1:1 or 1:1.66 ?

[babcom]

Hi, Please can somebody answer this:

I'm putting together a new rig for my brother. It includes a C2D @ 800Mhz and DDR2-667.
Will it bench higher in Sandra Memory Bandwidth if I run it 1:1 or 1:1.66?
And is 1:1.66 - (CPURAM) the correct divider to use if I want to run the ram at its full clock speed?
I will not be overclocking.
Thanks.

 

[Throckmorton]

You don't get more performance with 1:1.66. Leave it at 1:1

 

[JustaGeek]

With the FSB of 200MHz (800MHz "Quad pumped"), the 1.66 is the correct RAM:FSB ratio to run your memory at 667MHz.

It will most likely bench more in SANDRA memory bandwidth test.

 

[BonzaiDuck]

Tentatively, I beg to differ slightly, keeping in mind JustaGeek's remarks about my speculations over dividers that are " NOT(1:1)." JustaGeek knows his s*** -- just the same.

It looks to me that the bandwidth benches are almost a wash. At 1:1, with good memory modules, you can tweak the latencies and the command-rate to be rock stable and near-stunning bandwidth.

"Nefarious" (another member) proved to me that you can run Crucial B-stix latencies at 3,3,3,3,1T at DDR2-667 and get "read" bandwidth of around 9,800 MB/s. I just haven't had the nerve to try it yet, but I get 9,750 at 3,3,3,6,1T, tRC=9 at DDR2 = 704 Mhz. At 667 Mhz and 3,3,3,6 [etc.] -- I get close to 9,500.

This adds some complication and tedium to the over-clocking process, but I think it's worth it for giving more options and flexibility if you're going to over-clock at all.

Furthermore -- with the recent tweaks I've made along the lines mentioned, my subjective experience with a favorite game program leaves me with a profound conclusion that I cannot tell the difference now between Q6600 @ 3 Ghz and Q6600 @ 3.2 Ghz with CPU/DDR = 1:1 and the tighter timings and command rate.

Further, what was once noticeably slower at 3 GHz, now seems real snappy. I mean -- "ree-ull snappy" -- "bang, bang" -- "zoom, zoom."

 

[JustaGeek]

1:1, with the FSB of 200MHz, would mean the memory running at 400MHz.

Would it yield the same performance as 1.66:1 for the 667MHz? Pehaps, depending on the timings/latencies.

I always favour running memory to specs - so if the OP has the 667 memory, why not try it at both speeds with SANDRA, use it in the most commonly used applications, and decide what works best.

You can approach overclocking the memory from 2 different perspectives. One is the clock/multiplier, and relaxing the latencies accordingly. Relatively easy on NVidia motherboards, with practically unlimited memory multipliers/ratios.

The other one is keeping the memory at 1:1 with the FSB, and tightening the timings. This method seems more complicated, since it involves many more factors with the 4 basic timings, all the subtimings and the command rate.

Bandwidth wise, the simplest assumption would most likely be (and correct me if I'm wrong, Bonzai, since you've recently experimented with latencies):

667MHz at 3-3-3-8 = 800MHz at 4-4-4-12 = 1066MHz at 5-5-5-15 --> bandwidth wise.

A simplistic approach, but I hope it gets the message across.

 

[BonzaiDuck]

Yeah, JaG -- I can jump on your bandwagon of logic there. It seems about right.

With good memories -- d9's -- you can get tRAS to 6.

To give credit to the other approach, I've seen highly-overclocked systems (AigoMorla's) running the Crucial Tracers at close to DDR2-1000 and tCL = 4 with tRAS = 9. It would be interesting to see the bandwidth benchies for that setup. But figure his tRAS is 50% "longer" than mine, or maybe 33% slower (?), or I'm 25% "faster" with my tCL = 3. It would seem less confusing -- simpler as you imply -- in comparing his bus-speed to mine. But the proportions are roughly equivalent.

For me, right now, I'm just short of 10,000 MB/s for a "read" bench-result with the 1:1 divider. I just cut back on my Vcore, and may drop it another notch. So -- cooler, faster as "at least fast enough." No hurry about replacing processors, even if it is last year's "B3."

As to "getting the message across" -- I hope so. Another thread has someone desperate to avoid giving up FSB ground, but the general assessment appears to be "you're roastin' and toastin' your CPU as we speak!"

I posted this quote before. 45 years ago, I was taking drivers' ed at the local high-school summer-session. No air-conditioning, hot So-Cal summer mornings, you could fall asleep from the heat and boredom. They presented a lot of movies. One was titled from a Mario-Andretti quote: "Turn left to win." Which is to say -- "Don't push it too hard, and avoid being a burned-up hot-dog."