Core i7 920 Overclock...

Post by **Gilly** » 28 May 2009, 14:17

Unfortunately, Intel, AMD and nVidia want to push GPU tech and CPU tech as far as possible.

and also the game developers want to spend as little as possible to make games, and have to keep in mind, not everyone has bleeding edge hardware. So the game cant really be optimized for quad cores, and run like crap on the PC's that maybe 50% of the people will have.

You can argue the people that will be going to buy the game are probibly hardwared up, but not everyone is.

Post by **Paradigm Shifter** » 28 May 2009, 16:59

the DO is supposed to be a bit better at overclocking, but by no means a land slide from what I read.

Basically you woul want one if you were buying now but if you had the older one already you shouldnt lose any sleep over it.

But yeah a 4ghz OC sounds so nice, given that the i7 is clock for clock faster than my q6600 and an extra 400mhz OC that would increase my video encoding time by a good degree.

I cant wait to see what comes next and hope it overclocks as well as these do.

The D0 chips really shine with the amount of voltage they need to get to a given speed. It's lower than the C0 chips. However, once you break 3.8GHz, the amount of voltage you need to get 4GHz+ stable goes up dramatically. For example, my 920's stock volts (on the box) is 1.25v. I can get 3.8GHz on that. But I can't get 4.2GHz stable on 1.35v. 4GHz is almost stable @ 1.35v - a little more tweaking and I'd have it, I think, but temps are pushing it at 4GHz. I'd want fairly high-end watercooling to run 4GHz 24/7.

My 920 @ 3.2GHz encodes at the same speed as twin Harpertown's @ 4GHz, at 3.8GHz, it's about another 20% faster again. Which is bloody fast. ;)

...

I've read some bad news today - seems like Intel are going to kill off the 920 and 940/950 on LGA1366 in favour of Core i5... the only chips on 1366 are going to be wallet-raping Extreme Editions. :( Don't know if that's true, but if it is... :evil:

...

To be honest, I wonder if the 32nm chips are going to be even more voltage sensitive than the 45nm ones - HKMG doesn't like a lot of volts before it starts to degrade.

...

Vicious, with you saying about that thermal compound... I definitely want a tube of that now. How much is it?

Frag Maniac · Post by **Frag Maniac** » 28 May 2009, 21:06

...keep in mind, not everyone has bleeding edge hardware. So the game cant really be optimized for quad cores, and run like crap on the PC's that maybe 50% of the people will have.

It's not so much that it isn't possible to do both, more like so many PC game devs are still green at threading for quads effectively and expediently that they gloss over getting the general optimization done properly. No matter what kind of hardware the game can play on, there is still such a thing as general optimization that ANY spec can benefit from. Occluded texture compression, procedural rendering, careful scripting to avoid memory leaks, etc, etc. The list is actually much longer of things that can and should be done to optimize in general. I get the feeling many devs are learning on the job how to write for quads and because of it, losing touch with the fundamentals.

@PS,
That's also pretty much a sign that going beyond 3.8GHz is not really a good idea anyway, at least not if you care about long term stability and reliability.

I once posted a video tutorial NCIX did on OCing a Core i7 that was taken from advice written by one of the more tech savvy members of XtremeSystems.org. I got a response from one of the OC obsessed on the forum I posted it on saying it's ridiculous they advise not to volt them higher than 1.25v.

Mind you this is a guy that cares more about some OCing competition stats he brags about on a website that lists such things than giving tempered, wise advice to the average person looking to merely take an affordable CPU and balance it's speed to a top shelf GPU, without rendering it disposable.

Odd how this guy only ever raves about ATI GPUs for their extremely low prices when he can so easily advocate throwing caution to the wind like that. Seems those obsessed with OCing to that degree are better off getting things cheap though, because as the old saying goes, "The light that shines the brightest burns the shortest".

Post by **Paradigm Shifter** » 29 May 2009, 00:11

That's also pretty much a sign that going beyond 3.8GHz is not really a good idea anyway, at least not if you care about long term stability and reliability.

Pretty much, yeah. I'm not unhappy with 3.8GHz, anyway. Particularly how much faster encoding is with i7, even at 3.2GHz... ;) Even if I did watercool now, really the chance of me pushing this chip faster are slim. I'd be cooling for quiet and lower temps now... not more performance. Although there is something a little temping about saying '4GHz'. 8)

I once posted a video tutorial NCIX did on OCing a Core i7 that was taken from advice written by one of the more tech savvy members of XtremeSystems.org. I got a response from one of the OC obsessed on the forum I posted it on saying it's ridiculous they advise not to volt them higher than 1.25v.

Eh, given that Intel say the safe maximum is 1.35v, I'd probably say that 1.3v is 'safe' provided that temperatures can be kept down - i7 runs devilishly hot when the voltage goes up. At the same clock speed, an additional 0.075v means the difference between 70*C load and 85*C load. 70*C I'll live with. Even 75* on Core i7 given the limit of 100*C... but as soon as it breaks 80, it's time to back that chip down to cut down that heat. Or improve the cooling.

But certainly from my experience, the difference between 1.25v and 1.3v is fairly minimal in terms of OC achieved. So, therefore, unless serious overclocking is required, more than 1.25v is not needed. Of course, every chip is different. :)

Mind you this is a guy that cares more about some OCing competition stats he brags about on a website that lists such things than giving tempered, wise advice to the average person looking to merely take an affordable CPU and balance it's speed to a top shelf GPU, without rendering it disposable.

Odd how this guy only ever raves about ATI GPUs for their extremely low prices when he can so easily advocate throwing caution to the wind like that. Seems those obsessed with OCing to that degree are better off getting things cheap though, because as the old saying goes, "The light that shines the brightest burns the shortest".

Those who enter OC competitions (I haven't and likely never will) aren't really bothered about longevity, IMO. :)

However, buying cheap and getting lots of performance out of it always used to be the be-all and end-all of overclocking. At least until Intel decided to make their 'Extreme Edition' chips, and mobo manufacturers made 'overclocking motherboards' that cost the Earth... and so on and so on.

Overclocking always used to be about maximum performance for minimum expenditure.

Now, for companies it's all about bettering other companies... and for those with money, it's about having e-peen.

I'll always respect someone who can take a (within reason) budget board and (fairly) low-end chip through the roof more than someone who spends five or ten times that amount for only another 10% performance.

DaFox · Post by **DaFox** » 29 May 2009, 02:16

I'll always respect someone who can take a (within reason) budget board and (fairly) low-end chip through the roof more than someone who spends five or ten times that amount for only another 10% performance.

Reporting in.

Edit: Though my motherboard was like $280. BUT THATS ONLY BECAUSE it was the first P3X series board available in canada and I could not wait!

Edit2: My ram was by far the cheapest I could find though, Saved like $20 buying it.

Edit3: My 88gtx was free too.

Frag Maniac · Post by **Frag Maniac** » 29 May 2009, 04:01

Eh, given that Intel say the safe maximum is 1.35v, I'd probably say that 1.3v is 'safe' provided that temperatures can be kept down - i7 runs devilishly hot when the voltage goes up.

Keep in mind, safe does not imply consistent stability, it just means the chip won't be damaged with adequate cooling. The video tutorial I spoke of btw was using a 940 and water cooling too, so they were definitely of the mindset to error on the side of caution, which suits my way of thinking. When it comes to voltage limits, I like to leave some headroom just like I do with PSU wattage.

I had in mind a limit of 3.6GHz if I end up getting a 920 DO chip, though I may go as high as 3.8. I won't be using water, so I'll have to see what the temps are like. Even 3.6GHz though on a quad chip with QPI, MC, a fair amount of multilevel cache, and 8 thread capable HyperThreading is going to be screaming faster than what I have now, and plenty good enough to keep from bottlenecking the best GPUs, especially once devs start doing more threading for them.

packerfan · Post by **packerfan** » 29 May 2009, 04:38

That is cool, that diamond cpu paste stuff it is about three times the cost of artic silver.

Curiosity, how far does a 1.5g tube of paste go, like how many cpus can you do with a small tube of arctic silver?

Seems like interesting stuff if 1.5g of it can go far enough for the 7 bucks, but if it does save you a few degrees it could be worth it if you put it on lots of things.

Post by **Paradigm Shifter** » 29 May 2009, 12:28

Keep in mind, safe does not imply consistent stability, it just means the chip won't be damaged with adequate cooling.

Intel usually build in a safety margin on their published tolerances, so the voltage where electron migration starts to really damage the chip is probably more like in the 1.4-1.45v region. I've seen some of the heavyweight OCers get 920's up to some impressive speeds, but at the cost of feeding the chip voltages like 1.55-1.6v... way too high for my liking...

The video tutorial I spoke of btw was using a 940 and water cooling too, so they were definitely of the mindset to error on the side of caution, which suits my way of thinking. When it comes to voltage limits, I like to leave some headroom just like I do with PSU wattage.

Totally. I don't like have my PSU 'close to full' as it means there isn't any wiggle room. I'll never use Load Line Calibration on a CPU as it removes Intel's Vdrop/Vdroop from the board... while that might sound a good thing, Intel built in the Vdrop/Vdroop for a reason - voltage spikes when the processor switches from full load to idle state. Those spikes can be fairly high, and LLC makes them worse. Kill your chip, one state-change at a time. :roll:

I had in mind a limit of 3.6GHz if I end up getting a 920 DO chip, though I may go as high as 3.8. I won't be using water, so I'll have to see what the temps are like. Even 3.6GHz though on a quad chip with QPI, MC, a fair amount of multilevel cache, and 8 thread capable HyperThreading is going to be screaming faster than what I have now, and plenty good enough to keep from bottlenecking the best GPUs, especially once devs start doing more threading for them.

While your chip will likely behave a little different to mine, my temps @ 3.6 and 3.8 were the same, 3.8 didn't need any extra voltage, and running the CPU @ 200BCLK on the 19x multi meant that my RAM was running at spec, rather than slower than it was rated. :) So there is no reason not to try 3.8 if you don't have to feed it extra voltage. ;)

3.2GHz on an i7 is already really fast. Any faster is really just bragging unless you want something like video encoding done as fast as possible. (Which I do.) Or you have Quad-SLI or Quad CrossfireX that needs the fastest CPU it can to minimise bottlenecking from CPU.

Frag Maniac · Post by **Frag Maniac** » 29 May 2009, 21:33

While your chip will likely behave a little different to mine, my temps @ 3.6 and 3.8 were the same, 3.8 didn't need any extra voltage, and running the CPU @ 200BCLK on the 19x multi meant that my RAM was running at spec, rather than slower than it was rated. :) So there is no reason not to try 3.8 if you don't have to feed it extra voltage. ;)

3.2GHz on an i7 is already really fast. Any faster is really just bragging unless you want something like video encoding done as fast as possible. (Which I do.) Or you have Quad-SLI or Quad CrossfireX that needs the fastest CPU it can to minimise bottlenecking from CPU.

Well I hope I get the same results as you. Isn't the stock voltage 1.2v though? Seems you could try 1.25v and see how it goes. I don't think I'd want to volt any higher than that on air though.

Yeah 3.2GHz is fast, and would match up well enough to most single GPUs, but the more they make CPU intense games like GTA IV, the more it makes sense to go higher. It also allows for future readiness and helps the CPU get used to a higher speed, vs breaking it in at something lower than it can do.

From what I've read, CPUs tend to get comfortable with a speed you set them at after a while, just like a car gets used to the rpm you drive it in most. I certainly wouldn't want my CPU to pick up bad habits by getting too comfortable with a low frequency.

Post by **Paradigm Shifter** » 30 May 2009, 13:19

Well I hope I get the same results as you. Isn't the stock voltage 1.2v though?

My 920 at stock speeds (Optimised Defaults in BIOS) pulled 1.248v when idle @ 2.66GHz, 1.16v when idle at 1.6GHz (Speedstep) and 1.2v (2.66GHz, 100% load with LinX). So from that I'm guessing that the 'stock' Vcore is 1.25v...

Seems you could try 1.25v and see how it goes. I don't think I'd want to volt any higher than that on air though.

I'm at 1.275v (BIOS) 1.248v (CPUz) 1.246v (DMM actual) when idle, and 1.22v (CPUz) when loaded. So, basically, at 1.25v. :)

Yeah 3.2GHz is fast, and would match up well enough to most single GPUs, but the more they make CPU intense games like GTA IV, the more it makes sense to go higher. It also allows for future readiness and helps the CPU get used to a higher speed, vs breaking it in at something lower than it can do.

The Core i7 line can degrade quite fast if pushed too hard for too long. Although 'too hard' seems to be relegated to the 4.5GHz+ range...

From what I've read, CPUs tend to get comfortable with a speed you set them at after a while, just like a car gets used to the rpm you drive it in most. I certainly wouldn't want my CPU to pick up bad habits by getting too comfortable with a low frequency.

:lol:

I've not read that, myself. It's probably best to find a speed you're happy with when you first get that CPU, then stick to it. :)

Frag Maniac · Post by **Frag Maniac** » 30 May 2009, 22:52

Yeah I meant to say within reasonable temp limits regarding the speeds it gets used to. Odd that you're getting 1.25v stock, as most info I've read on these says 1.2v is stock.

I am resistant to the idea of going more than 1GHz over stock speed, which would be 3.66GHz. If I can get it to run just as cool and stable at 3.8GHz though, that would be icing on the cake.

Post by **Paradigm Shifter** » 31 May 2009, 14:26

Yeah, as I said, 3.8GHz runs fine at the same volts and temps as 3.6GHz for me. 3.2GHz runs with slightly lower voltage and slightly lower temps, but they're not enough of a drop to make it worthwhile over 3.8GHz. Particularly with stronger fans. :)

It's entirely feasible that the 'Auto' setting for Vcore on this board is giving the chip more than it needs, and that it would, in fact, be quite happy doing stock speeds at significantly lower voltages - I seem to remember one chap 'extreme undervolting' and having his 2.66GHz i7 920 running stock at less than 1.1v... both idle and load temps took a serious dive with voltage that low.

I know that 'Auto' on Vcore with this board likes serious overvolting when you turn BCLK up - leaving everything on Auto bar BCLK (with that set to 150 from a stock 133) means the board pumps 1.34v into the chip. Even though the chip isn't running that fast... it sure runs hot with that much juice...

Frag Maniac · Post by **Frag Maniac** » 31 May 2009, 18:50

Well you're using two fans on your TRUE, I may go with the Titan Fenrir, which just has one. So I don't know if my results will be the same.

I could go with a setup like you have and use two 4 pin fans attached to the MB though. I basically want it quiet at idle for movie watching but I think if you setup the speed control via the MB it's not auto regulated like the Fenrir is, you have to manually adjust it in the BIOS, which would be a pain every time I start a game.

What MB are you using btw?

Post by **Paradigm Shifter** » 01 Jun 2009, 13:14

Gigabyte EX58-UD5. It was a bundle deal with a D0 stepping i7 920.

I'm using a Zalman MFC-1 Plus to control six fans. On minimum they're close to inaudible (and temps, while high, aren't too bad...) while on high they're fairly loud but it's the 'whoosh of air' noise you get from high-power fans. Less annoying than the rattle my pedestal fan gives when it's cooling the room. (No AC, unfortunately, as the room where this i7 is has a nasty tendency to get up to 32*C or so in the afternoons, then not lose that heat until 3 or 4am...)

I did a test with another system yesterday - downstairs in a cooler room Q6600 @ 3.2GHz was idling at 30*C. Up to the room where the i7 is... 40*C. I reckon the 920 would lose 10-12*C off it's load temps, and probably 5-8*C off it's idle temps if I moved it downstairs.

Frag Maniac · Post by **Frag Maniac** » 01 Jun 2009, 20:41

Gigabyte EX58-UD5. It was a bundle deal with a D0 stepping i7 920.

I'm using a Zalman MFC-1 Plus to control six fans.

I'm looking at the Gigabyte MBs too. I actually like the slightly older X58s they have better because they offer up to 2000MHz memory standard without an OC. Most say you don't get much benefit over 1600MHz on an i7, but on most Gigabyte X58s you have to OC just to get that.

As for the fan control, I prefer a setup that is auto speed regulated. I get side tracked easily and being 51, my memory isn't what it used to be. I'd probably launch games forgetting to crank up the speed all too often, which can be hard on the CPU. I've still yet to see anyone carry the Titan Fenrir in the US though, why I don't know. Newegg carries 3 of their CPU coolers, just not that one.

Post by **Paradigm Shifter** » 02 Jun 2009, 01:40

Yeah, the Gigabyte X58 boards seem less... trouble laden than any of the others. I can't get Sleep mode to work when BCLK is more than about 10MHz over spec, though. Although that issue seems standard across all X58 boards. The Asus boards don't appear bad, but aren't as well specced as the Gigabyte boards and in the UK cost just as much.

I'm going to be building an as-small-as-possible ultra-low-power HTPC (dual core Atom with nVidia Ion mobo) soon, so need to order some more kit. Will be custom building a wooden case for it. Might stick up a build log if I can be bothered. ;) (Spending spree again... but this is a present for someone...) on the topic of the Fenrir, you got PM.

On the subject of fans, I have RealTemp in my startup folder. I leave the fans on ~9v most of the time, and only turn them up if they look to be getting warm, or down if I want it as quiet as possible. Crysis, Oblivion, Cinematic Mod Half-Life 2, GRID, DMC4... none of them get the CPU over 65*C average. Even on a day when it's idling at 50 'cause the room is 34*C - only things that really push this CPU is encoding (never breaks 72*C) or a stress test. I can't stand auto fan regulation after a board years ago that could never get it right - it was always either turning the fans off (system overheating very fast) or whacking them up to max and leaving it there even after the CPU was 20*C under the 'max' threshold.

Post by **g00seberry** » 24 Sep 2009, 12:07

I'm yoinking your thread.

I'm currently at 3.6ghz, just as a trial run. Voltages are, compared to you anyway, pretty high. Leaving them up here, then slowly reducing them.

Vcore : 1.31v BIOS, 1.280V load
QPI : 1.335v

Ratio : x20
BCLK : 180

I think temps are acceptable considering I should be able to lower the voltage. This is after 1 hour of prime.

I will put it up to 3.8 now and test for stability, if so then i'll reduce the voltages down as far as they can go. I don't think I'll try for 4ghz. People have said they've managed to do it on a UD3R, but i don't want to push it. 8)

Post by **Paradigm Shifter** » 24 Sep 2009, 12:33

First, don't trust Hardware Monitor for temps on a Core i7. Download RealTemp 3.00 and use that. Far more accurate. Also, use LinX as a stress test once you've tried Prime95. It will guaranteed get you temps 10*C+ higher than Prime, and it'll hammer the RAM at the same time.

Those temps are good for 3.6GHz at 1.3v, IMO. Provided they're correct.

Set the BCLK to 200 and the multi to 19x for 3.8GHz - for some reason the 19x multi on 200BCLK is more stable for me than 20x multi on 190BCLK. Needs more voltage to be happy on the 20x multi.

When I'm done with thesis, I'm gonna watercool this i7 to get temps down more, then I'll probably see if it'll do 4-4.2GHz without too much of a voltage boost.

Post by **g00seberry** » 24 Sep 2009, 12:48

I just tried 190BCLK x20 for 3.8ghz, and even with 1.337v on the vcore, windows would BSOD. So I'm back down to 3.6ghz now and just slowly reducing the vcore.

Not sure if I should change my QPI volts? They're on 1.31v atm. I read it should be about 0.33v less than your RAM? Dunno if that's true. My RAM is on 1.66v.

Trying Real Temp 3.00 now. The readings are almost identical to Everest. In fact Real Temp says they're 1-2C lower.

Have only run linX for 10mins, but temps are giving the same as Prime atm, max 67C on one of the cores.

I think i'mna try dropping the multi to x19 and seeing where I get, as 19x200 would be just what I want, will keep the RAM at optimum speed too.

Post by **Paradigm Shifter** » 24 Sep 2009, 13:02

I just tried 190BCLK x20 for 3.8ghz, and even with 1.337v on the vcore, windows would BSOD. So I'm back down to 3.6ghz now and just slowly reducing the vcore.

I did it 'quick 'n' dirty' - 1.35v Vcore, see where it'd boot, take it down until stable, then drop voltage to reduce the temps. But the 20x multi needs more Vcore for some reason. In my experience.

Not sure if I should change my QPI volts? They're on 1.31v atm. I read it should be about 0.33v less than your RAM? Dunno if that's true. My RAM is on 1.66v.

QPI within 0.5v of DRAM, or damage to the memory controller. I tend to call it 'within 0.4v' to be on the safe side. 1.33v QPI is good, might want to try 1.35v (safe maximum) as I couldn't get RAM stable until 1.35v with my Corsair sticks.

Trying Real Temp 3.00 now. The readings are almost identical to Everest. In fact Real Temp says they're 1-2C lower.

Ah, that was Everest? *wibble* Everest is fine.

Have only run linX for 10mins, but temps are giving the same as Prime atm, max 67C on one of the cores.

Set it to, say, 4096MB RAM, 50 runs (you don't have to let them complete until the final stress test, it'll take ages) and give it some time. Temps'll rocket.

I think i'mna try dropping the multi to x19 and seeing where I get, as 19x200 would be just what I want, will keep the RAM at optimum speed too.

Good plan.

Widescreen Gaming Forum