Which is better for overall performance?

[mach1canuck]

Basically which is better for overall performance-wise, that I will see day to day, and in games:

1. Overclocking your CPU as high as it will go, and using the correct memory divider to overclock your RAM using the tightest settings it will use.

 

2. Overclocking your RAM as high as it will go 1:1, using very loose timings. And use a divider to setup your CPU. (possibly CPU not being as fast as in #1) For example, I can get up to 285FSB/RAM, but I have to divide my CPU @8.0

 

What are your thoughts?

 

Do you have any better suggestions?

[Guest Kevin]

im no genious but i would imagine cpu over ram

[tristancarton]

try both. run 3dmark 2001 for comparison.

[mach1canuck]

try both. run 3dmark 2001 for comparison.

I heard 3dmark01 places a lot more emphasis on just the video card, not true?

[CosmicOne]

no 3dmark 2k1 is the best to test ur system including memory & cpu,whilt 3dmark 2k3 & 2k5 are only GFX tests.

 

aslo i was in the same situation as you on using 2.5GHZ cpu with memory @ DDR420 or use 2.47GHZ with memory @ DDR450

 

the first gave me 20200 in 3dmark2k1 & the second gave me 20600.

 

so i just stayed with teh second overclock coz it's better.

[JimX]

I'm not an expert on this, but I can calculate that within a certain range, 100MHz extra on the CPU is worth about 1 second off in SuperPi 1M, and 200 points more in 3dmark03.

 

My Winchester at 2.28Ghz did Super Pi 1M in 36 seconds, whereas the Venice I'm using now at 2.65GHz does it in 31 seconds. 3dmark03 was just under 8100 before, now just under 9000.

 

I'm now running the RAM 10MHz faster but with 2T instead of 1T timings so I don't think this has had much to do with the improvement. If anything the new RAM settings are hurting my performance because it gets worse in Everest memory benchmark so the improvement may be even greater if I can somehow get the 1T timing working.

 

Whatever the case, 2.6ghz and 2T is faster than 2.2ghz and 1T for anything practical. I just need to work out where the cutoff point is and see if running the RAM with a divider to get 1T back is faster.

[magg]

It's all about CPU speed. Dividers don't hurt AMD64's because of the on-die memory controller, it's only a negligible amount that it's reduced by. The biggest thing that reduces performance is taking your RAM to 2T. Avoid that at all costs...

 

CPU Speed > RAM freq. on AMD64's.

[Opponent]

Maxing out your CPU is the key. RAM overclocking is a minor detail.

 

Loosening the timings can easily negate any real world performance gains you get with higher fsb, you just have to test to see that for yourself...

 

Dividers per se don't affect the performance of A64, but depending on the divider and CPU speed your RAM Mhz may actually be different from what you'd expect, check out the tables in stickies for that. Also, many peeps including me get instability with dividers...

[rizorith]

I can attest to the above. A lot of people said run 1:1 no matter what.

 

I can run my computer at 255x10 at 9:10 or 240x10 at 1:1. The 255 gave significantly higher marks all around, except very slight decrease in some memory only tests.

 

Unfortunately I can't even boot into 5:6 so I don't know if I could try that divider.

 

But in the end, try both ways and benchmark it.

[Guest burningrave101]

With Athlon 64's the only thing that really affects performance is raw CPU speed. Make sure you find the very highest CPU speed you can achieve through dividers for the RAM or lowering the HTT multi or whatever it takes first and then you can lower the CPU multiplier and overclock the RAM or the HTT link.

 

RAM timings and speed only affect overall system performance by a few percent in benchmarks. Its nothing you can actually notice in real world tasks.

 

You can always tweak the RAM and other settings for a little boost here and there but in the end the CPU speed is all that really matters.

[Angry_Games]

a good divider like DDR180 or DDR166 and really tight timings coupled with extremely high cpu mhz = your best performance.

 

when you get into dividers like DDR133 and DDR100, it kills the performance to where even 200+ cpu Mhz won't make up for it.

 

I've tested down to DDR150 divider but I like to stay around DDR166 or DDR180.

 

of course...if you can get supa high FSB @ 1:1 with tight timings...that is your best bet.

 

what you really need to do is test for yourself with some BH-5/UTT @ tight tight timings on a divider (say 2-2-5-2 @ DDR166, 300HT) vs some TCCD with loose timings but close to 1:1 as you can get (TCCD, if it is good, will do 300FSB @ 2.5-4-8-4 etc)

 

the memory bandwidth difference will favor 1:1 high FSB, but the benches like 3dmark01 (which is excellent to use as a guideline but never a single bench as the be-all/end-all) will be more impressed by tight timings and high cpu Mhz.

 

or you might see no difference.

 

or you might only see a slight different with 1:1 favoring but it being extremely stressful and not as stable.

 

got to fool with the stuff for a while...there is no quick answer

[`clipse]

erm, is it true that there is no penalty for using a memory divider on all a64 platforms (given the ram at the same mhz, after divided)?