Granite Bay vs nForce2 (memory bandwidth)

[Flakk]

I am wondering which is faster. From the simple numbers, it sounds like nForce2 is faster than Granite Bay... nForce2 having 6.4gb/sec max transfer rate with dual DDR333 memory, and Granite Bay having 4.2gb/sec with dual DDR266. That kind of irks me because I just bought an expensive Granite Bay board and now I'm wondering if that was a waste of money. Would I have just as well off, or even better off going with an nForce2 motherboard?? I plan to pump my FSB up to 166 to take advantage of Dual DDR333 on my Granite Bay board.. do you think this will match nForce2?

Is there something else I am missing in my comparison.. because if not, looking purely at the memory bandwidths for each chipset, it *seems* clear that nForce2 beats Granite Bay.

 

[AndyHui]

The bus bandwidth to the Pentium 4 matches its memory bandwidth on the Granite Bay chipset: both are 4.2GB/s.

The bus bandwidth to the Athlon processor is only 2.7GB/s with the 333MHz FSB models. The Athlon cannot take full advantage of the additional memory bandwidth offered by the nForce chipset. Essentially 2.7GB/s is wasted.

RAM <-> chipset <-> processor

RAM 4.2GB/s <-> Granite Bay <-> 4.2GB/s Pentium 4

RAM 6.4GB/s <-> nForce 2 <-> 2.7GB/s Athlon XP

See where the bottle neck and what the problem is?

 

Is that just on paper and are they close in comparison in benchmarks, or does the p4 show a large gain in memory bandwidth?

 

[Nardman]

Bandwidth limitations are always theoretical but the conclusion is not, P4 systems handle higher bandwidth memories better than athlon processors.

 

[Flakk]

Ok i get it.. but that raises another question: Why did they even bother to use dual DDR memory if it doesnt do jack anyway. 2.7gb/sec is not that much, and in fact the same as a single channel of DDR333. A completely pointless chipset?

 

[Boogak]

Originally posted by: Flakk
Ok i get it.. but that raises another question: Why did they even bother to use dual DDR memory if it doesnt do jack anyway. 2.7gb/sec is not that much, and in fact the same as a single channel of DDR333. A completely pointless chipset?

The extra bandwidth does wonders for the nForce2's onboard video.

 

[mechBgon]

The extra bandwidth also allows nForce2 the luxury of guessing what data the CPU will ask for next, and pre-fetching it. It has little to lose by guessing, considering how much bandwidth it has available to do this. Look at the blue numbers in the chart on this page for examples of where pre-fetch really hits a home run.

I wouldn't attach a huge importance to what you call "the simple numbers," because as you saw in the other GB thread, the "simple numbers" would also lead to GB's DCDDR being faster than PC1066 RDRAM, but at least in the case of my infamous WinZip 8.1 maximum-compression benchmark, it's not even close. Life is just like that... theory is always getting run over by reality.

So check out the "real-world" benchmarks that suit your usage pattern next time, and let them be your guide. Since you already have a GB board, enjoy it and wring a nice overclock out of your P4.