3700+ San Diego Overclocking Hell

[asicman]

OK, here is a summary with all current info on this problem.

System:
3700+ San Diego
Epox 9NPA+ Ultra w/ 6/29 BIOS
2GB Patriot 2-3-2-5
550W Aerocool Turbine Power PSU
XP-90

MAX FSB: Isolated to 360 max. This is with lowest CPU mult and DDR divided all the way down.
MAX Chip: 240x10 and DDR 166 with VCore 1.55, Vdimm 2.8 passes Prime95 overnight. If I further divide down the DDR to take it out of the equation, and raise the FSB to 245x10 then Prime95 fails in 1 minute. So 5 MHz is the difference between running stable all night and failing in one minute. WTF???
Memory: Passes 2 hours of memtest at 2-3-2-5-1T at 220MHz. So it doesn't seem to be a problem.
BIOSes: 3/22 and 7/28 can't hit the MAX Chip above. Only 6/29 is stable at that speed.
Other: I have taken HT Frequency out of the equation too. 3x, 2x, 1x, it doesn't matter. 245x10 fails immediately. PCIe is confirmed locked with (100) with clockgen. The EPoX BIOS has no manual lock setting - it appears to be automatic. What else is there to do?

Thoughts: It has to be either a weak CPU or instable power. I will try to verify with a different PSU tonight, but might not have time. Ideally I could use a scope to montior the output of the regulator, but I'd have to do that at work.

Other thoughts / ideas / random suggestions all welcome.

 

[Markfw]

I have never heard of that PSU, that would have to be my first guess.

 

[asicman]

Originally posted by: Markfw900
I have never heard of that PSU, that would have to be my first guess.

It is modular and tested VERY well. HERE is a link to the newegg page for it. 12V1 20A, 12V2 18A. Plenty of juice... And HERE is a link to a review that tested it and showed it being VERY stable on the 12V rails... like under 1% variance stable. Of course, mine could still suck I suppose.

 

[Markfw]

Try loosening or changing the memory timings. In my case , I had then set to 2.5,4,4,10 like the specs said, but I was only running PC4000 ar 233, and it auto detected 2.5,3,3,8, so I left it there, and it seems to be more stable. Also, I don;t trust prime95, I use F@H as my detector (distributed computing)

 

[asicman]

I've loosened the timings to no avail. I've set them all to auto to no avail. It's not a mem issue, or memtest would show it. And at 250x10 I can't even boot into windows. BIOS screams at me, goes into safe mode, and says to dial back the CPU speed. So if 250x10 can't boot Windows, then I have no problem believing the Prime95 failures at 245x10. I almost think this isn't a 3700+ CPU.... But I'll concentrate on the PSU for now, I think.

 

[imported_MadHatter]

I had similar situation with my CPU, Athlon 64 4000. Prime stable at 248x11 HT 4x, but increase to 250 and it fails literally in 20 seconds. I have an Enermax EG565P-FMA REV.2.0 and Asus A8N SLi Deluxe. Just out of curiosity is it gaming stable? Although my OC would fail Prime, I could play BF2, Doom 3, and Far Cry for five hours straight.

 

[asicman]

Well, I'm not a gamer, and I haven't even loaded S/W on this sytem yet. I primarily do video editing/encoding and photoshop stuff. But superpi and Prime95 both fail at 245x10, and Windows doesn't boot at 250x10, yet this thing should easily be capable of 280x10.

 

[imported_MadHatter]

Ok, if Windows won't boot then I think your original assessment that the PSU is not providing sufficient power is correct. Please post your results if you get a chance to test the PSU.

 

[wanderer27]

Probably a stupid question, but the part # is ADA3700DAA5BN CABGE0515VPMW, right ? Last part may be different, my thinking here is you may have the 130nm 3700+ vs 90 nm 3700+.

Other than that, have you tried using just one stick of RAM and/or moving the RAM to different slots ?

 

[Shimmishim]

wanderer27 there is no 939 130nm 3700+ so that is not a possibility.

it is verypossible that you have just hit the max for the chip. the difference between being stable and not stable can be 5 mhz.

my chip does 2.85 but won't run 2.9 prime stable...

i think you might just have gotten a really bad overclocking chip. don't count that out.

 

[wanderer27]

Shimmishim, yep. I was thinking 754 for some reason

 

[Lithan]

You're kidding right?

Disable 1t.

 

[Markfw]

Heres a stupid question (I hope) whats the HTT multiplier, didn;t see that mentioned. use 3x to be sure, and yes the use 2T timing also.

 

[n7]

I do hope that's not 4x512 MB RAM.

Also, as has been mentioned, set the RAM to 2T & at 100 with very loose timings like 3-4-4 until you have messed with OCing the CPU more.

 

[SlitheryDee]

My PDP memory would pass memtes86 at 2-3-2-5 1T 245mhz flawlessly, yet windows not boot unless I used a divider. The fact is if I tried to boot at 220 FSB, CPU multi x10, and memory at 1:1 windows wouldn't boot. Right now I'm running perfectly stable at 272 FSB, CPU multi 9x, and the 5:6 divider for a memory speed of 223mhz. My memory is actually clocked somewhat higher than it would run 1:1. The only difference is the divider.

If memtest86 were correct, I should be running 245 FSB, 10x multi, and memory 1:1. The point is, Just because memtest says it's OK doesn't mean it'll work. (/lecture)

Try using a divider and/or 2T command rate. My Epox 9nda3+ won't boot windows without one or the other when I'm overclocking.

 

[asicman]

Holy crap, now we're getting somewhere!!! I truely thought that since my mem timings passed with flying colors in memtest that that couldn't be the prob.... I relaxed them to 3-3-3-8 2T and dialed back some of the other DRAM parameters by 1T, and what do you know, I'm running 250x10 Prime95 for 5 minutes... HUGE step since before I couldn't boot to Windows at 250x10 ! !

Thanks guys!! Will keep at it.

 

[asicman]

And just as I typed that, Prime95 crapped out after 6 minutes. LOL.... I don't think I'll get the new PSU tonight though!

 

[Markfw]

progress. Don't give up. OC'ing is a real art that takes experience, and a lot of trying. Otherwise they would start at those speeds.

 

[asicman]

Hitting a brick wall at 250x10 with these settings :

Vcore 1.55
Vdimm 2.9
RAM timings 3-5-5-10 2T (can't get much looser here)
HT 3x

Prime95 fails in 5 minutes or less.

 

[Markfw]

Use superpi, and tell me where it fails, ignore prime for now. 2M test.

 

[asicman]

ok... give me a couple mins.

 

[asicman]

I backed down to 3-3-3-8 2T since Prime95 was running a little longer at that setting.... SuperPi completes the 2M calculation error free in about a minute. (250x10)

 

[asicman]

4M calculation also finishes error free at the same settings.

 

[Markfw]

OK, go for 8m. If it passes, keep cranking

 

[asicman]

OK running 16M. 8M passed.... Very low memory usage with SuperPi I notice.