Confirming my settings

Hey all; not new to overclocking but I've never had nearly as good a result as with my new system, and I owe a lot of it to lurking in these forums.

I'd appreciate it if some of you could just have a quick look over my settings to make sure I'm not approaching any dangerous voltages/settings.

System specs:
Core 2 Quad Q9400 @ 3.76Ghz
Asus P5Q Pro
OCZ Reaper PC6400 CL4
XFX Radeon 4830

These are the settings I'm currently at in the BIOS:

FSB = 470
FSB Strap = 333
PCI-E = 100
Dram Freq. = 940
Latencies: 5 - 5 - 4 -15, tRFC is at 55
All other latency settings at "Auto"
Ai Clock = Medium
Ai Booster = 10 (Performance Level)

1.3500v = cpu (CPU-Z displays it at anywhere from 1.248 to 1.304 which kind of concerns me)

1.56v = PPL
1.28v = VTT
2.02v = RAM
1.34v = northbridge
1.20v = Southbridge
1.60v = PCIE
LLC = Disabled (I've noticed this seems to immediately make things unstable if it's enabled...seems to prevent booting right away)

I've ran OCCT and OCCT Linpack for 1 hour each with no errors. Also fired up LinX and ran it 150 times with no errors. My temps never go higher than 60 or 61 on any of the cores so that's fine. I'll do a 24 hour prime test soon to do a final check.

What I'm kind of concerned about is that fluctuating vcore...0.05v sounds like a lot to me but what do I know!

Thanks

EDIT: I should add I'm using a 550w Antec power supply which I've only read great reviews about...so I don't think that would be the culprit as far as the vcore fluctuation goes.

.05v core fluctuation is nothing, i would be worried if its more like .20v. can you tell use the exact model of the Antec? some antec's are better then others and some past one have a known history of having CAP's fail over time. i say your seeing for near 3.8ghz on a 45nm quad look good. you may be able to get 3.8ghz with a touch more VTT and CPUV. that is if you know your ram can handle the faster FSB speed.

Thanks for the reply! Good to hear about the fluctuation.

The power supply is an Antec NeoPower 550.

I ran Prime95 overnight the other day and I had an error in core 0 after about 7.5 hours, and then core 1 followed about 15 minutes later.

I upped the VCore a couple of notches, and ran Prime95 again and this time Core 0 failed after about 15 minutes, which I found kind of strange.

From my understanding, a failure on just one or two cores is generally a VCore issue?

I then upped the Vcore a few more notches again (it's currently at about 1.296v in CPU-Z) and lowered the FSB to 467, since I'm more concerned with real world performance and stability, and I know I'm not going to notice a 50-60mhz drop. Temps are stable at about 56 during the prime. Prime95 is running right now, and it's been about an hour and a half with no errors so we'll see. If I can get a 24 hour stable run I'll see if I can get away with a bit less Vcore since I dropped the FSB a tad.

If it fails again I think I'm gonna drop the multiplier to 6 and run Prime95 to see if it's the memory. If that's not it, I'm not sure whether I should raise the VCore some more, or try messing with the PPL/VTT voltages...I'm not really sure how high I can raise them safely either, as I'm getting a lot of conflicting numbers here and there. I know VTT shouldn't exceed 1.4v with 45nm chips, but not so sure about PPL, or Northbridge for that matter.

One thing I'm finding with the higher Vcore is that it's a lot more stable now. Still fluctuates a bit at idle, but at full load it's solid at 1.296V, and barely ever moves. I'm not sure if this is a good sign or meaningless but at least I know it's not a bad thing!

set your voltage to where your the 8/7.5hr stable, then just up the cpuvtt,mch. ther cores failing could be from to much cpuv or to little. its really something that needs to be worked on along with cpu vtt and mch since cpuvtt is partly dervied from the mch voltage. does sound like you might possible have ram problems though depending on the prime95 test, small/large/blend? 1.3v for that speed is low, your right about cpu vt, with pll i honastly dont know but i wouldnt go high on it either. with NB its not really that big of a deal but going to 1.5v for MCH would be to high IMO.

like i was talking about with prime95 but really didnt go to detail on. small FFT's covers the cpu, large goes into ram only, blend does cpu/ram/MCH(aka northbridge). now if it does fail in blend but not large or small then you need more MCH voltage. if you really want to test ram in more detail use memtest86, you can boot is now from a USB pen drive or just get the ISO to make it CD bootable.

Hmm, alright well if Prime fails again, I'll try working on MCH and VTT voltages once I make sure its not the memory holding me back.

On all the Prime tests it was blend since I wanted to test overall stability. If it fails again I'll try just small FFT to help me figure out if it's memory/cpu.

I've done a few memtest+ runs with no errors but only 2-3 runs, so I'm guessing I should try a longer test.

I also did the windows HCL Memtest earlier, two full runs (200%) with no errors, which took about 1.5-2 hours, but again that might've been too short I'm guessing. I'm also not sure how effective it is at detecting errors compared to Memtest+ but from what I've read it's fairly good.

ok so if your ran those 2 on the memory then i could be cpu or the MCH. run small FFT's use the inplace option and set it to the amount of L2 cache the cpu has. if the cpu passes then up the MCH voltage since you know it is failing in Blend.

Alright, Prime95 failed again after 4.5 hours on Core 0.

I didn't see an option for "in-place" with small FFT's. I assume you mean custom, checking the "in-place" option and specifying a small FFT range? Not sure what that range should be, but core 0 failed on the 32K test so I put the range 8-256K just to be safe.

Crossin' my fingers!

EDIT: Okay, just reread your post; when you said set it to the amount of L2 cache do you mean the FFT size or the "memory to use" box. Because if I check "Run FFTs in place" that box is grayed out.

If you mean FFT size that's 6144 which I'm guessing is large fft?

ok so if it failed at 32k that is in the cpu's L2 cache. try upping the cpuvtt/cpufsb and possibly the cpuv a touch, like one notch. other then that im starting to think your temps at those voltages are high enough to cause the core to error out. what are you temps like while it is loaded? i have preference to keep my temps 62c max but i aim for 60c max load temp.

when your in prime95 with the custom option if you use "memory to use" that tells prime95 to go to ram. you dont want to use that option, ur the "inplace" and specifying the range. since you using a Q9400 you want the range to start at 0 and max size would be 3072k. btw are you running enough instances of prime95 to test all 4 cores?

Don't think it's the temps. Full load in prime, they don't go above 55-56 and running LinX or OCCT, they never go above 61. (In fact only one of the cores reaches this high, the rest don't top 60).

I'm using 64 bit prime (Vista x64 here) and it runs a thread for each core so I'm certain it's testing them all; there's 4 windows, one for each thread/core within Prime and I'm only getting failures on core 0 or 1. From my understanding because these quads are basically two duo's stuck together, this just means that one pair of cores is a bit more stable than the other?

I'll restart prime with the values you gave me.

I'm curious though, if the L2 cache is 6MB, why would I test 0k to 3072k rather than all the way to 6144K? I'm guessing it's because that 6MB is divided equally between the two sets of cores. If that's the case, would I still enter 3072 if my Prime has a seperate thread for each core?

right use 3072 for the value, this way each thread will get up to that amount ensuring it uses all the L2. you could half that value since 3mb l2 is shared for each set of cores as you know. now as far as the numbering goes, i followed this from core temp. core 0 is on one set core 1 is on the other. making core 0/3 one die core 1/4 the other, didnt make sense to me. i just followed the temps since the temps for each set of dies should be equal or +-1c of each other.

Hmm, how did you tell that from core temp? Going by temperatures, 0/1 are one die, and 2/3 is the other for me, since 0/1 are usually at the same temp, and 2/3 are at their own. For instance right now at idle, Core 0 and 1 are at 35, and core 2/3 are anywhere from 33-34 so the two dies seem to be within that 1 degree or so difference you mentioned.

There may have been more than just that one core I mentioned that hit 61 but it's the only one I really noticed and it was only 1 degree more than the others; from what I've read, the temp sensors on the quads can be a bit flaky so I guess it could be that.

Either way, I'm gonna start a Prime test with the settings you specified and let 'er run for the night.

I can't thank you enough for the help Evilsizer, it's much appreciated.

that no problem,dude! hope you stick around with us for a long time... cause after lga775 with i7/i5 ocing is going to get even more fun!

yea i mainly watched the temps in coretemp to figure that out...well looking at other peoples ss's with quads that is...

Well, failed after about 7 hours, 50 minutes on core 1, so it seems to be CPU related. Good to get that cleared up.

You mentioned raising CPU VTT which I don't see in my BIOS. I do have FSB Termination Voltage which is what I was referring to when I mentioned VTT. I raised that a notch, and raised the VCore a notch and am running the test again. If it fails this time I think I'm just going to drop the FSB a bit until I'm at 3.7 even or so.

change the fsb strap to 400 and see what happens on the same voltages.

I think I had given that a shot before and it was unstable but I don't remember what the rest of my settings were so I'll give it another go.

I'm wondering if I should just do OCCT to try and detect errors. Waiting 8 hours for Prime to finish every time is getting to be quite the hassle. I've never ran it for more than an hour so I'm thinking it might be worth a try.

umm prime95 runs for as long as you want... some people go by 8hr prime stable some preferer 24hr stable.... OCCT does the exact same thing as prime and runs for as long as you want...

Oh I know that, but Prime seems to catch the error after about 8 hours so I generally have to wait at least that long to see how stable everything is.

Everything I've read indicates OCCT detects errors faster than prime but I've never ran it longer than an hour. I think I'll give it a shot.

If P95 keeps erroring out at a certain time check what fft length its at when it fails. Next time you test do a custom test that uses only that length. If it passes for 30mins or so then you might have made it over your hurdle. Then you can try the whole test again w/ a bit more confidence.

Hmm, that's a good idea but it didn't work for me. I'd run the same one or two sizes that would fail and they'd be fine for hours.

Either way I seem to have broken the 8 hour barrier.

I wasn't seeing any increased stability after raising the vCore 4-5 nothes (over a few tests) and didn't have any extra luck with raising FSB VTT so I figured I'd just bring it down to 3.7 and see if I could get it stable.

13 hours so far and no error! If I can make it to 24hr, I'm just gonna stick with this for now; don't have it in to me to spend another week tweaking for an extra 1-200mhz.