**Official** Core i7 860/870 overclocking thread

[-n7-]

@-n7-: Are you running the latest 64bit version of p95? On every board I have, from the 775 quads to these 1156 quads it takes less voltage to pass linx than blend, have seen others post the same thing. Normally if it runs 20 passes of linx max mem it will require a small bump in vcore to pass blend without exception on about 6 different quads and 4 different boards.

Seriously?

I've found the exact opposite to be true for my Q9550/Q9650/i5 750/i7 860.

For me, LinX max RAM always requires more vcore than P95 anything.
I've always used the x64 version yes, presently on 25.11

I suspect it has to do with my 8 GB though, as max RAM LinX w/ 8 GB is far harder on the CPU than 4 GB.

[OC Nub]

Seriously?

I've found the exact opposite to be true for my Q9550/Q9650/i5 750/i7 860.

For me, LinX max RAM always requires more vcore than P95 anything.
I've always used the x64 version yes, presently on 25.11

I suspect it has to do with my 8 GB though, as max RAM LinX w/ 8 GB is far harder on the CPU than 4 GB.

Ahh sorry, didn't see the 8gb of ram, that would make a huge difference. I have only tested using 2x2gb kits.

Have been playing with this ud6 board today and got a validation of 5ghz. Vcore was insane.

[OC Nub]

@-n7-, I tried linx using 8gb today max mem at 7160mb, it failed on the first pass, so I went up on PCH voltage a couple notches from 1.080 to 1.120v, made it through 3 passes in 18 minutes before stopping the test myself and deciding 8gb is not for me, I feel your pain. I don't know if it would have passed or not but at 6 minutes each pass I wont be finding out.

[irev210]

Corsair H50 cooler
P7P55D-E Deluxe
G.Skill Ripjaws 2000MHz ram

Results: (after ~8 hours of LinX)
172MHz BCLK
21x mult
3.62GHz CPU speed
2064MHz Memory speed

1.25Vcore
1.35VTT
1.6Vdimm

While nothing extraordinary, this system is dead silent and extremely stable.

Small update on the overclock:

168MHz BCLK
22x mult
3.7GHz CPU
2016MHz Memory speed

1.26Vcore
1.35VTT
1.6Vdimm

Can your chip run that overclock with a little more vcore and less vtt... ~1.25v?

Well sure, if I slow down the ram. 6x mult I am using on the ram gives me 2016MHz, hard on the memory controller.

My biggest issue is heat. Since I am only using two fans in the entire case.

CASE WALL | 140MM FAN (1200MAX RPM) | Corsair H50 radiator | Corsair 120MM FAN (1500 MAX RPM)


The power supply is that new seasonic X650, which the fan isn't even on most of the time.

It's pretty sweet, when the system is idle, the fans are spinning <800RPM (and the power supply turns off). Since this has SSD's, it is just about dead silent.

[PolRoger]

Well sure, if I slow down the ram. 6x mult I am using on the ram gives me 2016MHz, hard on the memory controller.

Ahh... I missed the divider that you were using and was just thinking about your BCLK...

I don't have that divider available with my X3440 or i5 and I seem to need ~1.3v+ imc/vtt with ~210+ bclk

[bot@xs]

i have the ftw and 860 and i can't get 200 bclk.
got a shacky 190 but i am running stable 180x21 for 3781 mhz
but i have to use high voltages to get there.
VCORE 1.4v
VTT 1.4v
PCH 1.075v

is there any point of raising the PCH?
i would like to use less power or get higher clocks

[OC Nub]

i have the ftw and 860 and i can't get 200 bclk.
got a shacky 190 but i am running stable 180x21 for 3781 mhz
but i have to use high voltages to get there.
VCORE 1.4v
VTT 1.4v
PCH 1.075v

is there any point of raising the PCH?
i would like to use less power or get higher clocks

I set PCH 1.125v on that board to help with stability at higher clocks.

[bot@xs]

i was actually able to lower the voltages and raise bclk to 185 for 3885 mhz
voltages now are:
vcore 1.4v
VTT 1.35v
PCH 1.05v
PLL 1.65v

[Scaniris]

Hi,

I don't know if you know this thread: http://www.evga.com/forums/tm.aspx?m=20029

I used the settings suggested by the EVGA product manager, and run my i7 860 @4.2GHz very stable!



I'm not at home for the moment, but I can also take a screen shot of my voltages if useful. The Vcore is at 1,36V, never more than 1,38V. I had to icrease it a bit because under @ 1,30 (recommended starting point in the EVGA PM tutorial) it was not stable with Prime95 (some errors in 3 cores).

So you may test these settings maybe it helps

[bot@xs]

yes please if you could. i saw the thread you linked too, it's actually mention by the OP in the first post
somehow, is no worky for me. i finally got it to a stable 4ghz but i have to increase VTT to 1.425v
i am not sure if i am compfy with that 24/7, how sensitive are these chips to high VTT's?
i usually go through a intelburnintest and linx session before i prime .. is that or

[p2501]

I can only judge from my x58 i7, but with 12GB I'm running over 1.4v on VTT 100% load 24/7 for 9 months now with no problems whatsoever. Latest LinX session was 66 runs before I got bored. So I'd say all you have to do is keep VTT within the usual .5v from DRAM and you're set, your chip will take it if you can keep it cool. If it needs 1.425 to be stable then feed it just that. Your board will go down first if the PWM section goes flaky or catches fire...

[Scaniris]

yes please if you could. i saw the thread you linked too, it's actually mention by the OP in the first post
somehow, is no worky for me. i finally got it to a stable 4ghz but i have to increase VTT to 1.425v
i am not sure if i am compfy with that 24/7, how sensitive are these chips to high VTT's?
i usually go through a intelburnintest and linx session before i prime .. is that or

Ok here it is, example with OCCT CPU LINPACK 64 bit with HT on:



Link to display large image: http://www.hiboox.fr/go/images-100/s...31a38.png.html

Here it's only 2 minutes for showing the requested data, after 1 hour the water temp is around 34-35°C with a air intake temp of 23°C, I was not in my private office this evening so it's cold in there, 18°C, brrr.....

In the EVGA tutorial, they talk about that:

* Bootup CPU VCore - 1.30000V (good for starting point, will likely need to increase for 24/7 stability)
* Eventual CPU VCore - 1.30000V (good for starting point, will likely need to increase for 24/7 stability.)
* DIMM Voltage - 1.65V
* Bootup CPU VTT - 1.400V
* Eventual CPU VTT - 1.400V

@ 1,3VCore, not stable with high CPU load. I increase the Vcore power until 1,36V and something using the ECP hardware control panel and it was working. What I do not understand is that the Vcore auto adjust during stress test, but anyway if I setup VCore @ 1,36V+, no problem with stability.

There is another difference with the tutorial: I use native PC16K memory modules, so I don't OC the memory, it's running @ 2000MHz 8-8-8-24 1,65V without OC, it's the reference setup for the GT.

Something happens when temps are higher, for example during a Power Supply test with OCCT: if I run a Prime95 8 threads, I get some errors on 3 cores. I don't know why, calculation stops on Core 0 thread 2, and core 2 thread 1 and 2. Maybe because temps or voltage are too high? Its not reported on the monitoring... never went over 80°C even after a hard time testing all CPu/GPU together.

Do you think the dual 8 pin 12V power connector on the Mobo may help for stability? My PSU is connected on both of them.

[Scaniris]

Yes maybe something else: I don't use this setup for daily use!! I'm concerning with power consumption and earth care. So I usually use a dynamic OC, with C State, Speed Step and Turbo Mode on, Bclk @ 166, mem @ 2000MHz 8-8-8-24 1,65V. I have more power when needed only, but also downclocking depending on the perf requirements. And I'll also test a 2 cores only cpu, because I don't see any application (except maybe Premiere...) that take benefit of 4 cores, I have more benefit with higher clock speed, I'll check this. Because.... with 2 cores only, you can get 4 x 133MHz increments with Turbo Mode on. 5x with a single core. So...

Bclk @ 166 - 2 cores - Turbo mode on - Native cs @ 3486MHz - Turbo cs @ 4150 = under the maximum TDP for the i7 860.

With such a reference clock setting, I can use the mem @ maximum stock frequency/latency/voltage values, and get only when required some additional clock speed, keeping benefit of SpeedStep and downclocking.

Well it's theory, but I will test it this evening for sure With application to monitor real load and compare for example compression time for a video, or audio file. Because this is my daily use: gaming, audio, video... not benchmarking And the benefit of a higher clock speed is probably higher for my apps than having 8 thread parallely, I don't know if my apps are able to use so much power.

[Scaniris]

And here, a screen shot with the 3.5 oc WITHOUT SStep/Turbo/Sstep:

[kiwi]

wow your is bad i got mine early too
i can do 4.2 ht 1.4v

4Ghz bad? lol

[Scaniris]

And now, same thing but with Speed Step enabled, Turbo Mode enabled, and C state enabled, you'll see how this positively affect the CPU behaviour:

4 cores running with OCCT at full load, you get a x22 multiplier, only 1 step increase with 4 cores running:



Then... you'll see that the GPU Z is reporting higher multiplier, x24 and x25 which correspond to a dual core use, and single core use:






Strangely... the RealTemp Turbo monitor don't report the multiplier increase that CPU Z reports... but maybe with so many monitoring tools opened at the same time... OCCT is not happy when starting for example because it's based on CPUID tech for monitoring during stress test ^^

[Scaniris]

I think this may be a balanced setup. With C state controlling the core status, we can get a dual or single core running turbo mode at maximum possible factor, don't need to disable manually cores into the bios so... balanced mode? When idle... multiplier to x9. Dynamic, balanced, power efficient. I like this one. The Vcore may change from 1,18 to 1,48 @ x25 multiplier...

[bot@xs]

wow scaniris, thanks so much for all the info. thats some great work
lots of insights here

[Scaniris]

Well I hope this helps, I'm not really an extreme OCer, and I'm mainly looking for balanced configurations. Some articles @ Tom's hardare and PC World are really interesting about OCing the P55 compatible cpus, but they mainly talk about I5 750. The architecture is the same finally and some good indications are given in their reviews, good starting point to try different ways for having a balanced system, corresponding to our personal use.

In my opinion, the OC is also very helpful for cutomizing the system for our own use. Some people see in OC only a competition tool. I think OC is a value added tool for daily use. And if we want to take all the benefit of this new architecture, coming from the Nehalem revolution, we should also consider to use C State, Turbo mode, advanced speed step... the Dynamic Datacenter benefit is huge in enterprise architecture, and we, hobby and private user, are able to access this kind of technology, do you imagine the chance we have? I think it's great to use these new features. We have more performance, yes. Every week I'm so happy to make some Poc with customers demonstrating how Nehalem EP (and coming soon Nehalem EX) CPU beats some Sparc based systems, and some old multi cpu monolitic architectures for less money, less energy consumption, with same (Nehalem EX) reliability and HA features (EX is definitely positionned by Intel in front of the Itanium... in the past it wasn't same story with Dunnington...^^ something is changing isn't it?). We have a dynamic and virtualized architecture, in our home pc. Fantastic isn't it?

I don't think it's a responsible attitude to run 24x7 an OCed CPU. Not because of the CPU lifetime. I don't care. But because of the power consumption. We can play with OC, it's a funny part of this hobby, but we don't need it for daily use. Each watt we are using to increase the clock speed need another watt for cooling... Running a 24x7 OCed system with big graphic cards and be pride of it? Be pride of the power invoice that we receive at the end of the month? Well... I don't share this pride with others running 100GHz and more @ home, behind a ''good'' reason like folding@ home and stuffs like this. I don't agree with that, I mean participate to such programm is very good, but should be done with a responsible system, not with a power eater! Because density in datacenters is more efficient than splitted OCed computers.... I have another PC, Energy star 5.0 certified and partcipate to folding@home and change sometimes to others. I'll never use my gaming system on a 24x7 basis for that ^^

Why do I use benchmark tools? Very easy: for fun, and to check if the performance increase provided by some OC upgrade is really useful or not. This is a decision assistant, to confirm or not a setup. If I have to increase the power consumption to get only 2 or 3 more percent... I'll not validate the setting.

So... sorry for this off topic summary of my overclocking approach. Just want to give another point of view about how to use OC embedded features

[PolRoger]

I don't think it's a responsible attitude to run 24x7 an OCed CPU. Not because of the CPU lifetime. I don't care. But because of the power consumption. We can play with OC, it's a funny part of this hobby, but we don't need it for daily use. Each watt we are using to increase the clock speed need another watt for cooling... Running a 24x7 OCed system with big graphic cards and be pride of it? Be pride of the power invoice that we receive at the end of the month? Well... I don't share this pride with others running 100GHz and more @ home, behind a ''good'' reason like folding@ home and stuffs like this. I don't agree with that, I mean participate to such programm is very good, but should be done with a responsible system, not with a power eater! Because density in datacenters is more efficient than splitted OCed computers.... I have another PC, Energy star 5.0 certified and partcipate to folding@home and change sometimes to others. I'll never use my gaming system on a 24x7 basis for that

Interesting perspective...

Like "bot@xs" mentioned you have some informative posts and screen shots.

[bot@xs]

Bclk @ 166 - 2 cores - Turbo mode on - Native cs @ 3486MHz - Turbo cs @ 4150 = under the maximum TDP for the i7 860.

so with this setting you would essentially have 1 core running at x26, 1 at x25 and two more at x22 multiplier. exactly like the turbo mode should function and still achieve 4+ ghz

but only 1 core is running at 4+ ghz

[Scaniris]

1 core running = 5+ / core = max x26 / w bclk = 166 >> max 4.316MHz
2 cores running = 4+ / core = max x25 / w bclk = 166 >> max 4.150MHz
4 cores running = 1+ / core = max x22 / w bclk = 166 >> max 3.652MHz

That's what is reported by CPU Z during real life testing and I suppose that is fitting with the turbo and c state purposes combined together, otherwise how to claim for a x26 multiplier if 4 cores are running, 860 is not designed for it.

[axffox]

I've given up with my 860 and bought a Xeon X3440, batch number L927B244T.. I'll pop it in my other board. hope all goes well!

[Constantine015]

I bought a core i7 860 and it's not as as I thought that It will be,for OC ofc.I need 1.39 V for 4ghz,Linx stable,VTT is 1.35.Cooler is Corsair H50,temps-33 idle,80-85 full load.

[Raul-7]

This is the most I've tried. Not sure if it's stable, but it can run SuperPi and so far hasn't crashed under intense gaming sessions.

Part of the reason I haven't completely stressed is due to the very high temperatures I get. As soon as I get some Shin Etsu X23 thermal paste I'll try and stabilize it.

EDIT: I'm debating whether or not I should keep this chip? I have an i5 750 on the side, but not sure if it's will stabilize with decent voltages.