More thoughts on the Paris. SInce it is a socket 754 chip and Athlon 64s using S754 will still be available, the motherboards for K7-replacement Paris will also have to handle Athlon 64.
This means that the K7-replacement Paris chip is using a socket with AMD64 capability, it will be plugged into motherboards that are AMD64 compatible and, according to several leaks, it will be based on the K8 core.
There is no cost savings associtated with removing 64-bit capability from the chip. It is possible that the reason it is "castrated" is so that it can run some yet-unknown Intel extensions to x86, which, for lack of a better word, I call SSE64.
SSE64 would essentially be an extension of SSE and SSE2 to give more 64-bit data processing functionality, but would not change memory addressing. Intel could then claim they have a "64-bit" CPU, even though it addresses only 31 bits at a time of memory in Windows. A slew of benchmarks would beat AMD unless the AMD chip were run under a 64-bit OS, which might not yet be available. Another reason Intel would take this route is that it would not damage their Itanium program at all -- enterprise servers need 64-bit addressability and SSE64 would not provide it.
All this is conjecture so far.
The problem for AMD is that many people couldn't care less about adding more than 2G of memory to their PCs, but they like the "64-bit capability" that Intel provides because it will speed up a few games, encoding and decoding programs and encrytption stuff. Big problem.
So I further conjecture that AMD knows the details of SSE64 and wants to implement it to match Intel. (An alternative would be to ignore it and force people to buy Windows XP 64 to get the same performance.) Anyhow, I suspect that since the K8 core already has enough registers and added 64-bit capable ALUs (K8 core does have additional 64-bit execution hardware), the K8 core could be modified to implement this hypothetical SSE64 but that would make it incompatible with AMD64.
Hence, the "castrated" K8 -- it will not do AMD64, but will implement this "SSE64." The other major change to the Athlon 64 is that Paris will only have 256K L2, but with integrated memory controller I expect it will easily outperform the old Athlon XP core at the same frequencies.
Another reason I am thinking along these lines is that Paris does not appear until mid 2004 at the earliest, but simply dropping the cache to 256K could be accomplished way before that. It should not take 6 months to reduce the cache and disable 64-bit, if thats the only changes. Also, there is a mid-2004 modifaction of the Prescott core scheduled by Intel, in which it will start using a new LGA775 socket. I suspect that is the first Intel chip that will have their poor-man's version of "64-bit extensions" that I call, "SSE64."
But wait, AMD could have a couple other tricks up their sleeves. One thing AMD could definitely offer customers who buy Paris-based systems is complete 64-bit upgradeability. By replacing the Paris CPU (probably called XP-?) with an Athlon 64, they immediately have a 64-bit system capable of running Windows XP 64 in 64-bit mode.
Another possibility is that the Paris CPU itself could be switchable between an "Intel mode" that is compatible with Intel's so-called 64-bit extensions and an "AMD mode" that would be compatible with AMD64. A special code purchased over the Internet could unlock the AMD64 mode in the BIOS. Voila! A 64-bit system.
I don't think Intel will ever make an incompatible TRUE 64-bit extension to x86. They may elect to stick with an SSE-like extension as I've outlined until they feel Itanium is ready to take over the desktop. (PS -- it will never be ready!)
But adding a few 64-bit instructions (add, sub, multiply/divide with 128 bit product/dividend, multiply then add, multiply then subtract, etc.) -- and then calling their CPU 64-bit seems exactly to be the kind of thing Intel would do to try and screw AMD.
Summary of this post:
1. Paris chip will fit into 64-bit motherboards and is a modification of the 64-bit K8 core.
2. CONJECTURE: The reason for disabling 64-bit on Paris is so that it can implement some future Intel x86 extensions that do NOT involve memory addressing but do require 64-bit registers and hardware. I refer to these extensions as SSE64.
3. It will take AMD at least 6 months to make changes to the K8 core to implement SSE64.
4. It is entirely possible that the Paris chip can be switched between two modes -- an Intel mode that is compatible with 32-bit OS's and Intel's SSE64, and the AMD64 mode that has full 64-bit data and addressing, and is compatible with either 32-bit OS's or 64-bit OS's.
5. WIth respect to (4) the possiblility of a purchased code being entered into the BIOS to enable AMD64 mode.
6. AMD will actively encourage Paris owners to upgrade their systems to 64-bit by replacing the CPU when Windows XP 64 retail is available.
7. Even if SSE64 is not the reason for K8 castration (i.e., disabling 64-bit), but AMD is doing it strictly to segment its products and pricing, the possibility still exists of AMD realizing revenue by upgrading using method (5) or (6).
Comments?
Petz
