Core i7/X58 Overclocking Thread

[dejanh]

4ghz @ UNDER 1.3v load with HT on in LinX @ 210 bclk? Something is fishy!

No, actually my chip does the exact same thing. I just did a 20 pass run maybe 2h ago. Same clock, same voltage. I used memory at 1691, 7-8-7-20-1T timings, uncore at around 3800MHz give or take a few MHz.

LLC was on auto in my run. CPUZ reported voltage showed 1.288V (not completly correct, but then again, only a DMM is really correct).

If however I set 20x200MHz, I have to use roughly 1.3V.

It's funny that OC with higher BCLK actually works better than higher multipler. Completly opposite of what we should be seeing.

No wonder all the chips basically clock the same across the versions (920, 940, and 965 XE).

[SteveRo]

No, actually my chip does the exact same thing. I just did a 20 pass run maybe 2h ago. Same clock, same voltage. I used memory at 1691, 7-8-7-20-1T timings, uncore at around 3800MHz give or take a few MHz.

LLC was on auto in my run. CPUZ reported voltage showed 1.288V (not completly correct, but then again, only a DMM is really correct).

If however I set 20x200MHz, I have to use roughly 1.3V.

It's funny that OC with higher BCLK actually works better than higher multipler. Completly opposite of what we should be seeing.

No wonder all the chips basically clock the same across the versions (920, 940, and 965 XE).

concur - all chips about the same (I have the 965).
Also - lower multi with higher bclock also gives lower temps.

[tot31]

concur - all chips about the same (I have the 965).
Also - lower multi with higher bclock also gives lower temps.

That's really interesting, I'll test this theory.

[JWMc]

concur - all chips about the same (I have the 965).
Also - lower multi with higher bclock also gives lower temps.

I have seen many people say this but I'd like to know why this should make sense at all?

Higher bclk = higher IMC stress and more bandwidth to feed the cores at the higher core clocks. There's no reason I can think of for a lower multi with higher bclk to lower the required voltage other than some bizarre motherboard implementation thing. I'm still a noob though.

[Lukee]

I have seen many people say this but I'd like to know why this should make sense at all?

Higher bclk = higher IMC stress and more bandwidth to feed the cores at the higher core clocks. There's no reason I can think of for a lower multi with higher bclk to lower the required voltage other than some bizarre motherboard implementation thing. I'm still a noob though.

Unless when you increase the multiplier, the volts needed to stabilize it, are considerably more.

Take the E8600. It required more volts to do 450 x 10 compared to 500 x 9. The higher multi required more volts, which imo we are still seeing with the i7 processors.

[SteveRo]

I have seen many people say this but I'd like to know why this should make sense at all?

Higher bclk = higher IMC stress and more bandwidth to feed the cores at the higher core clocks. There's no reason I can think of for a lower multi with higher bclk to lower the required voltage other than some bizarre motherboard implementation thing. I'm still a noob though.

look at the pic attached.
don't ask me to explain it - but I get lower temps and higher oc at high bclock/low multi. All this was on air. (noctua 12 with 2x120 in push/pull)

[JWMc]

look at the pic attached.
don't ask me to explain it - but I get lower temps and higher oc at high bclock/low multi. All this was on air. (noctua 12 with 2x120 in push/pull)

Thanks Steve. I'm still learning new stuff pretty much every day. Sorry I didn't go over your data in detail already. I just felt like there might be some magic to the EX58-Extreme+965 Extreme+3x1GB modules+Not Vista-64 you were using because my results were so different to yours but with a very similar motherboard.

[SteveRo]

Thanks Steve. I'm still learning new stuff pretty much every day. Sorry I didn't go over your data in detail already. I just felt like there might be some magic to the EX58-Extreme+965 Extreme+3x1GB modules+Not Vista-64 you were using because my results were so different to yours but with a very similar motherboard.

The UD5 is an excellent mobo - just as good as the extreme for oc based on what I have seen
The 920 with F4j or m and with turbo turned on gives you constant x21 cpu multi - this is a big deal!
Do any of the other mobo's do this - ggbt did a great thing when they enabled turbo full time.
Don't be shy about the voltages or the temps - i7s can take it like there is no tomorrow.
Stay within the cpu max voltages - don't worry about the high temps for benching - the cpu can take it.
It will not thermally throttle until 100C!
see below -

[Pisklink]

Yes, it has to do with the physics and what these new CPU's and the Mem Bus and COntroller can withstand. It's a differential voltage problem. If you start getting away from a .5v diff between the QPI/DRAM (Vtt) and the DRAM voltage it could cause damage. This was Explained by Dr. WHo ina post on here early on right around release time. He also said that like all OC'ing to do so at your own risk.

In this case, I'd say the less diff the better. Since we usually deal with around 1.65v on the DRAM, we need to keep around 1.3-1.4v on the QPI/DRAM. 1.35v on the QPI/DRAM is the Intel max spec if I'm remebering right. I have 1.3v set in mine right now and it's fine on the rig in my sig. Be careful with Vtt though, becasue there is a limit, and if you go too high damage to the IMC can result. I don't think we know where that line is yet, but some have been running slightly over spec voltages with succsess although it doesn;t mean they'll be able to do that for long periods of time. It's all about the risk you are willing to take.

I'm on the more conservative side, and I'm very careful what I say about voltage because voltage can cure some problems, but it also kills these parts if one goes to high. AFter a bunch of reading, I've developed a feel for it, but I'm very careful what I do.

Many thanks for the eplanation and for giving your reference voltages. I'm on the conservative side as well but was hoping to achieve 4Ghz with my 920. Unfortunately I'm still waiting for some WC'ing components and wiring (8-pin extender). I hope to receive all these components this week, so I can assemble everything this weekend and start to play with the settings. I just wanted to make sure that I don't screw up my new components immediately by not complying to the 0,5v differential.

[Lukee]

No, actually my chip does the exact same thing. I just did a 20 pass run maybe 2h ago. Same clock, same voltage. I used memory at 1691, 7-8-7-20-1T timings, uncore at around 3800MHz give or take a few MHz.

LLC was on auto in my run. CPUZ reported voltage showed 1.288V (not completly correct, but then again, only a DMM is really correct).

If however I set 20x200MHz, I have to use roughly 1.3V.

It's funny that OC with higher BCLK actually works better than higher multipler. Completly opposite of what we should be seeing.

No wonder all the chips basically clock the same across the versions (920, 940, and 965 XE).

My voltage fluxuated from 1.288 to 1.296 during the Linpack test.

I just started trying to Oc this chip. My next step is to increase the memory to over 1600 Mhz and try to stabalize that.

I turned off HT and tried to get 4.4 stable, and just couldn't. I think I need to adjust something else besides vcore since I ramped it all the way up to 1.53 v and still unstable.

I'm using a new board. Asus P6T Deluxe. I'm familiar with the Gigabyte board more. It seems you have more voltages to adjust compared to the Asus.

The 940 is more expensive than the 920 and reacts similarly, but I'm selling my 920. I can't hit 210 bclk with it unless I increase qpi/vtt past 1.75V which is crazy. It does work really well at 200 x 19 w/ HT on though.