I do not want to be argumentative
Nor I. Education is my primary goal - of myself and others.
DDR 166 or PC2700 does not run at 333MHz and DDR 133 or PC2100 does not run at 266MHz. The clock speeds are 400/3 or 133 1/3 MHz and 500/3 MHz or 166 2/3 MHz. (Like you stated. I never intended to be that precise. Thought that 166MHz would have sufficed. Even Fatburger did not slam me for that, in which he in my opinion is the Slam-King. Three cheers for FB!)
While you are technically correct on the fact that DDR effective speeds are not actual cycle speeds, they are in actuallity, specified in MHz. Look at your own muskin link it states "PC2700=333MHz FSB." This is simply common usage. People commonly refer to the AXP's FSB as 266MHz when it is in fact clocked at 133MHz with DDR technology. The same is true of PC800 RDRAM (400Mhz DDR) and P4's 400MHz FSB (100MHz QDR).
In addition, while some "purists" would like us to refer to PC2100 and PC2700 modules as DDR133/PC266 and DDR166/PC333, the DDRxxx rating refers not to the MHz clock rate of the module, but to the data bandwidth of the RAM chips in Mb/s/pin. So, in effect, your usage of the terms "DDR 133" and "DDR 166" are incorrect. It just so happens that DDR333 (333Mb/s/pin) chips run at 166MHz (333MHz effective). DDR333 chips are placed in sets of eight (non ecc/parity) on PC2700 modules to achieve 2666MB/s (333Mb/s * 8bytes).
The memory manufacturers decided to modify the original rating of DDR200 chip base PC200 modules to PC1600 to combat Rambus' rating of PC800 RDRAM. This is a political issue and the reason why I generally use both the chip rating (DDRxxx) interchangeably with the module rating (PCxxxx) when referring to DDR SDRAM.
As for actual vs. theoretical; I don’t put a lot of faith in only using Sisoftware Sandra or any single benchmark...
...I like the article over at Ace’s Hardware which has a display of average bandwidth in MB/s. (Ace utilized these figures using STREAM and Cachmen.) When looking at these you will notice that the Asus KT266A board is showing only 784MBs/s of the available 2666.667MBs/s.
I usually don't either, but in this case, we were talking about the capablities of the RAM modules themselves and their efficiency in conjunction with the rest of the system. While artificial, Sandra 2002 seems to perform this measurement better than any other. As has always been my arguement, the software, chipset and the rest of the system can cause bottlenecks. This is why I would like to see truely equivalent DDR/RDRAM platforms for the P4.
Look at your <A HREF="http://www.aceshardware.com/read.jsp?id=45000279" target="_new">Ace Hardware</A> figures again - both the PX4266 and i845D PC2100 based chipsets offer 2133MB/s bandwidth and the dual-channel i850 PC800 based chipset offers 3200MB/s.
DDR SDRAM:
Tyan P4X266 PC2100 did 1020/2100MB/s = 49%
ASUS i845D PC2100 did 1016/2100MB/s = 48%
RDRAM:
ASUS i850 PC800 did 1320/3200MB/s = 41%
ASUS i850 PC1066 did 1741/4266MB/s = 41%
It would still seem that DDR is more efficient than RDRAM on P4 platforms - even when they run asynchronously to the FSB which gives them a significant disadvantage.
I thought a thought, but the thought I thought wasn't the thought I thought I had thought.
Facebook
Twitter
Instagram