Overclocking the UD5 with 12gb of ram.

JonJ678 · 30 Nov 2009 at 23:54

nb/obviously a work in progress, will try to rewrite as a guide if successful.
Most people use 6gb of ram with on the X58 platform. However some people wish to use 12gb, and guides on doing this are scarce. I'm starting this thread to log my progress in the hope that it'll be helpful for someone with a similar system and/or someone will spot me doing something daft. Note that copying anyone else's settings is unlikely to work out as well as doing things from scratch as all hardware is unique.

System is an unmodified Gigabyte UD5 F7 bios, i7 920 D0 stepping (3913B213). Corsair dominator 1600@c8, one set of v3.2, one of v2.1. The processor is clumsily water cooled, ram and chipset on air. Stability testing on an initially clean install of vista 64 using ibt.

Stable settings
Initial, known stable from previous efforts.

Code:

200x20, turbo off, HT on
6gb v3.2 @ 1600c8
vcore 1.3
qpi 1.295
vdimm 1.64

Code:

[u]Advanced cpu features[/u]
[b]Clock ratio 20[/b]
Turbo disabled
Cores enabled all
Multi-threading enabled
Enhanced halt (C1E) disabled
C3/C6/C7 state support disabled
CPU thermal monitor enabled
CPU eist function disabled
Virtualization tech enabled
Bi-directional prochot enabled

QPI link speed x36
Uncore frequency x16
Isochronous support enabled

[u]Advanced clock control[/u]
[b]BCLK 200[/b]
PCIe 100
CIA2 disabled
CPU clock drive 700mV
PCIe clock drive 700mV
CPU clock skew 0ps
IOH clock skew 0ps

[u]Ram[/u]
Performance enhance standard
XMP profile1
System memory multiplier 8
DRAM timing auto (8-8-8-24-2)

[u]Advanced voltage control[/u]
Load-line calibration disabled
CPU vcore 1.3
QPI/VTT voltage 1.295
CPU PLL 1.8
PCIe 1.5
QPI pll 1.1
IOH core 1.1
ICH I/O 1.5
ICH core 1.1
DRAM voltage 1.64
DRAM termination/vrefs auto

4.2ghz very nearly stable with 12gb of dominator.

Code:

210x20, turbo off, HT on
6gb v3.2, 6gb v2.1 @ 1260c8
vcore 1.375
qpi 1.335
vdimm 1.64

Code:

[u]Advanced cpu features[/u]
[b]Clock ratio 20[/b]
Turbo disabled
Cores enabled all
Multi-threading enabled
Enhanced halt (C1E) disabled
C3/C6/C7 state support disabled
CPU thermal monitor enabled
CPU eist function disabled
Virtualization tech enabled
Bi-directional prochot enabled

QPI link speed x36
Uncore frequency x16
Isochronous support enabled

[u]Advanced clock control[/u]
[b]BCLK 210[/b]
PCIe 100
CIA2 disabled
CPU clock drive 700mV
PCIe clock drive 700mV
CPU clock skew 0ps
IOH clock skew 0ps

[u]Ram[/u]
Performance enhance standard
XMP disabled
System memory multiplier 6
DRAM timing auto (8-8-8-24-2)

[u]Advanced voltage control[/u]
Load-line calibration disabled
CPU vcore 1.375
QPI/VTT voltage 1.355
CPU PLL 1.8
PCIe 1.5
QPI pll 1.1
IOH core 1.14
ICH I/O 1.5
ICH core 1.1
DRAM voltage 1.66
DRAM termination/vrefs auto

4.4ghz settings, 15 passes ibt stable.

Code:

210x21, turbo on, HT on
12gb v3.2+v2.1 @ 1200c8
vcore 1.4375
qpi 1.375
vdimm 1.70

Code:

[u]Advanced cpu features[/u]
[b]Clock ratio 20[/b]
Turbo enabled
Cores enabled all
Multi-threading enabled
Enhanced halt (C1E) disabled
C3/C6/C7 state support disabled
CPU thermal monitor disabled
CPU eist function disabled
Virtualization tech enabled
Bi-directional prochot enabled

QPI link speed x36
Uncore frequency x16
Isochronous support enabled

[u]Advanced clock control[/u]
[b]BCLK 210[/b]
PCIe 102
CIA2 disabled
CPU clock drive 800mV
PCIe clock drive 800mV
CPU clock skew 0ps
IOH clock skew 0ps

[u]Ram[/u]
Performance enhance standard
XMP off
System memory multiplier 6
DRAM timing (8-8-8-24-2)
tRFC set to 100 on each channel

[u]Advanced voltage control[/u]
Load-line calibration disabled
CPU vcore 1.4375
QPI/VTT voltage 1.375
CPU PLL 1.7
PCIe 1.5
QPI pll 1.1
IOH core 1.14
ICH I/O 1.5
ICH core 1.1
DRAM voltage 1.70
DRAM termination/vrefs auto

Iterations. (now in code tags as the list is getting quite long)

Code:

1/Known stable settings, ver 3.2, six loops, passed
2/Known stable settings, ver 2.1, six loops, passed
3/Disabled XMP. Ram multi x6, bsck 180. 12gb installed. No post.
4/Bios auto reset to stock, v2.1 in primary slots, v3.2 in secondary. Three loops passed, not patient enough to run any more at 20 mins a loop. Don't care that much if it's stable at stock speeds. 

Starting from stable settings for 6gb, with ram multi down to x6 and bsck decreased. Aiming to bring bsck back up to 200 then optimise ram. 

5/bsck 140 boots, passes a couple of loops
6/bsck 160 boots, 180 boots, 200 boots. Now on roughly 1200c7. QPI up a notch to hold this, 8 loops passed.

Trying to move to 1600mhz, 

7/Try vcore 1.3, qpi 1.355V, dram 1.66. Ram set to 8-8-8-24-2. Optimistically going to set this to 1600mhz ram. Post, no boot. 
8/Repeat at 10-10-10-24-2. Post, no boot. Repeat at 8-8-8-24-2, 1.325Vcore. Post, no boot. 
9/Vcore 1.325, qpi 1.355 or 1.375, dram 1.68 no signs of getting into windows. Giving up on 1600mhz for a while, something I'm missing here. Clock drive perhaps

Aiming for 210x20 @ 1260mhz

10/210 bsck, vcore 1.3125, qpi 1.315. 1260mhz @ cl8. Froze within a minute of ibt. vcore to 1.325, system locked up. qpi to 1.335, vcore to 1.35 freezes a little bit further in. 
11/Decreasing pll to 1.5 removes all trace of stability. The freezing is similar to 210+turbo with 6gb.
12/increasing IOH, vdimm makes no difference. Trying 1.35V qpi, vcore 1.375V. 10 mins a loop instead of 20 now. XS suggest pcie clock (and possibly TRFC) is the bit I'm missing. Well, it didn't freeze. At about 60 degrees, ten loops at 11052mb stable at the second specified settings.

Pushing for 4.4ghz rather than for 1600mhz.

13/Turbo enabled boots to windows, expecting ibt to freeze shortly. It doesn't, rebooting instead of freezing. Vcore limited instead of qpi I think. 
14/ Regarding pcie. 105 won't post, needed a cmos reset. 102 is working better than 100. With 102 pcie and a notch on vcore ibt crashes but windows stays upright. More vcore perhaps. pcie 103 no better than 102. 
15/pushing more voltage, 1.4V vcore and 1.355 qpi. Managed a pb of one loop of ibt before the program crashed. 
16/Trying +800mV cpu clock drive in the hopes of avoiding more vcore. Marginal improvement. 
17/800mV clock drive + a bit more vcore has got 4 runs ibt stable. 
18/50ps cpu skew stops post and causes freezing in the bios. Recovered with cmos reset, much love to gigabyte for putting a button for this on the back panel.
19/Testing at 900mV clock drive. Seeing 1.35Vcore in windows. 5 passes, not convincingly different to 800mV. 
20/vcore 1.4125, qpi 1.375. Threw a hissy fit, refusing to post. I think coincidence, but maybe tied to the 900mV clock drive. A couple of cmos resets later it's back up and trying one notch more on vcore. System froze after 8 loops rather than ibt crash.
21/More qpi. Hard reboot, oddly. Ran a series of tests at 4.2, think this is my new 24/7 clock. Blaming turbo for the instabilities. 

Increasing bsck beyond 210 with turbo off
22/Tried 215 bsck, vcore 1.4 starting from stable at 4.2 settings. Hangs on post, needs cmos reset.
23/Frustrated. At 212 bsck I can't even get it to fail consistently. Some reboots, some freezes. Seemingly voltage independent. Going to leave it stress testing overnight at the believed stable 4.2ghz settings and pick this up again tomorrow. Discovered my computer had rebooted in the night.  

24/Bios deleted my profiles and pump saw fit to stall. Now trying at 1.4V qpi, don't think I want to go over this. Finally saw cpu temp go over 70. A bit sick of stability testing so I'm going to play with cpu-z validations for a little bit. Concluded from this that I need to set the right pll, trying 1.7.

25/About four loops stable at pll=1.7. Had an interesting blue screen which is either drivers or a broken cpu, installed drivers and not seen it since. Somewhere over 1.4V vcore at present, seeing 1.36V under load. Plan is to find stability then bring this down. Just had a different blue screen, looked like memory failing under my old P5Q so trying more vdimm. Think the higher qpi needs the higher vdimm. Linpack crashed after 2k s, no blue screen. 

26/trying 1.6 pll despite suspicion that vcore is needed. Stable for 600s, so going back to 1.7. Clock skew 50ps, refuse to boot, cmos reset needed. Sense of deja vu. 900mV cpu clock drive, same bluescreen previously blamed on ram. cpu clock drive back to 800mV.  

27/Decreasing qpi a notch, same bluescreen. Trying more pcie clock drive. Rebooted after 5 loops.

28/800mV clock drive on both, touch more Vcore and I'm stable for 15 loops at maximum memory. 

29/cas7 didn't make it into windows, one notch down on qpi or vcore crashes ibt.

30/unstable in normal use, needs one more on vcore or qpi. trying vcore first. 

31/said instability causing trouble. Dropped to 4ghz on fairly basic settings, 1.3V vcore/qpi, x6 ram. Checked stability over a fortnight or so of folding & general use, seems good.

Ram testing
6gb at a time, starting with v3.2. Vdimm <1.7, qpi <1.35 limits. Unclock multi to twice ram multi. Frequencies quoted are as sold, i.e. 1600mhz = 800mhz, I'll halve them all if I see reason to. Settings to auto except for the big five.

Testing frequency limits for diagnostic purposes, only values I'm particularly interested in are tightest timings at 1320 and at 1760 for obvious reasons.

>5 mins Memtest stable, 3x2gb v3.2
1750 c8 t1 pass, boots at 1800 but errors in test
1550 c7 t1 pass, boots at 1600 but many, many errors
1300 c6 t1 pass, boots at 1350, a few errors after a while

>5 mins memtest stable, 3x2gb v2.1
1650 c8 t1 pass but 1700 c8 t1 fails
1550 c7 t1 pass, fails to boot at 1600
1350 c6 t1 pass, multiple errors at 1400 c6

>5 mins memtest stable, v2.1 in primary, v3.2 in secondary slots
1650 c8 t1 pass
oddly 1700 c8 t1 passed 20 minutes even though one set couldn't do this by itself

References
Overvolting courtesy of ocforums
Electromigration from a metallurgy lab.
QPI voltage strongly suggests mismatched ram will be fine if I hit it with 1.45V qpi. Not sure I want to though.
QPI/Vdimm graveyard from xs
DDR3 frequency vs timings by Anandtech, concludes that 1600mhz c6 is ideal but broadly speaking that they don't matter.

Conclusions/theories
1600mhz at 4ghz is very difficult. I suspect the issue is differences in my ram as Paul managed it with matching sets. Need to try this on lower bsck.
Vcore and qpi take you to 200bsck fairly happily with the x6 ram multiplier.
210 bsck seemingly needed more vcore, qpi but nothing else. Suspect ioh increase to be a placebo.
210 bsck with turbo on benefits from pcie of 102 or 103. Needs significantly more vcore.
6gb with turbo gave random freezes at 4.4ghz. Now suspect more qpi would have fixed this.
pll changes can help, mine has to be 1.7 but this seems to be different for each chip. Similarly cpu/pci clock drive.
qpi and vdimm related, increasing one seems to need the other increasing and vice versa.
IBT (v2.3 at least) isn't very conclusive, ibt stability is a lot easier to get than system stability

JonJ678 · 1 Dec 2009 at 09:06

Cheers. I think theres something to be said for being as thorough as possible, this will probably take me a few weeks. I've had no luck at all pushing 12gb at 1600mhz, not yet pushed them in isolation. YGM btw, I don't want to push high voltages through the ram if it's possible I'm shipping you one of the sets. They're behaving nicely enough at 1260mhz, will put time into ram after I've learnt more about the cpu. There's a fairly convincing argument for 1333mhz c6 online so I may end up going for that.

I think there's something to be said for pci-e frequency, but perhaps not until significantly over 200bsck. I think I'm going to leave it at 102 or 103 for a while. IOH at 1.1 was a problem, but nothing over 1.14 seems to make any difference. TRFC at 100, will tighten it later.

What sort of voltage limits were you working to? I have a suspicion that qpi will have a chip dependent sweet spot around the 1.35V mark but it'll take as much vcore as i'm prepared to offer it.

Oh, not slept yet. Was CS:S good?

JonJ678 · 1 Dec 2009 at 15:42

@TB that may be a sign that I should start using auto more, or that you have a better imc than me. I'll stick with manual for the time being I think, though some of the settings could doubtless be handled better by the board. Do you happen to know what qpi and vcore it's using for this?

@Paul same voltage limits here, more willing to push vcore than qpi at present if only because I'm drawing an analogy between qpi and P45's nb where excess nb voltage caused me no end of problems. I'm at 2T, I confess ignorance regarding ram though. I have a vague concept of lower = better and that some numbers should add up to some other numbers, got to change that in the near future.

@Dave thank you man, I'd forgotten about that. I'm not using optimised defaults for the simple reason that I don't want it changing things that I don't know about, but clock drive is something I haven't touched yet. Next time I'm on the edge of stability I shall try this. In fact that's now, ibt has just failed on me.

I'll be very happy if someone benefits from this. Possibly more happy in the short term if it would stop failing ibt though.

JonJ678 · 2 Dec 2009 at 06:10

Thank you spikey, always interested to see what other people are running. Increasing clock skew was ruinous for me, otherwise our results are similar.

Very nice latencies on the ram, 1200cas6 is what I'm hoping for. What's your cpu temperature like, and are you going to try for any higher?

JonJ678 · 2 Dec 2009 at 17:03

@Paul, that's odd. Account seems to be behaving otherwise, could try through trust or my username @ hotmail.co.uk.

@Dave More than 1.66 vdimm doesn't seem to help, I think because the ram is running so much under spec at present. 1.7V didn't get me 1600mhz stable unfortunately. In truth I'm pretty much ignoring the ram while I mess around with the cpu clock. 4.4ghz is eluding me so far, starting to blame turbo again though it might just be time to play with clock skews. That's a good thread, though it is one I've run into. I think I'll add it to the references. And thank you, trying to explain my thinking. It's making for a very long OP.

@Spikey the general rule is "if it's stable, it's fine" though that's not erring on the side of caution. I think you should increase voltage until you hit diminishing returns then back off a bit, while largely ignoring temperatures. It'll start being much more difficult to get it stable at high temps so it'll mostly look after itself. Otherwise on air people try to stick below 80 in realtemp/coretemp. I'm loading at about 65 using water.

JonJ678 · 2 Dec 2009 at 17:18

Different temperature being measured, easytune is reading the case temp (bios reads this, keep it below 70 degrees). Software takes an estimate based on distance to tjmax, on i7 this is a different measurement to the bios one. So 70 according to easy tune/bios, 90 according to other software for the sane limit.

JonJ678 · 3 Dec 2009 at 09:34

Took the thing apart to move fans around, and in an interesting twist, it's now refusing to post at 150 bsck and has deleted all my bios profiles. Going back to stock for round 2. Odd.

edit: appears my pump wasn't running, tubing too kinked I think. Bios profiles still gone but otherwise it's sort of behaving itself. Need to get a flow meter of some sort.

JonJ678 · 3 Dec 2009 at 13:43

My first blue screen!

"A clock interrupt was not received on a secondary processor"

Internet seems to think this is vista being ****, which doesn't really help me. Oh, or that the processor isn't stable. Needless to say I'm enraged that "processor isn't stable" seems to be considered a complete answer. Not sure what to do about this. Installing drivers and rebooting, otherwise will try to get stability from clock drive and pll.

@TB I wonder what qpi voltage auto is setting

JonJ678 · 3 Dec 2009 at 17:07

Interesting, cheers Paul. It appears to have gone away with installing drivers, though I'm now seeing a different one. Told windows to stop rebooting so I'll be able to google it properly next time it shows up.

Throwing vcore at it doesn't appear to make it any more stable, I think the important thing to get now is the correct balance of qpi and vdimm, and hope this isn't a function of vcore.

JonJ678 · 4 Dec 2009 at 15:47

Something seems to be working, I'm 11 passes into ibt at 4.4ghz. Really hoping this turns out to be stable, there's some voltages I'd like to lower.

Still, for the first time in a few days I'm feeling confident about this working out well. 4.4ghz with 12gb is not a bad effort at all

JonJ678 · 4 Dec 2009 at 15:56

It's with turbo on (boo

). And yeah, I will probably try to repeat this with turbo off. At present I can't work out how to get over 210 bsck stable, I'll probably try turning turbo off and increasing bsck on the current settings in the near future.

There's still many, many voltages for me to play with yet so there's still hope for a true 4.4ghz. So much still to learn. Might be playing with ram next though, and that's probably an even bigger minefield than the cpu.

Apparently 5 hours of ibt means **** all. Multitude of crashes, some more visually interesting than others. Basically it can't cope with heavy processing load combined with multitasking, which is unfortunately what it spends most of its time doing. Down to 210x20 and solid so far, got a fair bit of work to do for a few days so will revive this thread and try to finish off 4.4ghz when I have time.

Damn.

JonJ678 · 19 Dec 2009 at 07:07

Cheers guys.

I'm currently at 200x20, 1.3V vcore, 1.3V qpi, turbo off, 1200mhz ram. This is far from the result I want, but as ibt has proven inadequate I'm trying days of real use and folding as a stability test. I think it's doing ok, the plan is to leave it completely alone for a week or so (running f@h) and see how it copes. I'll just attach the monitor to my laptop.

I think the freezing is related to the memory controller deciding enough is enough. I'm basing this on more qpi voltage generally helping, and on crashes/freezing mainly happening when a lot of data is swapped into or out of the ram. Starting virtual machines kills it fairly consistently for example.

Current belief is that ibt + prime plus a good week of normal use is needed to consider it stable. This is then reconsidered if it ever shows signs of instability that can't be put down to software. Oh well, I didn't expect to make it over 4ghz very quickly. Next thing to do (after the weeks testing) is to see what my ver 3.2 dominator can do and give PaulyD a shout.

Good luck guys

It ran folding@home for a week without any issues. Earnt me considerable numbers of points in that time, so I think it's probably time to try for more than 4.000 again.

JonJ678 · 2 Jan 2010 at 21:11

Could you post your full bios settings Brightside? The most obvious omission there is qpi, if it's still at 1.15V or so that's probably the issue.

Are you using load line calibration or anything else similarly exotic?

As a final check, this crash was with windows 64 bit host system, 64 bit linux in a virtual machine? bigadv seems to be very hard on the system, my hunch is that it's swapping such large amounts of data in and out of ram which causes the problem. Certainly my only issues at 4.4 were when combining bigadv with virtualbox, and my system is very sensitive to qpi changes.

I'll do my level best to help, another guy folding bigadv would be excellent

JonJ678 · 2 Jan 2010 at 23:01

1.35 may be too high actually. There's a balance between vdimm and qpi which seems to be processor specific, 1.3 qpi and 1.66 vdimm works for me, but 1.375 qpi needs vdimm of 1.7 for stability.

Other things to check are pll, while it starts as 1.8 and intel say 1.88V max, lowing it seems to help. I'm most stable at 1.7, other people have noticed a benefit to lowing it as far as 1.3.

Should have put this first really, but what's your uclock multiplier? It has to be at least twice the ram frequency. With x6 ram which I believe you're also using, x12 won't boot for me. x13 won't post either, x15 seems to be good.

12gb just makes everything a little more difficult.

JonJ678 · 2 Jan 2010 at 23:42

Cheers guys. The OP is getting a little incoherent, I'll reformat it in the nearish future. "Simple" 4ghz appears bombproof, I'll have to update the OP with them. Now testing 4.2 with ibt v2.4, hoping the new version will prove a better stability test than the last one.

You'll be alright MI, 12gb is exciting. How hot are the board heatsinks running? I can't get anything like these results without a 120mm fan aimed directly at them

edit: bluescreen => memory subsystem failing? I think that's how I interpreted it on my P45 board anyway

JonJ678 · 3 Jan 2010 at 00:28

Some of those look pretty high to me, but I'm struggling to translate evga notation to gigabyte. Is cpu vtt the same as qpi voltage?

Those are pretty scary chipset temperatures, I forgot you're using an matx case. Don't suppose anyone makes a waterblock for the matx evga board?

JonJ678 · 3 Jan 2010 at 00:41

Ah, I saw that. Didn't have a water channel over the mosfets iirc but achieved good results regardless. Dekez may do one at some point but I'm nothing like skilled enough to I'm afraid. Fans are about the only option then

I'll have a look as well, someone's probably written a gigabyte - asus -evga translation page. Otherwise I guess we get to.

JonJ678 · 3 Jan 2010 at 00:59

QPI PLL VCore (default: 1.1v, <1.4v is pretty safe)
What it does:

Keeps on-chip memory controller in-sync with bclk.

When to raise QPI PLL VCore:

* Try raising this along with Vcore and VTT, but in smaller increments.
* Helps stabilize higher CPU Uncore frequencies and QPI frequencies (in CPU feature)
* Try raising this when you increase memory clock speed via multiplier.
* Try raising when LinX produces errors after a few minutes without BSOD

I've done nothing to this at all, appears I may wish to. Think I'll try it in the near future.

Google confirms cpu vtt as qpi voltage. Intel say nothing over 1.35 at any time, google thinks under 1.4 is fine but 1.45 is pushing it. 1.5V is higher than I've seen anyone using outside of phase, certainly not on 3 sticks of ram, this may well be hurting your stability. That said, xs reckons anything under 1.6V is fine so don't be too concerned

JonJ678 · 3 Jan 2010 at 01:41

Not a fan of overclocking really slowly and carefully then

If you've got some time to spend on it you can probably get 4.0 with just vcore/qpi increases and the right uclock/qpi/ram multipliers. Then put pci-e frequency up by 2 or 3, put bsck up to 210 and start increasing vcore/qpi again. Good odds that'll be all you need to do.

Hopefully I'll be stable around 4.4 when you get your new corsair and I'll know more.

4 loops ibt at 4.2, then bluescreened. irql less than or equal to. Upped qpi & vcore a notch and will repeat. Five loops...

JonJ678 · 3 Jan 2010 at 02:56

Don't think more volts will help tbh, more likely lowing volts will be the way forward here. Hard to be sure though, as I don't know what settings you changed

As you may have noticed, I overclock slowly when I'm unsure what I'm doing. I quite like the approach of finding a profile from someone with the same hardware then tweaking it to match yours, but I think there's something to be said for the learning process.

Temps should be a lot better now qpi is lower than 1.5, either vcore or qpi increases do bad things for cpu temp. I'm not sure how it scales with temperature, you may find lower voltages => lower temperatures, and it's then stable at these lower temps but not at the higher ones before the voltage drop. Turning on vdroop will probably lower mosfet temperatures, you'll need to set a higher vcore in the bios but probably won't see higher cpu load temps.