QPI voltage

Souness · 22 Dec 2009 at 19:48

Got my i7 to 3.9ghz so far, without touching anything but the vcore.

I am planning to go higher but judging by the threads here, there seems to be a point where increasing vcore alone is not enough. I see other voltages quoted, most commonly the QPI voltage.

I'm a bit of an overclocking noob so I have played around with it a bit, to see if it had any effect. I tried whacking it up to 1.3v from the default of 1.15V (i think) and all that happened was my stable system now failed to boot! This happened all the way down to low 1.2x V, so increasing it does not help me overclock at all.

So yeah, it'd help me loads if someone could explain why you increase the QPI voltage (are there any signs when it's too low?), how you increase it (do you go up in increments or just whack it up high?), are there any negative effects of a high QPI voltage, and if it's always necessary in achieving high (4ghz+) overclocks?

My mobo is a EX58-UD3R, if it makes any difference.

Thanks

prt · 22 Dec 2009 at 20:29

there is a few good thread about oc on a udr3. got mine to 4.6 stable, however run 4.2 24/7 now

Souness · 22 Dec 2009 at 22:58

prt said:
there is a few good thread about oc on a udr3. got mine to 4.6 stable, however run 4.2 24/7 now

I've read quite a few good threads but I can't find one that tells you when you need to increase QPI voltage. Most just seem to quote numbers seemingly pulled from thin air, and I want to know the method/thought process used because when I set mine to a typical value it just won't boot.

prt · 23 Dec 2009 at 06:00

when i reboot my system tonight, i will have look see in my bios what settings i have got, as i rlly cant remember right now, been a while since i done it.

subzero · 23 Dec 2009 at 07:53

As far as i understand the QPI/DRAM Voltage is tied to the DRAM voltage and should have a 0.5v tolerance on whatever you have set your DRAM voltage to.

Its a safety feature so that you don't fry your chip i think.

So whatever your Dram voltage is set to your QPI/DRAM Voltage should be the same or can be lower by a 0.5v tolerance,but it should not be set higher than the DRAM voltage.

JonJ678 · 23 Dec 2009 at 17:51

QPI voltage is related to the memory controller on the chip. Consequently high ram speeds, tight timings or using all six slots require considerably more qpi voltage.

Intel specified maximum is 1.35V. Consensus online is to keep it within 0.5V of vdimm, which most people run at 1.65V as that's writtten on their ram. What voltage is required is a fairly complex function of ram, bsck, multipliers and temperature. Cpu temp tends to go up sharpish when you increase qpi. What vcore are you currently on?

If you are stable at 3.8ghz with 1.15V qpi, then increase it a notch when trying for 3.9ghz if vcore doesn't help. If however you are not stable at 3.8ghz, or have had qpi set to auto (thereby letting the board push whatever it feels like through it), then we cannot advise. Take it off auto and decrease bsck until stable, then overclock in small steps ensuring stability before increasing bsck further.

Have fun

Souness · 24 Dec 2009 at 03:42

JonJ678 said:
If you are stable at 3.8ghz with 1.15V qpi, then increase it a notch when trying for 3.9ghz if vcore doesn't help. If however you are not stable at 3.8ghz, or have had qpi set to auto (thereby letting the board push whatever it feels like through it), then we cannot advise. Take it off auto and decrease bsck until stable, then overclock in small steps ensuring stability before increasing bsck further.

Oh right, I didn't know that the auto setting was increasing the qpi voltage on its own (I foolishly assumed auto meant the same as default, doh!). This explains why setting it to 1.3v didn't work; I thought I was increasing it from 1.15v but in fact I must've been decreasing it from whatever auto had it set to. Schoolboy error.

Will auto push it higher than the intel limit 1.35v? If not then I guess there's no harm in leaving it alone, otherwise I guess I'll have to tune it manually.

Thanks for the explaination.

setter · 24 Dec 2009 at 03:48

Always avoid auto voltage, a lot of boards tend to feed things to much voltage if left on auto, this leads to higher temps which is never good.

Souness · 24 Dec 2009 at 16:58

Ok, I had a go at it today.

Previously I was running at 3.9Ghz stable, vcore 1.30v, vdimm 1.65V, and everything else on auto. I was 2 hours prime stable with these settings.

I changed the QPI voltage from auto and set it at 1.35. I got to windows ok then I bluscreened after a couple minutes of prime.

At a bit of a loss of what to try now. I tried the max recommended QPI and that didn't work. I have a chip that should be able to do 4 ghz with ease, yet I can't get it to boot at 3.9! Is it possible that I may have to increase the vcore when I set the QPI from auto to 1.35 too? What would you guys try now?

If it makes any difference my IOH is also on auto.

The advice is much appreciated guys, I have studied a bunch of overclocking guides but just get stuck at this point

.

JonJ678 · 25 Dec 2009 at 01:17

Start from a considerably lower bsck, most voltages manually set to default values, and move upwards in small steps only when the previous is stable. Otherwise you've no way of knowing which setting is causing the problem.

Critically more voltage doesn't always help, stability can also result from decreasing voltages from what was stable at slower speed. Having tried the "max" qpi voltage doesn't mean much.

Souness · 27 Dec 2009 at 17:37

Argh I think my chip must just be a dud then.

I tried like you said, starting from scratch without auto and moving up in small increments. When I get to 3.9Ghz it's just impossible to get it stable. I have the QPI set at 1.35V, which funnily enough seems to have been the limiting factor all the way through (I have hardly had to touch the vcore at all - it's currently set at 1.25v), and no matter what else I change I can't get prime to run for more than 12 minutes without bluescreening.

So I think the only way I can get 3.9+ stable is by increasing the QPI above the suggested limit of 1.35v. Kinda dissapointing as I thought it was from a decent batch of D0's.

Think it's safe to bump the QPI up another notch or 2 or should I admit defeat?

JonJ678 · 28 Dec 2009 at 00:14

What have you got your ram set to? Slower frequency/looser timings need lower qpi voltages so may be the way to go at this point

Souness · 28 Dec 2009 at 01:26

JonJ678 said:
What have you got your ram set to? Slower frequency/looser timings need lower qpi voltages so may be the way to go at this point

Atm I have it set to what it said on the box...

8
8
8
24
2T

It's 1600mhz corsair xm3 btw, so it's still underclocked. What timings would you suggest lowering it to?

robin66 · 28 Dec 2009 at 01:53

You could try 9 9 9 28 2T and test from there.

prt · 28 Dec 2009 at 06:01

what ram is it

Souness · 28 Dec 2009 at 15:56

robin66 said:
You could try 9 9 9 28 2T and test from there.

Will that harm performance much?

Bleh, tried with the lower timings @ 3.9ghz, 1.35 qpi, 1.65 vdimm, 1.25 to to 1.30 vcore (I was stable at 3.8ghz with 1.25, so I can't see this being the limiting factor), 1.12 IOH, and still can't get past 10 mins of prime without blue.

Guess I will try loosening timings further.