'd kill them.

They are expanding production.
http://www.extremetech.com/article2/0,1558,1722362,00.asp

Advanced Micro Devices said Tuesday that it would add Chartered Semiconductor as a chip foundry beginning in 2006.

The decision will add another source of production to AMD's own lines, which include Fab 32 and the upcoming Fab 36, both in Dresden, Germany
...

 

[T2k]

I think they need to raise the production level soon. It'd be good to have a second Dresden but financially it'd kill them.

They are expanding production.
http://www.extremetech.com/article2/0,1558,1722362,00.asp

Advanced Micro Devices said Tuesday that it would add Chartered Semiconductor as a chip foundry beginning in 2006.

The decision will add another source of production to AMD's own lines, which include Fab 32 and the upcoming Fab 36, both in Dresden, Germany
...

Thanks. They're getting ready to the dual core and Turion needs... 8)