Help OC my Athlon 64? (try 3)

R

runLoganrun

Gawd
Joined
Jul 4, 2005
Messages
601
Athlon 64 3500, Win
Asus A8N-SLI
Corsair TwinX 1024-3200XL

My first overclock. Stock is 2.2G and DDR400 with 11x cpu multiplier and 2-2-2-5 1T.
After a bunch of testing I changed it to 2628mHz, with 9x cpu multiplier, a 5:6 ram divider, and the ram running at 239mHz with 2.5-3-3-7 1T timings. I ran it through 3D mark 03 and it bombed horribly from the stock set up. Whereas initially I had scored a respectable 7362, I now went down to 4363! What is wrong with this picture? My other plan was to tweak it so I had 2580 with 2-2-2-5 1T timings and the ram at 215. I haven't tried that one yet though. To accomplish this I had my DDR voltage at 2.8v, and my cpu voltage at 1.55v, the max for this board.
Thanks for any help
 
hmm..im not quite sure..but just to inform you instead of making new threads all the time just do a new post with bump in it...
 
I would first find the limit of your RAM... since it is PC-3200, you never know where your max FSB will be... So find the limit of your PC-3200 TwinX ram, and start from there...

Reset the CPU multi back to 11x... and use memtest to find the limit fo the RAM...

Since you already know how far your OC on your winchester can go, you will be able to better judge your final OC and ram divider.
 
Mayhs said:
hmm..im not quite sure..but just to inform you instead of making new threads all the time just do a new post with bump in it...
Click to expand...

those other threads are empty, it wouldn't post my message, some kind of server problem, it just went to the white screen, it held my title though.

About Ram max: It's 215 at 2-2-2-5, and 250 at 2.5-3-3-7, both with 2.8v.
 
Now comes the question for you... which set of timings do you want to keep?

with the 2-2-2-5 timings, you will see real world performance gains: faster load times and things of that sort, with the looser timings at 250FSB, you will see higher synthetic benchmarking scores...

Another note is the idea of using a RAM divider while benchmarking... On my LAN rig machine, I would score lower while being overclocked with a RAM divider, than I would without the RAM divider... I am not sure if the RAM divider is what was causing the problem or not...

My preference is to keep the RAM divider at 1:1... I run my main rig with the 2-2-2-5 timings on my RAM, and keep it at 2.5GHz for everyday use... and when benchmarking, step up the timings to 2.5-4-4-10, drop the multi and increase the FSB as much as possible...

It all comes down to your personal preference.
 
With the 2-2-2-5 timings, which seems like the way to go, and it topping out at 215 I can only get a total overclock of 2365, which isn't much. Am I better off at this point loosening my timings, while still keeping it at 1:1? Can I push the DDR voltage above 2.8v with this memory? (it goes up to like 3.0v).
Thanks
 
It does seem to limit the processor with the low FSB at 2-2-2-5

The only way to tell about the voltage is to up it a little (2.9V) and see if you can surpass 215 @ 2-2-2-5 or 250 @ looser timings... with your CPU, i would venture to use the looser timings and to just go for a max OC at 1:1 timings.
 
So, I ended up with 2625 (250*10.5multi. w/ 2.5-3-3-6 @2.8v). I tested the memory at this speed for 12hrs on memtest. Everything was fine for a couple weeks and then I started getting errors, like my computer crashing and rebooting. I backed off the HTT to 238, and increased the multi. to 11, giving me 2618. I thought since I lowered the HTT that much that I could lower the RAM voltage as well, so I lowered that to 2.75v. I started testing it with Prime 95, and only ran for 4 hours before I got a fatal error. After that I bumped the RAM voltage back to 2.8v, and ran another test, small fft torture test, which ran for 9 1/2 hours before getting an error. Then I thought I should just lower the HTT a little more and lower the total clock. I did this, to 236 (with 11 multi. yields 2596). So the question is: Should I be increasing the voltage here and trying to get more memory speed (possibly going to a 10 multi.)? I read that I can safely take my memory up to 2.9. Or, should I just be doing what I'm doing and lowering the total clock? I did test the cpu to be ok up to 2632, but I didn't do it for much duration. It turns out that lowering the HTT a couple MHZ let it run on the small FFT torture test for 18+ hours with no errors.
 
What is the vcore? Maybe you are just trying to push something that isn't there.

Upping the vcore, if you can and keeping the temps around 50 during load is what I would try.
 
I haven't touched the vcore at all. I'm pretty sure my bios has that. I don't know what it is at now or how far I can move that. That wasn't something I read about in the articles I had on overclocking. I did try moving up the ram voltage to 2.90 and it barely changed the 2-2-2-5 max clock, it went from 215 to 218, and I bet it wouldn't even have stayed at that after some more grueling testing. So how much should I change the vcore? What does that do exactly?
 
BLiTzKRiEG said:
Now comes the question for you... which set of timings do you want to keep?

with the 2-2-2-5 timings, you will see real world performance gains: faster load times and things of that sort, with the looser timings at 250FSB, you will see higher synthetic benchmarking scores...

Another note is the idea of using a RAM divider while benchmarking... On my LAN rig machine, I would score lower while being overclocked with a RAM divider, than I would without the RAM divider... I am not sure if the RAM divider is what was causing the problem or not...

My preference is to keep the RAM divider at 1:1... I run my main rig with the 2-2-2-5 timings on my RAM, and keep it at 2.5GHz for everyday use... and when benchmarking, step up the timings to 2.5-4-4-10, drop the multi and increase the FSB as much as possible...

It all comes down to your personal preference.
Click to expand...

a faster cpu clock with a divider on the ram and lower timings = faster than 1:1 and lower timings and slower cpu...told to me by eclipse
 
You must log in or register to reply here.
Back
Top