Help me hit 4.2 GHz Stable 24/7

[PiLsY]

Maximum vtt (measured with dmm) + 0.5v. Youll need a fan on the ram over 1.75v. Your limit depends on the chips, but realistically for benching 2.0 to 2.2v and day to day 1.75v to 1.85v. Higher values suit micron ram, middle samsung, lower elpida.

Whats your vtt at the moment?

[LennyRhys]

Thanks for the tips PiLsY. 195x21 is running solid at the following; I just haven't had the time to properly tweak it yet so I could most likely get it a bit lower:

CPU +120 (1.36 load)
VTT +120
vDIMM 1.64
PLL 1.617
Amp 0.700
IOH 1.18
VCCA 1.134

I can't believe that a mere 5MHz more on the bclk requires such an ENORMOUS jump in voltage for cpu, vtt, and dimm. Speaking of multis, I think my chip much prefers the odds to the evens.

Incidentally I wasn't aware that turbo makes CPUs run hotter, but even at 1.36v load it briefly touches 82C, and my air cooling isn't stellar. If I were to slap a lapped TRUE on that, I'd be looking at 75C max.

One thing I thought I'd mention - when I drop the CPU multi, the vcore takes a dive too (seems to be linked). Is that supposed to happen? When I boot into windows with default vcore seleceted in BIOS at 200x12 etc, it's at like 1v when I check it in CPUZ. Prime just craps out within seconds, naturally.

[PiLsY]

Thats an odd one dude?

How high can you go at 19x? It tends to be that as you increase vtt youll have to pull up vdimm with it for the uncore. They dont like to be too far apart. Theres a sweet spot where the cpu likes a particular ratio of vtt:vcore:vdimm. Its hard to find though. Anyway see where 19x takes you before you need to add vcore.

My board doesnt like 193 to 199 bclk. I lose loads of cpu mhz if I want it stable in that range. Seen it on 2 other bbloodrage too, could be common to all of them I suppose...

[LennyRhys]

Tried 197x21 without adjusting any voltages and it was priming for about 20 mins before I got an error, which is quite good I think.

Suffice to say, I decided to leave it at 195x21 since it's stable...or so I thought. Earlier tonight, I was just browsing the net (XS funnily enough) and Aero suddently turned itself off. I was like...huh? Then two mins later, the PC spontaneously restarted. C'mon Mr.Bloodrage...you can take several hours of heavy priming, but when it comes to fifteen minutes of light browsing, you throw in the towel? LOL

I just upped vcore and vtt to +140mv to see if that little boost will do the trick. Weird.

[PiLsY]

Honestly try over 200 and drop the cpu multi. I bet your board has the same nasty instability as mine. Youre describing exactly the same symptoms as I see between 193 and 199bclk.

[LennyRhys]

Seems my board or chip just doesn't like 200 bclk, but it's really ok - I'm actually surprised the thing does 4GHz so easily and comfortably (without any random shutdowns, too). Maybe later I'll find some patience to try over 200 bclk again, but for now I need to mod my PSU!

Cheers

[CobraXP]

Seems my board or chip just doesn't like 200 bclk, but it's really ok - I'm actually surprised the thing does 4GHz so easily and comfortably (without any random shutdowns, too). Maybe later I'll find some patience to try over 200 bclk again, but for now I need to mod my PSU!

Cheers

Dont feel bad. I have a friend that just lost his old EVGA E758 mobo out of warranty. He was running at 4.2GHz 24/7. He ordered a new FTW3 and no luck at 4.2 only 4.0

[PiLsY]

FTW3 is just the old LE with sata 3. Check them out side by side. The LE was an atrocious board and I cant see any changes of note between it and the FTW3.

LennyRhys - got your pcie at 102mhz? Helps with bclk...

[LennyRhys]

Yeah, PCIE to 102 was one of the first things I did

[PiLsY]

Just checking .

Im at 23 x 191 for day to day clock - 200bclk really is just a number, there's no performance attached to it. As long as you have your cpu and ram in the right speed range through multiplier manipulation then bclk itself influences nothing.

[LennyRhys]

I usually try to have the ram running at its rated speed (or higher if it can handle it), and set everything to capitalize on that...call it a quirk. Ever since my first Micron D9s running at 950MHz 4-3-4-9, I've been obsessed with getting the best memory frequency and latencies that any particular kit will allow.

Since I now have 1600 C9 memory (which it seems can do 8-8-8 no problem, but not 7, nowayhosay) 200bclk would give me that, if nothing else.

[PiLsY]

How high does it go at 9-9-9 and 8-9-8?

I use Ballistix 1333c7 : 1910mhz 7-8-7 1.82v / 1910mhz 8-9-8 1.72v / 1910mhz 9-9-9 1.58v. Its a really early kit so I dunno what the newer stuff is like. Cost me £60 new unopened in a clearance bin nearly 18 months ago. Best buy ever .

[CobraXP]

FTW3 is just the old LE with sata 3. Check them out side by side. The LE was an atrocious board and I cant see any changes of note between it and the FTW3.

LennyRhys - got your pcie at 102mhz? Helps with bclk...

I suggested my friend a Bloodrage . He prefers evga. The NB runs 70c in idle and 80c+ under load. The board looks cool though

[homefry]

I suggested my friend a Bloodrage . He prefers evga. The NB runs 70c in idle and 80c+ under load. The board looks cool though

maybe the evga nb temp sensor is actually on the chip.

i think our sensor is actually behind the mobo "near" the nb chip.

[CobraXP]

maybe the evga nb temp sensor is actually on the chip.

i think our sensor is actually behind the mobo "near" the nb chip.

The heat sink is so hot you cannot keep your fingers on it for a long time. I think you could really fry on egg on it... or atleast warm up a bagel.

[LennyRhys]

I tried 195x21 again and...to put it succinctly: FORGET IT!

I got a "page fault in non paged area" BSOD so I immediately slackened off the timings on my memory to 9-9-9 at only 1560MHz (rated to do this at 1600MHz with only 1.6v)

Immediately ran memtest with 5GB utilized to make sure my memory was ok, and I also ran a bit of Prime95 too.

This morning to double check, I put Prime95 on, and BAM... a "page fault" BSOD slaps me in the face before Prime even goes green LOL

Back to 191x21 for the moment...maybe I'll have to do stay at 191 bclk like you PiLsY

My ram does 1600 9-9-9 flawlessly at lower bclk; haven't tried pushing it yet but I might give it a shot and see if I can get the 10x mutli out of it for 1910 cas9

Is your Ballistix kit 3x2GB? If so that's amazing - £60 is an absolute STEAL!

[PiLsY]

Yeah 3x2gb - got it in ocuk clearance section in april 08 . Possibly best value bit ive very picked up for a pc .

I''m still surprised you cant do 200bclk. I've managed it on every 1366 setup ive tried.

200x19 @ 9-9-9 same voltages doesnt work?

[LennyRhys]

I got 200x19 stable at +220mv on the uncore and 1.7ish on the ram, but I was unhappy with the CPU clock. 200x21 will require 1.4v at least on the CPU (I think).

Seems I've got a "new" problem to deal with now anyway, and it's being a right pain in the AR$E if ever there was one - getting the same page fault 0x00000050 BSOD regularly. I've set my system back to stock for now, so if I get the BSOD again it's NOT GOOD!! Either I got glue on my ram when I was modding it or I've borked one of my system files, or my GTX470 ram is damaged...there's no way to know for sure

I have to use the G44 bios now because P09 won't work with my XMS3 modules installed. Challenge for the day: diagnose the BSOD fault and deal with it

EDIT: surely after 20 passes of LinX @ 4GHz with all memory, I can rule out faulty hardware?

[PiLsY]

Sounds like memory to me - timings not quite right or volts a little low.

Linx really only stresses the cpu, doesnt do much for uncore/ram/bclk.

Try prime95. Select "in place", then custom. Uncheck the in place box and change memory to use to 5200mb. It may take a while to hit 100% load as it frees up ram but be patient and it will get there. Let that run through to the end of the 512k test, reboot and repeat. This is the only way to generate good ram and uncore load. Any problem with vtt/vdram or mem timings and that will error out.

[LennyRhys]

Cool I'll get onto it right away. I'm just flummoxed that I've been getting no BSODs (none at all) and now I'm getting it several times a day...

cheers

[PiLsY]

BSODs

0x00000009 : IOH or IOH vcca voltage too low / Northbridge overheat
0x00000050 : memory timing/voltage or uncore ratio/voltage
0x00000101 : Vcore low
0x00000109 : incorrect vddr (+/-) or vddr:vtt ratio wrong
0x00000124 : incorrect qpi/vtt voltage (+/-)

[LennyRhys]

Cheers for that PiLsY. Just did as you suggested with Prime and it took about 90 mins to get to the end of 512K, no errors, going to reboot and do it a second time.

The more I look into this, the less I'm convinced it's hardware related - the x50 stop error is a notorious one because (according to microsoft) it can also be driver related, which I'm suspicious of... I'm passing stress tests right left and center, but a simple thing like opening a web browser prompts the BSOD. Argh!

[Heavybomber]

These errors cannot be directly related to certain hardware malfunction. E.g 0x00000050 can be caused by anything related to memory highway or vCore itself(e.g trashed TLB lines)
The only exception is MACHINE_CHECK_EXCEPTION, which is certainly located in processor and indicates low vCore.

[therager]

I got 200x19 stable at +220mv on the uncore and 1.7ish on the ram, but I was unhappy with the CPU clock. 200x21 will require 1.4v at least on the CPU (I think).

You should try 210x19 with lower memory multiplier - x6 mem/x12 uncore (thus resulting 1260mhz mem/2520mhz uncore). Lower multipliers will allow you to use default Vtt and Vmem. If everything is ok - buy a better memory .

These errors cannot be directly related to certain hardware malfunction. E.g 0x00000050 can be caused by anything related to memory highway or vCore itself(e.g trashed TLB lines)

And with one and the same settings in bios, but in a case of instabillity, you may reproduce almost all types of errors .

[LennyRhys]

@ therager, my system has already passed 8 hours of prime at the following settings (191x21, memory x8 1532Mhz 8-8-8-24, uncore x16) with NO blue screens or crashes:

cpu +80mv
vtt +100mv
vddr 1.64v
IOH 1.12

everything else default voltage. Only a few days ago did I start getting the annoying x50 stop error at least once every day, so I've been trying to reproduce the results I had before.

Last night I ran 191x21 with exactly the same clocks, but slightly higher voltages, and one thread errored out in 10 minutes... not quite the stable system I had before?

cpu +90mv
vtt +120mv
vddr 1.65v
IOH 1.18
VCCA 1.134

Well I've taken out the XMS3 (I'm suspicous of it, VERY suspicious) and put in my 2x1GB of D9JNL which will tell me if the XMS3 is the cause of the problem.