[Resolved] 2600K Overclock Take 2

[mrjoltcola]

So my first 2600K build resulted in 4.4Ghz max stable. Mildly disappointed given that Newegg sells a 4.6Ghz bare bones kit, 24-hour tested. I'll either end up selling the slower of the 2 rigs, or give one to one of my developers so it isn't a loss, but the heck if I'll surrender this easy!

I'm about to try again, with a second 2600K and an ASUS V-PRO Z68 M/B this time around, with addition of an H80 cooler instead of the H60.

I'm thinking of getting the 1866Mhz Vengeance kit this time, even though it is more expensive than the 1600Mhz. Should I bother? Will it affect my OC?

Any arguments about whether I should opt for a specific M/B, or why I should go for the Corsair H100 cooler over the H80? (I saw a graph that showed the H80 seemed to be the sweetspot for cooling, a full 10 degrees cooler than the H60, at load, but only 2 degrees off the H100 with dual fans -- not sure what the H100 would do with 4 fans, but I don't want 4 fans...). Any stats out there that show specific motherboards having better OC results? Besides the ASUS P8Z68 V-PRO, anyone have feedback on EVGA's Z68 models?

UPDATE: By using the Auto-overclock feature, I was able to get well into 5.0Ghz. Heat is now the limiting factor above 4.6Ghz, but no instability for 10 minutes at 4.8Ghz, so I'm satisfied there is nothing below-average about my chip. Ordered the Corsair H100 cooler to replace the H60.

 

[aphelion02]

If you care all that much....wait for SB-E? I basically think what you are trying to do is stupid, especially the part about the RAM.

 

[BonzaiDuck]

If you're going to do water-cooling, why pick a cooler that might actually underperform my NH-D14? Why not buy a Swiftech or other kit? You may spend more, but you'll squeeze maybe 10C degrees lower at load temperature and a room-ambient range.

I can't guarantee that your cooler "underperforms," anymore than it's a tad more on the
favorable side of mine. But I thought I saw some benchies with the Corsair [isn't it?].

Point of it being, I get to 4.74 with my Noctua, and TCASE doesn't exceed the spec. On a hot day -- maybe a couple C degrees and that's it . . .

Of course, maybe you actually got a CPU in a different performance quartile of the the production distribution. But my thoughts could still be relevant to your choices. If it's important, stick to your budget. I go all-out on air-cooling. But if I switched to water, I'd get a kit that performs significantly better than that one, even though there's still a lot to be said for it . . .

 

[Idontcare]

With BD and SB-E right around the corner, heck even IB isn't that much farther out, you would be so much better served by just holding out and getting something else that is equally fun, if not more, than doing a SB part duex IMHO.

 

[mrjoltcola]

If you care all that much....wait for SB-E? I basically think what you are trying to do is stupid, especially the part about the RAM.

You think being an enthusiast is stupid? Thanks for nothing, please go troll for a fight somewhere else. Anyway...

There is nothing _stupid_ about what I'm doing, except, perhaps, thinking I'd have a thread without at least one such comment like yours.

What I'm doing = Building a Sandy Bridge rig, having fun, and trying to get bragging rights. If you aren't into that, stay out of the thread?

 

[mrjoltcola]

With BD and SB-E right around the corner, heck even IB isn't that much farther out, you would be so much better served by just holding out and getting something else that is equally fun, if not more, than doing a SB part duex IMHO.

Thanks. Well I had actually expected to skip SB-E or IB altogether, given the timing. I can't keep waiting. I'm still running a small fleet of Conroe X6800. I have an engineering firm so I outfit my guys with identical models. I used to buy from Dell, but decided to build in house this time, as I've begun doing with servers as well. Our old model is Dell Precision 390 with X6800 Core 2 Extremes. I skipped the 2nd gen Core 2, and the 1st gen Nehalem, so here we are.

So my plan was to build a premium SB box, if it turned out well, I'd build a few identical for my guys. It is fun, it is one reason my guys like to work for me, and none of us think it is _stupid_ or frivolous. I spend more on software in a year than hardware, trust me. We are hackers and engineers and enjoy it. I ponder why I feel, all of the sudden, the need to defend myself in the "CPUs and Overclocking" forum. Thanks guys!


>>"If you care all that much....wait for SB-E? "

EDIT: Extreme Edition models run a lot more, if history is an indicator SB-E will probably run $600 more than a 2600K. The 2600K is certainly a better value.

 

[Idontcare]

Ah, now that I understand the context of your situation. More 2600K's are in order stat

 

[TakeNoPrisoners]

If you want cooling, you need fans. That is true with water and air cooling as far as I know.

I was only able to get to 3.9GHz with my PII with one fan on my Hyper 212+. When I put another fan on I can now hit 4.1 GHz.

I think you should take a look at your case and make sure you have the fan capacity maxed out, with PWM controls or a fan controller so you can adjust for noise.

If you really want that super overclock just buy a top-of-the-line air cooler such as the D14 or the H100 with all four fans.

You just cannot get extreme overclocks with complete silence.

Those large 120mm and 140mm fans really arn't very loud. You can hear them move large volumes of air, but they don't need to spin up fast to get that job done which makes them quiet.

 

[mrjoltcola]

I have the Corsair 600T Graphite Special Edition (white) case, with 3 fans, including the Corsair fan with the H60. I have the glass window on the side panel, though it also came with a screen. It isn't exactly silent, but is decent. I hear you on fans, except I had hoped water would get me to where I wanted be. So far my box gets unstable at 4.6Ghz @ 40 degrees C, so I don't think temperature is the limiting factor. It can't be. The CPU runs hotter under load at 4.4Ghz than it does when it blue screens at 4.6 idle.

>>Ah, now that I understand the context of your situation. More 2600K's are in order stat



So I'll report back on my findings with the 2nd build. Not doing a complete rebuild, just M/B + proc. Thinking of just keeping my current RAM, aphelion02's comment sort of sobered me up. I was just wondering if a RAM change would be in order, and am wondering what you think (aphelion02) is stupid about what I said about the RAM? Did you mean the mere act of overclocking, or did you mean considering buying new RAM? Budget isn't so much of an issue, I have so much hardware at this point, and I'm selling some of my Dell workstations and servers, so there is some $$ coming in for replacement.

 

[VirtualLarry]

Why are you overclocking engineering workstations? I sincerely hope you aren't building bridges, or skyscrapers, or space shuttles. You get my drift.

 

[mrjoltcola]

Why are you overclocking engineering workstations? I sincerely hope you aren't building bridges, or skyscrapers, or space shuttles. You get my drift.

Sigh. I keep teeing 'em up and you guys keep taking cracks at me.

None of my workstations at the office are overclocked. I have my box at home that IS though I run a mild profile most of the time. I stress test it, then back down several multipliers.

I'm not talking about overclocking a fleet, I'm talking about taking the best processor from a batch of builds, for my home fun box.

 

[grkM3]

Your cpu is not holding back your overclock.running 16gb of ram is.try running 4gb 2x2 and see how much higher your chip will clock.

 

[Idontcare]

Why are you overclocking engineering workstations? I sincerely hope you aren't building bridges, or skyscrapers, or space shuttles. You get my drift.

Don't tell Via, they design their chips using farms of overclocked Intel cpu's.

The systems that were in cases were water cooled Core i7s, overclocked to 5GHz. There are two folks at Centaur who build each and every one of these machines, and overclock them. You and I know that overclocking both Nehalem and Sandy Bridge results in much better performance for the same dollar amount, but this is the first time I've seen overclocking used to speed up the simulation of microprocessors.

Just because it can be done badly and poorly (OC'ing) doesn't mean it can't be done competently and correctly.

It just means you have to know a thing or two about what you are doing, including knowing how to verify you've done it right, and then there's no magic or mystery about it.

 

[mrjoltcola]

+1 Idontcare!

I know of firms running Alienware "everything", laptops and workstations, all nicely overclocked.

 

[mrjoltcola]

Can you believe this? After all my whining... I must have the worst OC skills in the forum. Everything I'd read told me to set things manually, like Vcore, etc. and that is what I've been doing.

So just on a whim, I decided to try the ASRock Auto-Overclock and went to 4.6Ghz stable for 15 minutes under load! Prime95 Large 8-thread for 15-minutes, until temp equalized to 82C, then followed by several minutes of SuperPi. No errors with SuperPi, no blue screens and no instability.

It gets better, I went back into BIOS, chose the auto 4.8Ghz setting, rebooted, and am running Prime95 right now while typing this! Now I've no idea if I'll make it through 24-hour torture test, but this is so stark of a difference! I could not even boot without BSOD at 4.7, and Prime95 would crash the box at 4.5Ghz.

Unbelieveable. The Vcore is only 1.335v at load. All that setting to 1.4v wasn't on the right track. So my chip isn't a lemon after all?? Now I'm puzzled what setting was the key here.

 

[mrjoltcola]

Now it is really apparent that I need to upgrade the Corsair H60 cooler.

@4.8Ghz, after 10 minutes of Prime95, CPU Temp hit 90C. I stopped the run once it hit 90, but still no crashes. Now I'm thinking about the H100. Two hour ago, I was thinking about $500 worth of new CPU and M/B!

I do note that SuperPi performance seems to level off after 4.4Ghz, the numbers at 4.8Ghz aren't significantly better (11.3 seconds for 4 threads 1M run). I note more of a performance gain going from 1333Mhz DRAM speed to 1600Mhz with SuperPi, so perhaps SuperPi is bottlenecking on memory transfers now. Maybe see how 1866Mhz DRAM fares at the new CPU speeds.

CORRECTION: SuperPI wasn't bottlenecking, I screwed up and compared an 8-thread run with 4-thread runs from last week. HyperPi reset the threadcount on me.

 

[StrangerGuy]

You think being an enthusiast is stupid? Thanks for nothing, please go troll for a fight somewhere else. Anyway...

There is nothing _stupid_ about what I'm doing, except, perhaps, thinking I'd have a thread without at least one such comment like yours.

What I'm doing = Building a Sandy Bridge rig, having fun, and trying to get bragging rights. If you aren't into that, stay out of the thread?

If you think people here will actually be impressed by your antics, think again.

 

[aphelion02]

You think being an enthusiast is stupid? Thanks for nothing, please go troll for a fight somewhere else. Anyway...

There is nothing _stupid_ about what I'm doing, except, perhaps, thinking I'd have a thread without at least one such comment like yours.

What I'm doing = Building a Sandy Bridge rig, having fun, and trying to get bragging rights. If you aren't into that, stay out of the thread?

First of all, I didn't see the part about you giving your 2600k to your developers when I first read your post, so it seemed like you were willing to just take a huge loss on selling a rig for a few hundred Mhz of bragging rights, especially considering your comment about overclocking RAM which is next to useless on these Sandy Bridges, even in very intensive applications. If this was in your original post and not part of the edit, I apologize for my reading comprehension fail.

Secondly, you specifically asked for people's thoughts, and you got my honest opinion. If you are so thin skinned you can't take any dissenting opinions, you have no business posting on any internet forums. I don't need any snobbism about being an engineer, seeing as I'm a damned good one myself.

 

[Idontcare]

Now it is really apparent that I need to upgrade the Corsair H60 cooler.

@4.8Ghz, after 10 minutes of Prime95, CPU Temp hit 90C. I stopped the run once it hit 90, but still no crashes. Now I'm thinking about the H100. Two hour ago, I was thinking about $500 worth of new CPU and M/B!

I do note that SuperPi performance seems to level off after 4.4Ghz, the numbers at 4.8Ghz aren't significantly better (11.3 seconds for 4 threads 1M run). I note more of a performance gain going from 1333Mhz DRAM speed to 1600Mhz with SuperPi, so perhaps SuperPi is bottlenecking on memory transfers now. Maybe see how 1866Mhz DRAM fares at the new CPU speeds.

I apologize if I've already asked you this, but have you lapped the H60 or would you consider it? The surface of my H100 turned out to be rather concave leading to a large pocket of TIM residing between the IHS and the water-block. I lapped it (and the CPU) and got better results (~5C drop in temps).

As for the magic of what the auto-OC is doing, yes definitely pursue that. You, we, need to know what the crucial difference is.

On my MIVEZ I found out that the enable "PLL Overclock" removed the same barrier you appear to be experiencing with your chip. But you've already enabled that IIRC, right?

 

[john3850]

My superpi numbers just keep going up as they should as I increase my multiplier.
If you pi time stops going up there is a reason.

 

[Idontcare]

I do note that SuperPi performance seems to level off after 4.4Ghz, the numbers at 4.8Ghz aren't significantly better (11.3 seconds for 4 threads 1M run). I note more of a performance gain going from 1333Mhz DRAM speed to 1600Mhz with SuperPi, so perhaps SuperPi is bottlenecking on memory transfers now. Maybe see how 1866Mhz DRAM fares at the new CPU speeds.

My superpi numbers just keep going up as they should as I increase my multiplier.
If you pi time stops going up there is a reason.

I missed this part of mrjoltcola's post.

Sandy Bridge contains ECC on some memory paths and circuits, it is far more robust in detecting errors and correcting for them rather than just crashing and giving you a BSOD.

The downside to this is that you can optimize your overclocks to such an extreme where the chip is "stable" thanks to ECC but your performance is not increasing because of the delays involved in resending data once errors have been detected on the fly.

The only way to ensure you aren't operating in this ECC-enabled-stability regime is to increase the Vcc on what already appears to be a stable OC and verify that the bench numbers are not improving as you increase Vcc.

At 4.5 GHz, running 4 threads on my 2600K, here's my GFlops versus Vcc (measured with multi-meter):

Vcc........GFlops
1.190V....N/A (system BSOD's during LinX)
1.293V....N/A (LinX stops on error detected)
1.325V....95 GFlops
1.329V....102 GFlops
1.334V....108.2 GFlops
1.339V....116.6 GFlops
​

Voltages above 1.339Vcc at 4.5GHz yield identical GFlops.

Note I am "100% stable" at 1.325V, I can run IBT for 12 hrs and not crash or BSOD, I also don't crash with any other apps at that voltage.

But I'm clearly forcing the chip to operate in an unoptimal environment, the ECC is kicking in to take care of the internal inconsistencies at the expense of requiring the computations take longer to be completed.

 

[AdamK47]

Are there any articles that go over ECC when overclocking? I'd like to know what I'm getting into once (if ever) I get an SB-E. Detecting and preventing it will be one of my top priorities during the overclocking tests.

 

[mrjoltcola]

If you think people here will actually be impressed by your antics, think again.

What antics? I'm not the one insulting people here.

I just don't take to being called stupid by someone I never met. Insulting someone over a PC build is juvenile any way you cut it. You can get your point across without that.

If I were trying to impress you it wouldn't be by telling you about my lack of ability to overclock my build. Plus I'm sure there is very little I could do to impress you. I am here because I _was_ having a lot of fun at this, and like to carry on friendly discourse about it. But for some reason I have a knack for bringing out the best in people online... I don't understand why people just can't be friendly, and leave it at that.

 

[mrjoltcola]

My superpi numbers just keep going up as they should as I increase my multiplier.
If you pi time stops going up there is a reason.

I made an error there. I was comparing numbers from last week when I had run 4 & 8 threads. It does indeed continue to drop as I retested with the same settings in one sitting.

SuperPi on single thread indeed continues to drop the further I clock up. At 5.0Ghz now I'm getting approx 7.3 seconds. Does that sound about right?

 

[Idontcare]

Are there any articles that go over ECC when overclocking? I'd like to know what I'm getting into once (if ever) I get an SB-E. Detecting and preventing it will be one of my top priorities during the overclocking tests.

Other than references to its existence, and the reason for its existence (which are peppered through many of Anand's reviews), there are no "guides" that I know of.

And it's not something that can be disabled like HT or the cache itself, as I understand it.

But the simplest way to ensure you aren't in that regime is to up your Vcc one or two notches at the same clockspeed and just rerun the bench. If the number doesn't materially change then you know the lower Vcc was both stable and providing sufficient voltage to avoid performance-limiting ECC-induced resends.

(also it may well be mobo dependent as well, as others have noted that can't seem to get their SB to behave the same way but they are using a different mobo...when you are operating at the hairy edge of stability the power-control quality may be making the difference between getting a BSOD or reboot versus remaining stable but having lowered performance)