Peter Gerassimoff (pgerassi@ticon.net) on 2/3/03 wrote:

...

>FYI, Itanium3 (Madison) is only going to be 1.3GHz for the 6MB version and 1,2GHz
>for the 3MB version.

Have you told Intel yet? Their public statements are that Madison is going to clock at 1.5 GHz.

YOur predicted SPEC scores are 25% too high. You should multiply
>them by 80% or SPECint/fp 6MB: 1000/1720 and 3MB: 920/1600.
>
>You should also give Opteron up to 2.6 to 2.8GHz by end of this year on PC3200.

Really? Do you have a reference for this? The highest speed I've seen AMD claim for any Hammer this year is 2.6 GHz ("2.4 and 2.6 GHz in Q4", IIRC), and since that one statement I haven't seen anything beyond 2.4 GHz in Q4 (suggesting that the previous statement included 2.6 GHz as a best-case possibility which later started to look overoptimistic).

Furthermore, I've seen no claim that the on-chip memory controller can handle anything faster than PC2700 memory.

>Also the SPEC scores you cite are for 32 bit mode only.

That does seem to be the case, at least if one believes Freb Weber's quoted 1202 for 32-bit SPECint at 2 GHz. Paul has expressed skepticism that Fred's other claim (that the score will improve by 15% - 20% in 64-bit mode) will pan out, because he thinks that it will be difficult to get the 64-bit (K8-specific) compilers up to the standard set by the 32-bit Intel compiler used to obtain the 1202 score.

Adding 64 bit speed up
>of 10%

That may either be a bit conservative (based on Fred's statement) or a bit optimistic (based on current performance differences between the Intel compiler and, e.g., gcc).

and PC3200 speed up of another 10% (conservative),

Or imaginary (again: reference, please?).

2.0GHz Opteron gets 1440/1400,
>2.4GHz gets 1650/1600, 2.6GHz gets 1775/1700 and 2.8GHz gets 1900/1800.
>
>Given these Opteron has a good chance of beating Itanium3 in SPECfp

Dream on. Opteron won't even beat McKinley when Madison releases, since Opteron will still be at 2 GHz (or at most 2.2 GHz if Madison's late and the faster Opteron is early). And unless you can substantiate your claim that Madison won't run at 1.5 GHz, there's no reason to expect Opteron to be able to challenge it in SPECfp at anything less than about 3.2 - 3.4 GHz.

and Itanium3
>may fall behind many other 64 bit CPUs in SPECint

"Many" others? Not Alpha (EV7 is a bit slower than McKinley so even EV79 in 2004 likely won't quite catch Madison, and EV68 is only about 5% faster than McKinley so won't nearly match Madison). Certainly not SPARC or MIPS (they're much slower than McKinley). Not PA-RISC either (faster than SPARC and MIPS, but notably slower than McKinley). Even POWER4+ will have to bump up its clock rate significantly to match Madison (it probably can do so, but whether that will happen before Madison's release is unclear).

as Opteron runs far ahead of all of them.

That one I'll agree with.

- bill