Overclocking my i7 920?

[bumfire]

Currently doing another Orthos stress test with the voltages pulled back a bit.

 

CPU down to 1.36875 selected, e-leet reading 1.34 under load

DRAM pulled to 1.65 (rated voltage) at rated speed

 

4 copies of orthos with affinities set to stress all 4 cores (1 as "CPU and ram", 2 as "CPU and some ram" and the 4th just CPU)

 

I get these temps:

CPU 52

VREG 67

System 29

Cores are reading 66/64/63/62

 

Anyone think I should tweak anything further? This is getting to be a range I'm happy with.

 

Thx.

 

Teej

 

I would stick with that if its for everday usage, but I think your cpu voltage is a little on the high side.

 

I run my 930 at 3.6 something and for WEI I get 7.7 for everything except my video card which gets 7.5, so my base score is 7.5 :(

 

using the kit in my sig.

 

What was your cpu freq again ? still at 20x200 ?

Edited December 15, 2010 by bumfire

[Teej]

I would stick with that if its for everday usage, but I think your cpu voltages are a little on the high side.

 

Thanks...one thing I've learned is that I should mention...I have dialed the bios down from that 1.36 whatever to 1.34375....but more importantly that's with vdroop on - under load, vcore is down to ~ 1.32.

 

Still higher than some have had, but with the temps being so cool under a load I'll never see...I'm not worried at this point. It should still outlive its usefulness. :D

 

Def appreciate the input though. I did some overclocking in the early days, but I think the last system I really overclocked was a spare Celeron in the days when the 300mhz part easily went 450.

 

Yeah, it's been a while. :D

 

 

Yeah, I'm running 20x200

Edited December 15, 2010 by Teej
eta

[bumfire]

TBH, I would actually dial it back a touch, as your cpu will probably last, but I think with havign such a high OC all the time, I think your motherboard might actually give out before your cpu does.

 

I personally would dial it back a touch to say 3.5/6 similar to what I have mine at, but keep a profile for 4gig, so if you want to jump ingame, you can just set the profile on the fly and you have 4 gig without having to reboot etc.

 

Thats what I would do, because anything over 3 gig, with these cpu's you just dotn notice it for day to day stuff, only when you game or do heavy video editing etc will you notice that you pc performs better with an overclock over 3ghz, but for day to day stuff, anything above 3gig you simply cannot tell the difference, i.e using your browser at 3gig is exactly the same as it is at 4gig.

 

So personally, I would downclock what you have, as you know your cpu can hit 4 gig and handle it without taking a giant poop on you, so for the sake of longevity, I would dial it back until you actually needed the extra speed.

 

Thats just me and how I would do it, I know others like to run maxed out 24/7, but these cpu's are so good they will be still at the top or near the top of the tree for a few years to come, and their is no point in killing it by running it at 4gig 24/7 and only getting the full use of 4gig maybe a couple hours a day when you play a game.

 

As I said in another thread somewhere, I took mine upto 4 gig to see if it could do it, which it could, but I havent got the balls nor the money to run it at 4 gig 24/7.

 

You have a good chip there, let it live longer by selectively using the 4gig OC when you really need it, your cpu and motherboard will thank you for it in the long run.

 

Anyway, back to the voltages, I know you have a 920 and me the 930, but they are basically exactly the same underneath, with mines at 3.6ghz, I have my vcore at 1.180v in the bios, once in windows it goes down to 1.168v as show by cpuz and a couple other apps.

 

I know we are at different clocks and each chip is different, but to me 1.36875v is a little high and something I wouldnt be happy running for long, but its good to see you were able to bring it down to the 1.32 range, given time you might be able to find a sweetspot just below the 1.30 mark, say 1.28-1.30 ? I think that might be possible, if not 1.32v is alot better than 1.37v ;)

 

Your temps on all cores for 4 gig are excellent, especially with a H50, that is really good temps for that especially if its when all cores are under 100% load.

Edited December 15, 2010 by bumfire

[Teej]

TBH, I would actually dial it back a touch, as your cpu will probably last, but I think with havign such a high OC all the time, I think your motherboard might actually give out before your cpu does.

 

Interesting thought. Why would that happen?

 

I'm running the PCI clock and RAM at stock speeds...

 

voltage regulator is staying under the 70C temp I've heard as a "best below" line...(It gets into the 60s under 100% load, but tends to stay 40s/low 50s at idle or in a DCS engine that primarily uses a single core)

 

What's getting stressed on the MB?

 

I did find one area I was venturing into danger zone territory though...for a little while I was having > .5v between vtt & vdimm. (I've since narrowed that gap).

 

I'll have to see how low I can go if I turn vdroop off. The way I'm understanding it...with vdroop (where I am now) you set a max voltage (in my case currently 1.34xv) and the system may/will drop below that under load. With vdroop off, you're setting a floor voltage and the system will end up running a bit above that point.

 

Given that...I'm guessing I could probably drop down to 1.3 or lower w/ vdroop off, knowing that as the load went up the voltage would rise a touch instead of drop. I'll save what I've got and give that a romp.

 

As to the temperatures...it probably helps that we're in winter now and my house temp is in the low 60s F (so...uh...about 17C...mebbe slightly warmer in the mancave)

[EtherealN]

Depends a lot on the mobo, I would say. One of the things in mobos that can die early is the voltage rails feeding the processor. It's not as bad as it was in the old 775 days and it's ancient FSB, but it can still be an issue.

[Teej]

This is a good thread...and I'm being a bit verbose for several reasons...none of which is that I think I know everything. :)

 

I'm documenting this so that others (eg Mower who's been following along) might pick something up...and so that others with more OC experience on recent processors might find fault in what I'm doing. I've found y'all quite helpful and hope some of what I say pays it forward.

 

So I went ahead and disabled vdroop, and right away turned the bios down from 1.34xx to 1.3v. Once I got the system up and running with orthos, I started pulling out voltage using eleet every few minutes of stable running.

Got down to a full .1v drop from where I was before I started running into issues (I had even it pulled down into the 1.23 setting...briefly). Now, FWIW I don't consider it 'working' unless I can peg the processor at 100% with Orthos and still bounce around in browsers and apps (obviously orthos is at lower priority so when I do something, the system responds). If it's not stable, it's no good...even though nothing I ever do in the real world (gaming, office apps) will thrash the system like Orthos does.

 

After a bit more playing...I figured I'd shut off hyperthreading - been meaning to and just kept forgetting. I was a bit surprised to find it shaved a full 5-7* C off my core temps. My temps were pretty cool before, but now...whoa...

 

After 10 minutes of 100% operation (the temps stopped climbing after the first minute or two)...the cores are at 53 / 49 / 50 / 49 with a CPU reading of 41C and a system 30C

 

Seems I have had to back off the timing on my RAM a bit even at stock speed to achieve max stability with the OC though. While it's supposed to be 8-8-8-24 @ 800, I'm currently running 9-9-9-25 @ 800. This let me pull back some RAM voltage as well.

 

So currently I've got VCORE set at 1.25v and under max load it's rising to 1.29. I'll certainly agree that's far healthier than letting it idle at 1.35 and pull down to 1.32 under load! The temps are just...stunning. 20 minutes now and no further changes in core temps. Assuming it gets to 30 and still doesn't crash as I bounce around a few websites...I'll be quite pleased.

 

Yup, it made it and then some. Now that it's cooled down from stress testing... I see CPU @ 18C (room temp!) with the cores at 31-34, 1.28v.

Edited December 15, 2010 by Teej

[Steel Jaw]

Seems I have had to back off the timing on my RAM a bit even at stock speed to achieve max stability with the OC though. While it's supposed to be 8-8-8-24 @ 800, I'm currently running 9-9-9-25 @ 800.

 

I tried EVERYTHING except that.

 

And it worked !!

 

Crap, cant beleive I forgot about RAM timings.

 

Stable o/c right @ 3.4.

 

Now what kinda performance hit would I see for loosening those timings though?

[Teej]

Nice! Glad to hear you got it going (faster)!

 

Stretching the ram timing does have an effect, but not all that much in most cases. Remember, there's plenty of memory sold at a given speed that has those timings stock. :P

 

You/I/both might have better luck with a separate fan blowing on the ram, given how long it took mine to go unstable.

[Steel Jaw]

The o/c is stable BUT 64 bit A10C now runs very choppy, so a big performance hit is the result so far.

[Teej]

Hmmmm. I've only tested FC2 since dialing back the ram timing...testing now.

 

OK, I can't discern any significant difference between mine with stock timings and extra voltage vs current. My "test case" was to use the takeoff training mission. I've done this a few times through the overclocking process so I know what to expect.

 

With my graphics settings, when I start the mission I'm sitting on the runway at 28fps. As I accelerate to takeoff, it generally hits 50fps by the time the wheels come up (more of the airport behind me) and locking in at 60 (maxfps) over the departure end. Circling back around over the airport and town I drop to 27-30 again....which is unchanged from my "best case" overclock.

 

Actually, that was 32 bit. Testing in 64 bit actually gave me a small boost again. It did early on...then for some reason I had identical performance...but now the boost is back, primarily over the town, less so on/over the airbase.

Edited December 15, 2010 by Teej

[Steel Jaw]

Disregard: Realtemp running with A10C was causing the jitters.

 

FPS are a little better but nothing exciting. Prolly still a GPU issue.

[Teej]

The o/c is stable BUT 64 bit A10C now runs very choppy, so a big performance hit is the result so far.

 

Let's see your e-leet cpu, memory, and overclocking screens...like these. Maybe throw in voltages too.

 

[Teej]

Ah. Gotcha. Yeah, A-10 definitely needs some GPU horsepower...although you can tweak a lot of settings. I personally don't go above 4x on aniso or AA - I can't tell the difference higher than that. I also don't use the HDR setting.

[Steel Jaw]

I like your clock setting.

 

I have left all voltages on auto except dim which is at spec 1.65 v and +150 vtt mv.

 

What is your vtt and cpu voltages.

 

EDIT never mind, you have a 950, I have a 920.

Edited December 15, 2010 by Mower

[Teej]

From what I've heard the 920 should likely be capable of 4.0 depending on what you have for a cooler.

 

I went with the 950 because there was a bigger sale on it at the time so it only cost me about $20 more than a 920 would have, and in case I never OC it, it was 15-20% faster stock.

[Steel Jaw]

Well I got my 920 to 3500 with a +200 mv vtt and dimm 6.50, timings opened to 9-9-9-25. Stable. Good heat, never crosses 50'C. That's my comfort zone with out pushing voltage more since I cannot afford to replace the CPU.

[Teej]

I hear ya. That's not bad.

[RSharpe]

The o/c is stable BUT 64 bit A10C now runs very choppy, so a big performance hit is the result so far.

 

Make sure you have enough voltage. Sometimes I've been able to set the clocks high, and the computer is stable enough to boot up, but will start to stutter when processing if it's not getting enough voltage. This is very noticeable when hyperthreading is on.

[Teej]

I saw that too, Rsharpe...but didn't make the connection at the time. Completely forgot until you mentioned it.

 

Mower: I assume since you have the X58 it has the option for vdroop and stuff?

 

I'd suggest turning vdroop off...hyperthreading too, if they're on. It's really not gonna help A-10, whether it "hurts" or not (and I'd say adding 5+ degrees C on mine counts as "hurting" when I don't often run apps that can take advantage of HT) and start at about 1.25v. This is WAY BELOW intel's max spec (~ 1.3675 or something like that) so you won't cause any problems right there - that's not overvolting it.

 

You might need to bump up (or be able to bump down) a little from that...but leaving it on auto may not let it float up enough to stabilize your system.

Edited December 15, 2010 by Teej

[Steel Jaw]

If I turn off vdroop then that would allow the voltage to float around alittle when the CPU needs more/less...?

 

3.5 was freezing with +200mv and CPU volt 2.775...bumped it to 3.0 so I wonder if still safe to lturn vdroop.

Edited December 15, 2010 by Mower

[Teej]

Well...I hope you mean 1.2775 and 1.3...because if you put 2-3v on an i7 it would probably let the smoke out.

 

That said...you have to look at it this way...

 

With vdroop ON you set the maximum voltage. Under load, the cpu voltage will decrease...so you have to set a higher than desirable level - much as I was doing when I ran it at 1.37 and even up to 1.4 for a little while. That's not a good thing.

 

With vdroop OFF you're setting the minimum voltage and expecting the system will rise up a bit above that. So by setting mine to 1.25, it generally runs at about 1.28, touching up to 1.29 under 100% load.

 

The key takeaway is that you have to set the vcore higher than you'd like when using vdroop ON.

 

Even with vdroop off, it's not going to let it go all the way from 1.2x up to anything "dangerous". If you set it to 1.29, it would probably float up to about 1.33 or so (guess) which is still well below Intel's max volt spec.

 

I have several non-auto voltages, FWIW

On mine, I have vdimm set for 1.65 (spec for the memory)

QPI PLL Vcore got a slight bump to 1.125

QPI CPU Vcore did as well, 1.875 (stock is 1.8 and the allowable range goes well past 1.875)

CPU VTT got bumped to 1.250 - From what I've read, you need to make sure this is within ~ 0.45 of vdimm (ie if vdimm is 1.65, this needs to be at least 1.2)

ICH Vcore nudged to 1.1 (from 1.05)

Edited December 15, 2010 by Teej

[Steel Jaw]

Got that, thanx.

 

When I get stability @ 3500 I will post the specs.

 

Fine tuning with VTT as I do not want to raise the CPU beyond 1.3, vdroop now off.

[Steel Jaw]

What sort of max VTT is safe, I am now up to 275mv

[Teej]

Right. With vdroop off I would expect you would be able to pull it below 1.3. Certainly don't go above that w/ vdroop off.

 

I don't believe anything I've posted exceeds specs for the parts in terms of voltage, so I feel pretty comfortable with it.

[Teej]

I don't think I like the 275...but lemme either check the source I was reading or maybe someone else will chime in. That's getting added to 1.1v (at least on my board) which means your actual VTT is now 1.375...so don't raise it anymore at this point. Lemme get a link and follow up in this post.

 

Here's the major source I used:

 

http://www.evga.com/forumsarchive/tm.asp?m=100494809

Edited December 15, 2010 by Teej