Memory Bandwidth vs FSB

So I was wondering, been tinkering with my new C2D (lovin it btw lol )

I can get up to 3.6ghz on the cpu with some fair adjustment, but I can do it a few different ways.

1) 400x9 = 1600fsb then run the memory @ 2:3 ratio for a DDR1200

2) 450x8 = 1800 fsb memory @ 1:1 DDR1125

3) 516x7 = 2064 fsb memory @ 1:1 DDR1032

Wasn't sure which direction to go. The 400x9 seems a it more stable and I can get my memory stable at a higher frequencey and still keep my overall CPU but I wasn't sure which way would be more useful.

Have yet to do full on testing with this, just wanted some opinions and other people's thoughts

The bottom one would probably be the best although, personally, i'd do the top one for stability and actually have the memory running at DDR2-800 speeds (however, if you can use the 2:3 ratio, i don't see why not!). With option 2, if the FSB frequency is 450Mhz how will the memory run at DDR2-1125 if you're using a 1:1 ratio?

SoCalDJ said:

Have yet to do full on testing with this

Click to expand...

that will tell the truth

but, unless the memory timings are wildly different, the 3rd option is gonna win. don't underestimate the significane of fsb speed for intel

So with a higher FSB is definatly going to equate to a faster system, but I guess my next question is,

Is it possible to actually saturate a FSB that high? I mean 2000 is way over what is suggested in the NEXT line of procs with their 1333fsb. Wouldn't a straight memory bandwidth be better than higher, yet unsaturated fsb?

Edit - Oh and would running the divider really show me a loss in preformance bacause it's no longer running 1 to 1 with the fsb?

PS - You caught me mansize!!!! LOL! It was like 3am when I was trying to add, it should be a 2.5:2 ratio for the DDR1125, heh

ok, so let's say you're running the ram at 1:1 with a 516mhz fsb

the fsb is quad pumped.. so 516mhz is 2064mhz effective, which equates to 16.5gb/s of bandwidth (64bit wide bus)
memory would also be running at 516mhz, but is dual channel (128bit), double pumped. also 16.5gb/s of bandwdith

so right there, you're already saturating the bus with memory bandwidth, assuming you can peak it out. add in the bandwidth to the chipset for video card, hard drive, lan, etc for when you're playing a game....

also, faster bus = lower latency

take either 400 or 516. 516 will have a little worse latency, but you should reap more benefits from bandwidth. get the program mbench and do some testing at each. 450 will definitely lose.

Ok did a buncha testing all freakin day lol. Here's the results for all those who care lol.

I know these are all synthetic benches but hey what else was I going to use to measure such small deltas in preformance

9x400 1600fsb 2:3 mem @ DDR1200 5-5-5-15

Super Pi = 14.29s
3DMark06 was a tad slower (like 10 points or something like 3048 if I recall correctly)




7x515 2030fsb 1:1 mem @ DDR1030 4-5-4-12 (Tried tighter but it just wouldn't go without extreme voltages )

Super Pi = 14.44s I assume Super Pi is mostly based on memory preformance as opposed to overall fsb

3DMark06 was a tad faster (10 points about 3058 total)



The only other things I really noticed difference wise was that overall the higher fsb won, except in Super Pi.

Guess I'll be staying at the higher Fsb. Though for some reason I can't even get it to boot @ DDR1200 anymore, even with duplicate settings, odd, it was over 10 hours Orthos stable before....

Maybe one day I'll find out, but till then FSB it is!

hmm that's interesting. I'm not sure if sandra reads latency differently or what, but on boards without a fast memory enhance option, 400fsb with 2:3 should be in the low-mid 40's in terms of latency, while 500 1:1 should be about 60 or so.

Maybe it is because of the test it's running, "random" instead of "linear" ?

Is there another program I could use to test latency with?

Not too sure, I just wanted to see where I stood on which OC to stick with.

Wish I could get the timings a bit lower @1030, but I couldn't even get it stable @ 4-4-4-12

already have 2.35vdimm and don't feel like going much higher for everyday use. Oh well, it'd be pretty insignificant increase in preformance anyways.

For 24/7 use, i would choose non of those settings cos youre only giving your hardware stress for small gain in performance...... In your shoes, Ill run it 400fsb x 9 and use 4:5 divider ( DDR1000) 5-5-5-15 @ 2.1v...... I just think 1200mhz on the memory or 515fsb on the motherboard is too much stress for 24/7 operation.... D9's even if used at only 2.2v do degrade fairly fast and your stix not able to boot at 1200mhz anymore is a sign.....