Can I do better than this overclock? How important is CMD rate?

[Fryguy1]

Note: this is pretty long, but worth the read, epecially if you are looking to overclock your athlon64

Ok, well I started overclocking my system today, here are the specs:

Athlon64 3000+ newcastle
MSI K8N Neo Platinum
1Gig Corsair XL
Raptor 74gb
Radeon 9800pro
Another random IDE hard drive installed (doesn't matter)

To start the process, I set my memory divider very high, so I could isolate the cpu. Using clockgen and about 2 hours of my time, I found out that my cpu maxed out on stock voltage at 247mhz (a 470mhz overclock). Being on air cooling, increasing the voltage didn't do much for me at this point but make more heat. If I REALLY wanted to I could have squeaked out 250-255, but I would have been running 60C+ under load. As it is now I'm never above 55, and I'm happy with that.

So I started to introduce the ram into the equation. And I found out some interesting facts:

1. My Corsair XL absolutely will not boot at 233mhz or above with CMD rate at 1t. No matter what I used for voltage or other timings, it wouldn't boot.
2. Even at 232, 231 mhz, with "respectable" timings (3-4-4), it was having some seriously problems with stability, even with voltage cranked on my board (2.85v)
3. If I concede command rate to 2t, it runs at 247 with 2.5-3-3 timings pretty perfctly (haven't primed it overnight to verify stability, but it's close enough).

With this said, and given that I'm NOT dropping my multiplier to get this high of an FSB, should I just stay at 247 with the 2T command rate? Or is it worth it to pursue getting 1t command rate working at a much lower fsb?

and if 247 2.5-3-3 2t is the best, and after I verify that it's 100%, where can I go from here?

 

[AristoV300]

Corsair XL will run up to 253 @ 2.5-3-3-6 on 2.75 Vdimm.

 

[maluckey]

It all depends on what you want to do. It's normal to need about 7-10 Mhz FSB to overcome the difference between 1T and 2T command. If you can gain more than 10 Mhz, then use the higer (numerically) setting. You will still be faster. If you are trying to squeeze the ultimate efficiency at a given Frequency, then tightest timings and lowest command rate. I tend to set the entire system to the absolute fastest timings, then crank the FSB.

 

[Fryguy1]

maluckey, setting fastest timings and then cranking fsb only takes me to around 210-215 or so. Seems QUITE lacking when I can be gaining a couple hundred more mhz on the cpu by doing it another way.

So here are my options:

2-2-2 212fsb 1:1 divider, 1t command rate

2.5-3-3 247fsb 1:1 divider, 2t command rate

2-2-2 247fsb 166divider, 1t command rate (ram at 205mhz)

Or I can try to push the fsb and turn down the cpu multiplier to 8 or 9, but let's worry about that after. Am I going to have to benchmark the above 3 results?

also, in regards to pushing the fsb higher.. if I can turn the fsb up to say 274 (the "even step" up so my cpu stays the same 1 multiplier down), and I have to turn the HT multiplier to 3 to make it work, then it's basically not accomplishing anything right?

 

[maluckey]

Sorry for the delay,

You would need serious difference in CPU raw power to overcome loose RAM timings. It would have to be on the order of well over 100 Mhz in most cases. I don't know why your RAM isn't up to snuff, but it does seems a little weak. Since your RAM is a weak link, I'd suggest slightly looser timings, but not sloppy. Go for the middle ground. The way to find the middle ground is relatively straightforward.

Try this: start with a 1:1 ratio. Lower your CPU multiplier to something low, find the RAM maximum at tight timings and high voltage. Next, benchmark it in SANDRA, and your favorite games, along with anything else you can throw at it that stresses the system. Write it down...all scores and settings.

Next, Crank the CPU multiplier as high as you can run the CPU with stability, and repeat the testing.

Finally, loosen the RAM timings a little, crank the FSB as high as it will go, and repeat. all the above.

I'm of the opinion that you should not go for anythng other than 1:1 ratio. It kills gaming FPS (if you are a gamer).

The reason I suggest testing, is because I've found that most Bench progs are useful for stability, but nothing that I've found (not even 3DMark) can predict 3D behavior better than actually running the 3D app. I find that Il-2 and UT2004 are great progs for finding weakness. Either of these progs can make a system fall to it's knees if cranked to max details and resolutions.

 

[ectx]

I thought cmd rate is used for via chipset only? This shows how much I know.