What's holding me back? ram?

Question

I can only hit stable at 3.45Ghz (15*230) to run 1:1 and anything higher, memetest got errors. Tried increased to ddr 2.8v, loosen timming at 2.5-4-4-8 but no luck. When running test 5 got tons of errors. I didn't try prime95 since memetest didn't pass... If set to CL3 my system won't boot (Can't set CL2.5 to CL3? Lousy ram?) So I decided to run dividers to see how far my system can push.
Currently [email protected] (15*250) cpu temp idle at 50c load at 70c (readings taken from speedfan v4.2)
Ratio 5:4 ddr@333
Cpu voltage - 1.45v
DDR voltage - 2.8v
AGP voltage - 1.6v
Spread spectrum - Disable
AGP/PCI clock - 66.67/33.33
Dram timing manual 2.5-3-3-8
Running prime95 either small or large both failed. Increased vcore up to 1.5v, agp 1.7v and timming at 2.5-4-4-8 also failed when running prime. Now guys what could be the problem here? Even running at ratio 3:2 ddr@266 also won't help! Thanks 

Answer 1

The whole point of using memtest is to ensure your memory is not the limiting factor in your overclock. In order to really test your memory, you need to do this.
Run a 4 hour loop of standard tests (1-7)
Run a 24 hour loop of test 5
test 5 needs something like 32 (or 200 (or 1 million)) iterations to test your whole stick of memory, however many iterations it is, 24 hours should be just enough time to complete that)
No offense but you have cheap memory so it will not overclock very far. If you want to overclock with a 1:1 FSB:MEM ratio then I would recommend buying ADATA Vitesta PC4000 memory. It's cheap, I've used it in my P4 system, and I've seen it work in other people's P4 systems.
Therefore, in order to utilize your cheap memory, you're going to need to run a divider (keeping your memory at 200mhz or lower).
If you think just because you're running a divider, now you can raise the FSB to 250 (keeping your memory at 200) then you're wrong. That's a significant increase in CPU clock speed and chances are it's just too far of an increase or it will require more of VCore to run at that speed. With certain pentium 4's you can do this sort of thing but not all of them. I had a 2.4C that went straight to 3.0ghz without a bump in VCore. You have a prescott so that changes the scheme of things. FYI: I've seen prescotts overclocked real far, so I'm surprised you can't get this far.
Here's what you do. Run a 5:6 divider (the one you have now, where your memory is 200mhz when your FSB is 250). Set your FSB to 250 and run memtest the way I told you to. This will test your memory at it's highest possible frequency, which means, when your FSB is lower then 250 your memory will be lower then 200 and that means that your memory is safe at 200 so it's safe at anything lower then that.
Now set your FSB to 200, keep your VCore stock. Turn the computer on, run prime95 "in place FTT's" at priority 10 and make sure it doesn't fail for as long as you feel comfortable (note: this is testing your CPU at stock speed and your memory at 166 I believe). The reason I say test it for as long as you feel comfortable is because it's stock speed so as long as it passes for like a half an hour you know you're okay. Now raise the FSB from 200 to 205, do not increase VCore. Run prime95 as stated again for 1-3 hours. If that passes, raise the FSB from 205 to 210, again, do not increase VCore. Keep raising the FSB in 5mhz incraments (without raising VCore) until prime95 won't pass anymore, (note: the longer you run prime95, the greater the chance it will fail!!! so keep this in mind when cutting off your tests at the three hour mark, i've seen tests fail at 5, 7, and 12 hours in). When it doesn't pass anymore, raise the VCore to the next lowest VCore (1.525 to 1.5375 (or something lower if there's something in between the two of them)). Re-run prime95 with the new VCore for as long as you can tolerate and if it passes raise the FSB and try again. If it fails, go back to BIOS and raise the VCore again to the next lowest incrament. Repeat this process until you finally pass prime95. then go back to increasing FSB by 5mhz incraments, again not raising the VCore with each iteration. Then when you come to the next failure in prime95, begin raising the VCore again. Eventually you will come to a point where you cannot stabilize your computer in prime95 regardless of how much VCore you give it or you can't get to the desired VCore because temperatures are too high. If that's the case, back the FSB down to the last stable FSB/VCore and that will be your max overclock.
What should you have learned?
You need to watch your temperatures the entire time when running prime95, never take your eyes off of that. Too much VCore will fry a chip but extra VCore contributes to electromigration which in turn decreases the life of the chip. Therefore, think before you act. An unreasonable amount of VCore, although it stabilizes the chip, may eventually end it's life preemptively.
You must record each successful iteration of FSB/VCore so this way when you encounter an irreperable failure you can revert back to the fastest previous FSB with the correct VCore