Dear Elmer:
On x87 code, it is slower than a Duron, FOOL!!!!!
Even when optimized for SSE2, it barely beats out an equivalently clocked Athlon. If the Athlon is optimized properly, it can not do even that. See Tim Wilkins Primordia. In QMC low (the scores posted on JC), it gets run over by Athlon 1.32GHz PC143 SDR. And what will it do against Nvidea's Dual DDR PC2100 or Micron's Scimitar on those heavy bandwidth apps you seem to like? When doing heavy multitasking work (Sysmark2000 (even here it has optimization assists)), it gets blown away by 1 GHz Athlons without DDR. When on a 760 with DDR at 1.2 GHz Athlon, it gets crushed Big Time.
All around, the 1.2 GHz Athlon on 760 DDR PC2100 CAS 2.5,3,3 smokes P4 1.5GHz on i850 Dual PC800 RDRAM. What happens when PC2100 CAS 2,2,2, PC2600 (166 MHz DDR), or even PC3200 (200 MHz DDR) arrives? Micron is building those speeds as I post to the 3D GPU market (they even have a PC4000 (250 MHz DDR) in the works). They claim that PC2600 will be out Q1. Ali, Micron, and Nvidea claim that their Athlon chipsets will support it in Q1. AMD could get their 760 and 760MP up on it as well and their CPUs may already be able to support such a speed (claims have been made that some Durons already will run just fine on 133 DDR (PC2100) and Tbird will run just fine on 166 DDR (PC2600) with appropriate multipliers). And we have yet to talk about Palomino or the fact that 760 supports up to 4GB of DDR SDRAM whereas P4 has to do with a mere 1GB (just think of what happens when a simulation needs 1.1GB for a working set? Yeah, page swapping is a real drag, not to mention that as memory amounts rise, RDRAM systems slow down and (DDR)SDRAM systems don't).
As to integer scores, just look at rc5 (ROL (and for that matter most shifts and rotates) is used by many communication programs (speeds up (inter)net, HAH!). P4 is slower than any other CPU. Extrapolating from the 1 GHz P3 score, a 733 MHz Celeron would outrun it possibly even a 700 MHz one. A lowly 600 MHz Duron would also send it to the cleaners. Ah to be wasted by the low end of the competitors offerings. Shame on Intel. The new catch phrase will be BAP4W4 or BAP4WF (Buy a P4, What for? (a hood ornament?)).
No, the tests used were x87 tests and code with lots of shifts, rotates, and branches. Here a P4 is smoked by its own relatives (even a lowly Celeron at half the clock).
Pete |
