Core 2 Duo, DDR2, overclocking

[naddie]

As with any new platform, there is a new learning curve to get over. It took me some time to learn AMD's Athlon 64 no-FSB/HTT/bus nuiences.

 

Now that Core 2 Duo is on the scene as the best enthusiest CPU "eva", is there any tutorials or write-ups on how to OC this bad boy? I tried my best to understand the memory dividers and how DDR2 memory works in terms of how the memory speed is related to Intel's FSB (and how this "quad" multiplier applies to the whole schema).

 

Can someone either break it down or point to a place where they do just that?

 

Also do memory dividers affect performance on the Intel platform? Or is it a non-issue as with the AMD platform? If dividers doesn't matter, does that mean overall CPU/FSB/memory speed is the determining factor for "teh win"? How does tight latencies at DDR2-800 compare to loose latencies at DDR2-1067 and beyond?

 

Sorry so many Qs. Just that going over to the Intel side just isn't confusing enough, I now have to contend with DDR2 memory (higher latencies, but faster memory speeds).

[Rroff]

Traditionally intel platforms have performed best when the memory divider was 1:1 while overclocking - not having the low memory latencies and HTT of AMD systems an increase in CPU speed can easily be negated by poor ram performance - to see what I mean get a good GPU and raise the CPU speed via FSB with the ram on 5:4 until your back at the normal RAM speed (400MHz w/ PC3200) and benchmark with quake 3 (a good one to see CPU performance rather than GPU performance) then set the divider to 1:1 with the same FSB speed (needs good RAM) and you will see a significant performance increase... this applies to other games/applications as well for the most part.

 

NOTE using quake 3 isn't entirely 100% valid way to compare as quake 3 seems to be very sensitive to the system FSB, if you had an unlocked intel CPU and dropped the multiplier on the CPU back but raised the FSB you'd quite likely still see a performance increase.

[naddie]

What do these ratio refer to?

 

The FSB of Conroe platforms are now at 266, not the 200 of P4s, right?

 

Currently P4s are using DDR2-800, which is a ratio of 1:2, which is (FSB*2)*2 ?

 

So with Conroes, is it the same formula?

 

I took a look at how I would OC a E6600. If it is true that 4GHz is possible, it looks like my FSB would be 445 Mhz (445*9 = 4005 MHz). If that is true, how would I work out the memory?

[Rroff]

my ratios were based on later 478 chips, I haven't played with conroe yet but I'm assuming that to a degree the same rules apply even if the numbers are different.

[Thraxz]

Your idea isn't faring very well. What you're pointing at as evidence for your hypothesis is not really evidence at all for what you're trying to claim.

 

It, rather, points to the over arching and entirely too obvious idea that the faster your ram runs the fast your computer will go. It has nothing to do with 1:1 ratios but instead is just, plain, higher RAM MHz. Of course, on the old crappy 478 platform running at greater than 1:1 ratio didn't yield much performance because the FSB actually had LESS bandwidth than the memory. I found this out first hand with my old P4 rig. I was running the the ram on the DDR500 divider and the FSB at stock 200 and benchmarks showed that the performance was BARELY increased above 1:1. Sisandra even spit out a message on it's memory bench that my memory was being FSB limited.

 

How about that .?

 

What do these ratio refer to?

 

The FSB of Conroe platforms are now at 266, not the 200 of P4s, right?

 

Currently P4s are using DDR2-800, which is a ratio of 1:2, which is (FSB*2)*2 ?

 

So with Conroes, is it the same formula?

 

I took a look at how I would OC a E6600. If it is true that 4GHz is possible, it looks like my FSB would be 445 Mhz (445*9 = 4005 MHz). If that is true, how would I work out the memory?

 

Most of the DDR2 800 RAM is currently capable of DDR 1000 OCs.

 

Your memory at the DDR800 setting would run at DDR1780 SO you'd just drop it back to the (7/6?) DDR466 divider to get DDR 1038. You can even drop it back to the DDR433 (13/12?) divider to get a DDR 964. Hell even the DDR 400 yields a DDR 890... a nice ram OC.

 

I realise that the FSB at stock is is not 200 anymore so my numbers aren't right but either way this gives you the idea that you'll be running BIG dividers to reach 4GHz. Gimme time and I'll get REAL numbers up.

[Rroff]

eh?

[Thraxz]

WITH THE STOCK 266FSB THE RAM SPEEDS ARE AS FOLLOWS FOR DIVIDERS:

 

1:1 - DDR533

16:15 - DDR566

9:8 - DDR600

6:5 - DDR633

5:4 - DDR666

4:3 - DDR700

3:2 - DDR800

 

SO With the FSB of 445 you'd get the following RAM speeds at these dividers:

 

3:2 - DDR1335

4:3 - DDR1187

5:4 - DDR1113

6:5 - DDR1068

9:8 - DDR1001

16:15 - DDR949

1:1 - DDR 890

[naddie]

So the math is...

 

WITH THE STOCK 266FSB THE RAM SPEEDS ARE AS FOLLOWS FOR DIVIDERS:

1:1 - DDR533 = 1(266 * 2)/1

16:15 - DDR566 = 16(266 * 2) /15

9:8 - DDR600 = 9(266 * 2) /8

6:5 - DDR633 = 6(266 * 2) /5

5:4 - DDR666 = 5(266 * 2) /4

4:3 - DDR700 = 4(266 * 2) /3

3:2 - DDR800 = 3(266 * 2) /2

 

SO With the FSB of 445 you'd get the following RAM speeds at these dividers:

 

3:2 - DDR1335 = 3(445 * 2) /2

4:3 - DDR1187 = 4(445 * 2) /3

5:4 - DDR1113 = 5(445 * 2) /4

6:5 - DDR1068 = 6(445 * 2) /5

9:8 - DDR1001 = 9(445 * 2) /8

16:15 - DDR949 = 16(445 * 2) /15

1:1 - DDR 890 = (445 * 2) /1

 

Excellent. Also, I was afraid that there weren't that many dividers. This is very uplifting news.

 

Also, is a FSB of 445 MHz even possible with the current i975 chipset? I've read easy OCs of 400 MHz. But what about 445 MHz? Or even more? Anyone figured out the ceiling with a low-multi Conroe? (E6600 and below? 9x and below?)

[Thraxz]

You'd do well to note the crazy O'Cing done with the ES conroes over as XS forums. That'll tell you a lot about the Conroe/MoBo limitations. Keep in mind they are using pick of the litter chips and do not exactly reflect the production ones we'll see.