Petz, <anandtech's numbers were done on pre-release 3-RIMM i820 board.>
So were Tom's. His review was made on October 5th, 1999, well before the 2 RIMM motherboards were released.
<Intel had to slow the memory interface to get it stable>
No they didn't, at least not to solve the 3-RIMM problem. What Intel did was limit the number of open banks, but that limit was already put in place before the 3-RIMM problem.
The only change they made between September 29 and the actual release date of mid-November was to take away one of the RIMM slots. That has nothing to do with the speed of the memory interface. In fact, because of a peculiarity in the Rambus protocol, a two-RIMM motherboard performs slightly better than a three-RIMM board.
<And anyhow, with the SysMark 98 business benchmark, RDRAM/i820 was only 2.4% higher than PC100/BX, much less than the performance increase from PC100 to PC133 on the Athlon.>
Funny how the same jump from PC100 to PC133 also decreased the performance of Quake. But I guess you can attribute that to Via's poor AGP implementation.
By the way, I've noticed a certain dualism regarding RDRAM. People complain that Intel would have never achieved the high SPECmark numbers with Coppermine without the use of RDRAM (among other things). They say that comparing Coppermine on RDRAM to Athlon on SDRAM isn't fair. Yet I've noticed people, probably the same ones, also poking fun at RDRAM because it allegedly fails to outperform SDRAM in independent benchmarking.
You can't have it both ways, you know. Either RDRAM demonstrates no performance advantage over SDRAM, or RDRAM gives Coppermine an unfair advantage over Athlon. Which is it?
Tenchusatsu |
