Sign in with
Sign up | Sign in
Your question

Memory 'Sweet Spot' Question

Last response: in Overclocking
Share
June 17, 2008 8:48:20 PM

As those of you who live on the forums probably know, I'm currently overclocking a new computer (thanks for the help!).

I fixed the heating issue (got a new heatsink - Xigmatek pushpins just fail), and the OC issue (This Q6600 can barely do 3.5, and 3.6 is right out), and now have a memory question:

This probably gets asked often, with slightly different numbers, but here goes:

I have 2x2GB (4 GB total) PC 1066 G.Skill ram with 5-5-5-15 timings.

My processor's FSB is overclocked from 266x9 to 389x9 Is it better to have my ram as:

a) 1037 Mhz (Or somewhere around 1034-1037, I can't remember) with a 4:3 ratio and timings near 5-5-5-15 (before overclocking)

b) 778 Mhz with a 2:1 ratio and lower timings (presumably I can overclock timings more at a lower speed)

In other words, is the 1:1 'sweet spot' performance boost enough of a difference make taking a 260ish Mhz drop in RAM speed worth it? (even with the slightly better timings at a lower speed?) At what point would it stop being worth it?

Thanks!
Zithras
June 17, 2008 9:27:10 PM

I'd stick with 1:1, since if there is any difference, it's not notable, as shown by grasky and his series of tests. If you lower the timings as well, I reckon performance will match that of higher speeds, but the difference is pretty slim.
June 18, 2008 5:26:16 AM

Thanks for the advice. I've just about finished on the processor, and it's now time to start poking at the RAM.
Related resources
June 18, 2008 7:39:06 AM

z - your calculations are slightly wrong. Running the RAM at DDR2-778 IS a 1:1 throughput ratio, as long as your RAM is running in dual channel mode. There's really no "sweet spot", it's just that since all data to/from RAM has to go through the FSB as well, there's not much point in having the memory bus throughput higher than the FSB or vice versa.
June 18, 2008 11:02:27 AM

1:1 is the golden setting for any FSB. because FSB sets the limit.

lower you CPU multiplier to 8x and push 450FSB on it.so you can get extra memory performance.
June 18, 2008 11:02:59 AM

do you mind list all your hardware spec and BIOS setting for me?
June 18, 2008 11:23:33 AM

I would leave it at 1:1, but you may be able to lower your timings since you are not running at 1066.
June 18, 2008 12:20:53 PM

+1 for 1:1, some BIOS put it as 1:2 like Gigabyte. Same thing. What's your Q6600 VID? Maybe we can help you push it to 3.6, some more voltage on the vcore or the nb may help. Push it to 9x400 for 3.6 if the extra voltage stabilizes it and run ddr2-800 with 4-4-4-12 timings.
a c 197 K Overclocking
a b } Memory
June 18, 2008 12:59:46 PM

Zithras wrote:
I fixed the heating issue (got a new heatsink - Xigmatek pushpins just fail),
There's a TRUE type mounting kit available that let's you do away with the push pins.
What kind of HSF are you using now?

and the OC issue (This Q6600 can barely do 3.5, and 3.6 is right out),
My Q6600 (VID = 1.2625 volts) is running at 3.6 GHz with a TRUE/S-Flex HSF at 1.45 volts, drooping to 1.40 volts.

and now have a memory question:
memory timing question follows

Early last year, I put a system together using an eVGA 680i motherboard, E6600, and 2 GB of Crucial Ballistix RAM. I got to 3.3 GHz (367MHz X 9) pretty easily. Default memory timing was 5-5-5-15-2T. I worked it down to 3-3-3-7-1T (Orthos stable after 24 hours). Memory benchmarks indicated a 6 - 7% increase in memory i/o. :bounce:  However, system and graphics benchmarks indicated no significant improvement in perfomance. :( 

My suggestion is to get your memory running at 1:1. Then set the timing to 4-4-4-12. If it runs with stability, stop there. There are more productive uses of your time.

a b K Overclocking
a b } Memory
June 18, 2008 3:03:58 PM

Quote:
My suggestion is to get your memory running at 1:1. Then set the timing to 4-4-4-12. If it runs with stability, stop there. There are more productive uses of your time.



For Truth. Don't worry about the last Mhz on your Memory - Get your FSB/CPU Overclock stable, then set your memory to run accordingly. Personally, for a 3.2 Target, I would suggest a 400(1600) FSB, 8 Multi, and RAM at 1:1. In real terms, it shouldn't make any difference except on some synthetic benchies. But I like my math to be neat, so... <shrug>
June 18, 2008 4:07:10 PM

Same here, so I'd only stick with 266MHz, 333MHz, and 400MHz FSB.
June 18, 2008 4:11:36 PM

Evilonigiri said:
I'd stick with 1:1, since if there is any difference, it's not notable, as shown by grasky and his series of tests. If you lower the timings as well, I reckon performance will match that of higher speeds, but the difference is pretty slim.



What exactly IS your avatar...I think it IS canned hams !
June 18, 2008 4:14:35 PM

royalcrown said:
What exactly IS your avatar...I think it IS canned hams !

Uh?! Now that's a first...

It's a riceball. A RICEBALL. Wrapped with seaweed. In Japanese, it's called onigiri and it's very very common in Japan. It's also very common in Hawaii.

Geez...canned hams?
a b K Overclocking
a b } Memory
June 18, 2008 5:04:21 PM

...and they can have umeboshi inside...
June 19, 2008 12:32:15 AM

same to Intel. only one thats exclude from this rule is AMD CPUs and the new Nehalem.

the link you provide dont give anything that to do with memory multiplier and the result. the TRD value is for the MCH not the actual ram it self. you should have done your homework and reading better. just to let you know you will only gain a few or even just 100MB/s increase rather then 5/600MB/s increase if the actual FSB is increase at to the same value.
June 19, 2008 12:33:05 AM

what i mean about you should do your homework better is learn about how the CPU communicate with the Ram.
June 19, 2008 7:36:41 PM

You could always go for

425 MHz x 8 = 3.4GHz or 438 MHz x 8 = 3.5GHz as this will be 1:1 but the memory will be slightly overclocked (so memtest it to check its stable) and the FSB will be a bit higher (may need to bump the FSB volts 1 notch).
June 19, 2008 7:47:40 PM

since i know that 1:1 will give the better performance. so i wont spend more money to buy a chip thats got more than 8X multiplier since i will get best at 8x so i have 800mhz on memory which is more then enough for anything wit Cas 4.

but higher multiplier will give you the option to higher speed on CPU if the mobo is limiting it from getting to higher FSB.
June 21, 2008 5:35:09 AM

iluvgillgill said:
same to Intel. only one thats exclude from this rule is AMD CPUs and the new Nehalem.

the link you provide dont give anything that to do with memory multiplier and the result. the TRD value is for the MCH not the actual ram it self. you should have done your homework and reading better. just to let you know you will only gain a few or even just 100MB/s increase rather then 5/600MB/s increase if the actual FSB is increase at to the same value.


iluvgillgill said:
what i mean about you should do your homework better is learn about how the CPU communicate with the Ram.


iluvgillgill said:
since i know that 1:1 will give the better performance. so i wont spend more money to buy a chip thats got more than 8X multiplier since i will get best at 8x so i have 800mhz on memory which is more then enough for anything wit Cas 4.

but higher multiplier will give you the option to higher speed on CPU if the mobo is limiting it from getting to higher FSB.

My reading comprehension is just fine, thank you. And I didn't even need 3 posts to explain myself.
June 21, 2008 6:15:23 PM

Scotteq said:
...and they can have umeboshi inside...

Doh! Totally forgot about that. I haven't had an onigiri for about 10years. I mean a real authentic Japanese onigiri.

Anyways, enough derailing the thread.
June 21, 2008 6:53:03 PM

can it be make into a sweet taste?
!