1:1 or 4:5 vs mem bandwith/improvements??

[skyguy]

Specs:

-E6300 w/Zalman 9500
-Gigabyte 965P-S3 w/Thermaltake Northbridge cooler
-OCZ Plat DDR2-800 @ 4-5-4-15
-1:1 @ 430x7 = 3.0 ghz

Runs rock solid, temps are fine.

However, I've been hearing that going to a 4:5, even with 5-5-5-15 timings is better on the 965P than going 1:1. And the resulting increased RAM speed and bandwidth is far better overall for system performance (and games) than simply measuring performance by SuperPi times @ 1:1.

Can anyone substantiate this? All the recommendations here seems to be 1:1 (@ CL4 for my RAM), but is this isolated to particular mobos or chipsets? I've been searching through XS forums but nothing so far on the S3 mobo specifically.

So on P965, bandwidth > timings ?? If that's the case, then why all the fuss about having to run 1:1?

Or is the case of how big of a latency penalty to go asynchronous, and whether it's better to stay in sync.......?

What do the OC experts here say???

 

[Guest]

I ain't no 'expert' but I can gives you a few answer.

Most of the time, anything related to OC is much more platform specific(read 965) then motherboard specific.

For myself I don't know about running with memory divider. I often recommend 1:1 because I know it will work and offer good performance.

If you see people on XS running 4:5 then they're probably right. It is true that timings don't do a whole lot(to an extend) on the C2D expect in a few benchmark. But it is also true that increase bandwidth doesn't do much more for the C2D. Anandtech tested that from 1066 to 1333 you would see a 2.4% average performance, with like 7% in video encoding.

I am somewhat old school since I didn't get my new platform but back in the days, asynch wasn't always recommended, so I recommend 1:1 most of the time because you can't go wrong =).

My advice would be to try to squeeze as much mhz from the CPU as this is what will yield the most performance in real life. If you hit a wall with the cpu then you can start trying to extract as much performance from the ram.

Sisoft sandra benchies and such don't tell you that much, I think you should be looking at your regular usage, and try different setting and see whats the fastest for you in real life.

 

[rwaritsdario]

Why are you asking this instead of going for it and testing it??

The reason why 1:1 is recommendable for OCing is so the memory wont be the limiting factor in your overclock.

 

[Mondoman]

I've got to agree with dario here -- test both ways on the apps that matter for you.

 

[skyguy]

Well, I *think* I have more headroom in my memory than my mobo. I seem to have hit a 430 FSB wall on my S3.....at least, I haven't spent tons of time trying to get past it.....I gave up after a few "no-boot-to-Windows" and just went back down. So if my mobo is limiting me and my memory can go higher, a 4:5 might work in this case??

As for testing, what do you recommend? I normally run SANDRA, SuperPI and so forth.....but honestly, I don't have a benchmark in "real applications" like WinRAR, Photoshop, etc. Is there something already out there that I could use instead......a system benchmark program, and app benchmark program, etc???

Or would I just have to create some files and get out a stopwatch?

Thanks.

 

[rwaritsdario]

So if my mobo is limiting me and my memory can go higher, a 4:5 might work in this case??

Yes, quite possibly considering the godlike modules youve got.
Use the regular SANDRA, SuperPI and the like plus a few game benchmarks if youre into them. Theyll give you a accurate prediction of real world performance (that alluring 4% :lol: )