new guy here, would like some help with AM2 3800+ 64 X2

Hi all!

Just joined the site and I'm hoping you all can clue me in a bit better to where to go from my current situation. I read the sticky for this section and tried to perform the isolation max tests. My problem is, I can't bump the FSB clock with out having to nudge both the Vcore AND the DDR2 ram voltage.

System: Gigabyte M59SLI-S5
AM2 64 3800+ X2 (Windsor)
Arctic Cooling Freezer 64 Pro
4GB OCZ Plat DDR2
750W Thermaltake PSU

here is the CPU-Z for the last stable settings with everything stock except the voltages and the FSB freq. Obviously, just playing around a bit with the settings. Edit: Stock temps only hit 32C under load, with the setup you see in CPU-Z still only hitting 36C max.



Whats the deal with voltage, specifically vcore and ram voltage? As I said above, it seams I cant bump the FSB freqs without bumping both the VCORE and mem voltages on my setup. I personally don't like the idea of going to high in the voltages on either, obviously more clock + more volts = more heat.
Whats been the experience on this board with voltages on a Windsor 3800? What about ram? I dont like the idea of having to boost ram voltage too much either. If this is normal, then please say so, and excuse my inexperience.


Thanks in advance!

nm, i figured it out. had to kick the memory down from ddr2 800 to ddr2 667 and kick the multi's for HT and MEM down to 4 (from 5). working like a charm now

I've found with DDR2, it's best to let the ram do what it wants in BIOS and take advantage of the Learn or Train DQS built into the bios. The bandwidth we get from DDR2 is more than what the majority of our apps will use. You can tweek lats later after you find the solid high clock of the CPU. If you get past the 2720 mhz wall I hit then . I upgraded to a bigger PSU since I last tried so I may be able to go further. If you are setting your own lats then raise TRAS from 15 to 17 or 18. This only costs a little on bandwidth but you will gain it back within the next 10Mhz raised on the CPU Clock (HTT).

edit:
Locate your LDT, or HT multi and drop it to 4x or 800MHZ you have it running 1160 not which is pushing the NB a little. You may find that you will be able to oc a lot further.

thx

dropped all the mults for HT and Mem to 4, that definately helped. But it wasn't quite stable. I dropped the mem speed from 800 to 667 and that made everything run stable. now to push the cpu a bit

alright, need some thoughts on what I should do here. This is where I sit right now.



I can run all benchmarks such as 3dmark06, pcmark05, sandra tests, etc.
But when I try to run two instances of prime95, one for each core, core #0 always fails after 10-12 minutes and core #1 continues to chug along. Is it also common to see a 4 degree difference in temps between the two cores?

Any recommendations on where to go from here?

THX!

i dunno whats up with your bank cycle, but its 7 for my ddr, maybe you should try lowering it a bit? unless thats how ddr2 is supposed to be.

otherwise looks good, u got it set at 1.35 vcore? maybe try upping it abit?

as far as I know, the limitation on DDR2 is the timings right now, unfortunately the clock cycle is where it needs to be.

i'm sitting at a 1.375 (ish) vcore, next bump would take it to 1.40 (stock is 1.2)
I'll try that and see what happens.

thanks for the help serlant.

Stock is actually 1.25 to 1.35. Cool and Quiet will bring it down somewhat but your oc would be shot! 1.4-1.425 is the max for a 90nm with excellant cooling with temps in the 40sC max load. I would not take a 65nm above 1.375 unless it's temps are in the lower 40s! Just my 2 cents and personal limits!

I'm not inclined to raise to voltage either, any other recommendations cajun?
edit: tried the 1.4v still had one core error in Prime95. So i reset it to 1.375.

Blake,

Here is where I had to go to get my stable 2.75 with mine. My memory blows and will not do over 440 (880ddr2), so I had to drop to ddr2533 or cpu/8 in bios, but then I got 4-4-4-12 timings outta the stuff to compensate. Heres my setup:

275X10, 1.475V
4Xhtt
ddr2 533, 4-4-4-12 1T @ 1.9V

It is on an ABIT K9N-sli with an Nforce 570 chipset if that helps you out some. Rig #1 in my sig..

The main problem I've found with OCing the AM2 stuff is that you cannot isolate the Ram to find out its max. You need to drop the mem divider and find the cpu max, then up the divider and begin to raise fsb until it fails, knowing your CPU SHOULDN't be the issue. IDK about your mobo, but with mine, the dividers change as you drop cpu multies.

If you have any questions I'll be happy to attempt to answer them.

ive got tons of features to play with in the bios on this gigabyte board. But what I ended up doing was dropping the CPU clock 2mhz (my board gives me choices in 2's, i.e. 202, 204, 206 etc.) I just got finished running an instance of prime95 on both cores with out issue for 4 hours, so I'm happy. I'll try to push it further when I have time.

dont have time to post CPUZ right now, but the short is..

1.35v on vcore
248 x 10
HT & mem @ 4
mem dropped from DDR2 800 to DDR2 667.

3800+ X2 64 @ 2470
HT link @ 989
Mem @ 412.3

Don't be shy about dropping the mem divider again and pushing that FSB if your CPU will handle it. I find the PC MUCH more responsive with the higher clock speed and absurdly low mem speeds than the other way around. Besides, dropping the htt to 3X and mem to 533 is the only way to isolate the CPU if you want to find its max. My board just happens to allow a really high htt for some reason.. it'll go something like 1300+ htt.

Now would I like to have my cake and eat it too??.. You betcha.. That is why I am in the market for some decent clocking ddr2...

yeah my board actually posted at 1500htt, then as soon as it tried to load windows from the hard drive, it crapped out and i changed the multi back down to 5x. i just did a bios reset and forgot to change it back

ok interesting new development.

The 4 hours of prime95 yesterday (i stopped it because I had work to do) gave me a little confidence. So i dropped my HT and MEM multis and established that i really can push my cpu to 2600 (its like 2597 or something like that) if I wanted.
I still want to keep my mem and HT somewhere around normal though, so I dropped the FSB back to 250 and replaced the ht and mem multis to 4. I tried running Prime but it only ran for an hour. Which didn't suprise me, I hadn't quite found a fully stable spot at 250. So i ticked the clock down 2 which lasted 4 hours yesterday (10 x 248). I ran prime overnight, and low and behold 248 x 10 only ran for 6 hours before it had an error.

Couple of questions:

On the torture testing, how much is ram tested? Could a ram error cause that test to end?

Could I be running too much vcore v?

could i be running too little V in my ram (im +.1 right now, i found out my ram likes 2.1V for overclocking and I'm sitting at 2.0)?

would unstable memory clocks cause the test to end as well?

thats all I can think of at the moment. if anyone has recommendations, let me know!

thanks again

I believe the OCZ plat runs STOCK at 2.1, not only for OC. I had a pair myself, and my abit board would not post until I put in another set and changed the bios to 2.1, as that is what the newegg description said. So, yes, you're mem voltage may be a factor.

Second, what temps are you getting at 2600 and is it still at 1.375V?

Have you dropped the cpu multi and mem divider to check your max ht? Check it with 5X, 4X and 3X(just to be sure 4X isn't limiting you)

I would fool around with the cpu max until I had it perfectly stable before you begin to raise everything else back up, that way you know it is not a prob with vcore etc. Get rock solid stable with each piece you isolate so you know what to rule out when you do the full system OC.

And yes, unstable mem would cause prime/Orthos errors. It has set conclusions that should be reached with each test that it checks against, and unstable memory could cause miscalculations as well as the cpu. Blend uses more mem than small ffts, but small still uses it a bit.

And.. BTW.. I'm sorry.. Mine is at 1.475, not 1.375...90mn...

ill do another run at 2600 and get a CPUZ for you. Temps were fine IMO 43 under load from Prime95.

In terms of max HT, I'm really not inclined to push HT too much. It ran stable at 1100MHz, which is the highest I pushed it to. I'm trying to keep it around the stock 1000Mhz.

FYI- check the rev. # of your OCZ plat. Rev 2 operated anywhere from 1.9 to 2.1 stock. I have rev 2 which booted at 1.9 stock on my Gigabyte board .


edit: just got back from work ran Prime 95 all day. 2500 stable now, had to up my ram to 2.1V