Q9450 topping out at 423fsb, any help appreciated

[negev]

My Q9450 seems to have an fsb wall at 423fsb.

424 causes a rounding error in P95 after about 5mins, but 423 runs for over 2 hours at 1.35vc (1.33 real), 1.35 vtt

I've tried upping vcore to 1.375 (1.39 real) and vtt to 1.425 but still the rounding error after 5mins. As I'm trying for a 24x7 overclock I don't really want to push these higher on air..

GTL's are 10,5,10,10 which I believe is optimal for this processor, NB is 1.6 (can't really exceed that on air), VDIMM 2.09, PLL 1.58, SB 1.275

Is it worth playing with the GTLs to see if I can get it stable above 423, or is the rounding error a clear indicator of the cpu failing due to lack of vcore?

[Xello]

My Q9450 seems to have an fsb wall at 423fsb.

424 causes a rounding error in P95 after about 5mins, but 423 runs for over 2 hours at 1.35vc (1.33 real), 1.35 vtt

I've tried upping vcore to 1.375 (1.39 real) and vtt to 1.425 but still the rounding error after 5mins. As I'm trying for a 24x7 overclock I don't really want to push these higher on air..

GTL's are 10,5,10,10 which I believe is optimal for this processor, NB is 1.6 (can't really exceed that on air), VDIMM 2.09, PLL 1.58, SB 1.275

Is it worth playing with the GTLs to see if I can get it stable above 423, or is the rounding error a clear indicator of the cpu failing due to lack of vcore?

The day they have a program like Prime95 that allows you to actually diagnose individual system instability patterns will be a happy day, my friend, but i'm afraid there's no way to do that just yet. In other words, if you're getting round errors, there's no way to be sure what setting is causing it.

Sucks, i know!

Now, what multi is this using? For your VTT, remember that VTT must always be lower than Vcore. I use 1.35v for all my settings. Raja from Anandtech measured this himself and says that 1.35v actually gives 1.27v though, maybe this is something that needs to be fixed.

Something i would try is go to 424fsb and change your NB GTL to +3 steps, see if that does anything.

[negev]

I'm using 8x and the CPU333 multi, I don't think the ram is the problem as I've tried lowering the multi so it's well below stock and the exact same thing happens.

I'm trying your +3 step suggestion now, thanks :)

[negev]

That does seem to have helped, NB is crazy hot though, 66C under load :/

[HDCHOPPER]

ya your vtt is kinda high and same for pll but work on your vtt the lower you can keep this number the better

and ya hot ! thats why thiers a copula nipples on the nb hahahah good luck man really !!

have ya tried skipping a few fsb numbers thier might be a hole

still eyem sticking with cooling the nb below 50c and dropping your vtt

[Xello]

Yeah keep VTT at 1.35v, and yeah that's way too hot for NB temp, from what i've read, are you sure it's accurate reading?

Try experimenting with the NB voltage.

[negev]

Been priming at 3386MHz 423x8 for over an hour, i kicked vtt down to 1.275 about half an hour ago and its not crashed out yet..

As for NB, it's currently at 66C as reported by Aegis but isn't crashing.. HWMonitor reports the same (labelled as TMPIN2). In HWMonitor there is also TMPIN0 which is 56C and TMPIN1 44C, perhaps the reading is wrong and one of these is really the NB.. would be a bit strange for Aegis to get it wrong.

I'm surprised the NB gets so hot really, as I AS5'd the whole thing.

*** DAMMIT *** P95 just crashed, starting again with vtt at 1.35... fingers crossed...

[Xello]

If the board is uniformly undervolting the VTT then 1.27v would actually give 1.2v so yeah best to leave it at 1.35v.

I use Everest for my NB temp (it's called Motherboard under the temp bit), although Aegis agrees with it. However i know one or two people that experience some inaccuracy with Aegis so perhaps try Everest and see what it reports. I use Everest at the moment because it has all sorts of useful info you can display in the system tray, though i'm sure they will be updating Aegis in a little while to have more stuff like this :)

If your NB really is getting that hot i'd reduce your NB voltage considerably, you might even gain stability this way as i found that with certain settings it didn't like certain NB voltages, even if they were high enough to be stable.

TMPIN0 on hwmon is your cpu so 56c isn't bad.

For further tweaks i'd try changing NB voltages and also your NB GTL if you can't get stability with +3, drop to +2 and if that doesn't work try +4.

I'm not sure how hot the NB can safely go, i've read 45c but i've had it between 50 and 55c myself with no problems, perhaps Saaya can give some advice there.

[negev]

If I drop the NB voltage one notch using Aegis, P95 crashes immediately..

it's actually only at 62C atm. I'm going to leave 8x423 running to see if i can get this 12hr prime stable, then when thats done ill play with the nb GTL and see if i can push higher.

[negev]

terrrifick, now it won't prime at 400x8 for more than half an hour, something has died a horrible death :(

[negev]

feckin thing wants more voltage for 400x8 now, i think i've wraith'd my cpu :P

[HDCHOPPER]

ya vtt can be a cpu killer

[Xello]

If you went back to 1.35v you haven't killed your cpu with your VTT... Even 1.42v wouldn't be anywhere near enough to kill it, and if you put it higher than that it wouldn't post because as i said before the VTT needs to be lower than the vcore for the cpu to operate.

What vcore were you using?

[negev]

I went up to 1.39 vcore trying to get 424x8 to run for more than 5mins without a rounding error.

Anyway 400x8 is stable again, I loosened some memory timings (timings that were quite aggressive and hadn't been thoroughly tested anyway) and gave it one extra notch of nb voltage and it's happy.

With the new settings 416x8 has been stable for 2h 30, I think im going to leave this running for a full 12 hrs so that I know these settings are rock solid before I start trying for higher again.

[negev]

Overclocking is so tedious lol

I'm not looking forward to tightening the ram timings when ive found the max fsb, 12 hours for each test of each setting change... painful

[Xello]

I was priming for about 3 weeks 24/7 to get my overclock settings, you don't have to tell me :D

1.39v is fine, anything below 1.45v isn't going to cause you any problems. Don't be afraid to push it up that high, as long as your temps are fine. I've been told that people are running at 1.5v and more on air for months without problems, but i err on the side of caution with a cpu as expensive as this.

What memory do you have? What are the stock clocks, timings and voltage? To start with you should always have your memory at its stock profile, don't tighten etc and then try to overclock your cpu. Do the cpu first, then the memory when you're happy with the cpu :up:

For the 12hr thing, personally i've never had something fail past 7hr 20mins, so i tended to use 8 hours, depending on what stage of testing i was at. For example while tweaking settings in an attempt to get something i was happy with i would maybe run 3 or 4 hours for each adjustment to make sure it was stable, then when i had that setting at my final goal i would go for 8 hours. If it was stable then, i'd move onto my next setting. When i had all my settings at what i wanted i'd do a 24hr blend for confirmation. I found that to be a more efficient way of doing things - when i first started i was doing 24hr runs for each adjustment i made to make sure it was stable, but as you could imagine that would take years to get anywhere..

[negev]

I have 2x2GB G Skillz PI sticks, they're at stock timings 7-7-7-18 but the MHz are above their stock because of the fsb overclock. If this prime im running at the moment fails I'm probably going to drop the ram multiplier just to rule out the ram as being the source of failure.

Your methodology sounds sensible, I've had failures at 3 hours, 6 hours but never after 8 hours. I have actually just figured out why I've been having such trouble overclocking, I didn't realise that the new quad version of prime95 only stops the thread for the core that errors, not the entire thing. So when I overclocked to 3.33 before, I left prime running overnight and 12 hours later switched the monitor on, saw the green icon and the running threads and assumed it was stable, but looking back now (as the same settings for 416x8 failed today after about 3hrs) it was probably very likely that one or two of the threads had error'd.

Oh well we live and learn :D

[negev]

Ok thats it, no more overclocking until I get water.

This is the second time that the Blackops has decided one of my RAID5 disks has failed while running prime, so the excessive heat from the NB is obviously a problem.

I'm just thankful it only affected one drive, but it's still very annoying as it will now take nearly 48 hours to rebuild the array during which time my computer will be slow as :banana::banana::banana::banana:.

UGHH!

[Xello]

That's this kit here then?

http://newgskill.web-bi.net/bbs/view.php?id=g_ddr3&no=30

If they're at 7-7-7-18 for 1333mhz i'd leave that alone, that seems perfectly fine and shouldn't be causing you any instability. If you're going over 1333mhz because of the fsb increases, you might need to up your ram voltage a little, but i can't advise you on what is safe for this kit, you'd have to find out what other people have been safely using. My kit is rated 1.8v stock and i can go up to 2.15v no problem, though i'm at 1.95v just now, but that's on a corsair kit with samsung IC's.. I don't have any experience with G skill.

The one setting you might want to look to at some point under memory timings is the Performance Level (tRD) as this directly relates to the NB itself. You can use memset to find out what your PL is at by default. Raising this a step or two can bring stability with a performance hit on some occassions, so worth keeping in mind. My default is PL8 however i can go to PL6 at 400fsb and PL7 scales for me up to my current max of 460fsb.

Sorry to hear about your raid failure!

[negev]

Nah this one:

http://www.overclockers.co.uk/showproduct.php?prodid=MY-025-GS&groupid=701&catid=8&subcat=145&name=G.Skill%204GB%20DDR3%20PI%20PC2-12800C7%201600MHz%20(2x2GB)%20CAS7%20Dual%20Channe l%20Kit%20(F3-12800CL7D-4GBPI)

They're rated up to 1600MHz, I was priming at 1664 which is probably fine but now my raid has failed so I'm back to 400x8 waiting for it to rebuild.

I need water on the NB, peaking at 66C is way too hot and all that heat is obviously radiating down to the SB causing the raid to mess up.