Would you/Should I Delid?

[flexy]

I am pretty savvy when it comes to modding, but we all know that delidding and applying CLU on a CPU has "some" risk.

Here is the situation:

4770k now at 4400Ghz stable at 1.220VID
(I spent a LOT of time to get this to 4500, but no dice)

Temps with heavy synthetic stress testing (OCCT, AVX etc...) "very hot", up to 93. Yes fricking Haswell... Means there is no way I can up VID at the moment. Temps under normal conditions fine obviously.

Now I know that delidding might well drop my temps a lot, yet I don't think I would be able to get higher than maybe 4600, even after a delid. I am on air! Would you delid a 4770k for a potential gain of say 200Mhz or is this silly, should I just be fine with my 4.4g?

 

[Elixer]

For a possible 200Mhz bump, nope, and yeah, that would be silly.

 

[Soulkeeper]

The only reason in my mind to delid is if you're really bored and can afford to buy a new cpu right away should things go wrong. I had 1 failed delid out of 4 where my shaky hands and eyes failed me.

 

[NTMBK]

No, and no.

 

[BonzaiDuck]

What air-cooler are you using?

Also, the prevailing advice is to turn off the AVX2 option when stress-testing these processors. I was skeptical at first, but there's no way that instruction-set extension would ever be used to the same extent as offered in the stress-tests.

 

[flexy]

What air-cooler are you using?

Also, the prevailing advice is to turn off the AVX2 option when stress-testing these processors. I was skeptical at first, but there's no way that instruction-set extension would ever be used to the same extent as offered in the stress-tests.

I am aware of that and TBH running some extremely far-fetched FPU/AVX test for an hour "to test the system" is about the most silliest thing to do ever. It has nothing to do with the real world whatsoever.

I am using my trusty, old Thermalright Ultra 120 Extreme. I also have a EVO 212 but this one goes on the old PC for my wife.
(Current case and airflow SUCKS, but I got a new one coming so this might well improve temps too).

 

[BonzaiDuck]

I am aware of that and TBH running some extremely far-fetched FPU/AVX test for an hour "to test the system" is about the most silliest thing to do ever. It has nothing to do with the real world whatsoever.

I am using my trusty, old Thermalright Ultra 120 Extreme. I also have a EVO 212 but this one goes on the old PC for my wife.
(Current case and airflow SUCKS, but I got a new one coming so this might well improve temps too).

The TRUE was 'good in its day." Folks are going to think I'm some sort of sales rep for EVGA because I've posted repeatedly about this, but I found a cooler similar in size or shape to the TRUE which outperforms a Noctua-D14 by 6C in bench comparisons and ~5C in my own tests. It's the EVGA ACX (formerly "SuperClock") cooler. It's about $52 at the Egg right now. You'd want to replace the fan, and you'd be wise to at least spend $7 on a ThermalRight rubber accordion duct for it -- short of making your own duct-box (as I do).

That being said, I recently acquired an i5-3570K and was thinking I might delid it. But that's a symptom of my personal "goal displacement" -- too eager to engage in modding projects just for the sake of it.

Those IB and Haswell chips DID vary in how the heatspreader and TIM were applied -- or equally what sort of temperatures they would generate. But I'd focus on better cooling and avoid any de-lidding risk.

 

[flexy]

I have seen reviews/reports where the TRUE performed 6C better than the EVO on Haswell systems under max load/temps. This is why I chose tp keep the TRUE. (It's as simple as that 6 heatpipes are certainly a little better than the 4 the EVO has).

Looking at the EVGA AXC, I know it's silly to judge a cooler from "looking at it", but it looks very close to the TRUE (as you mention already). I cannot see the EVGA cooler being significantly better/worse than the TRUE.

As for the TRUE itself, I think age shouldn't be a criterion. Just from the weight and the nr. of heatpipes it is clear that it will outperform the EVO.

 

[BonzaiDuck]

I have seen reviews/reports where the TRUE performed 6C better than the EVO on Haswell systems under max load/temps. This is why I chose tp keep the TRUE. (It's as simple as that 6 heatpipes are certainly a little better than the 4 the EVO has).

Looking at the EVGA AXC, I know it's silly to judge a cooler from "looking at it", but it looks very close to the TRUE (as you mention already). I cannot see the EVGA cooler being significantly better/worse than the TRUE.

As for the TRUE itself, I think age shouldn't be a criterion. Just from the weight and the nr. of heatpipes it is clear that it will outperform the EVO.

I just provided the revelation about the ACX for your edification.

In a rank-ordering of test results cross-validated from two or more different review comparisons, it looks like this -- approximately:

ACX
D15
D14
U14S
EVO

The EVO falls behind the D14 by about 5C. The U14S falls behind the D14 by between 1 and 2C. The other information about the remainder, I already provided.

My best guess and understanding about it: You can make the EVO match the D14 with a simple duct-box to the rear exhaust fan, together with some decent improvements in airflow with the fan choices for the latter.

How the TRUE performs in relation to the EVO, I can't say from the available data because I hadn't paid attention to the TRUE since I first used it in 2007. I'd merely guess that it's the same or better by some margin.

For the Haswells, I've been watching both AiOs and Heatpipes for a solution that will reduce the temperatures you report. If the solutions fall short, I rely on my "grains of rice" approach: lapping, diamond-paste or CLU, ducting with case pressurization and good airflow. The grains of rice could add up to 10C better cooling.

Among the "grains of rice," I've explored beefier fans which some others wouldn't consider. But I don't think I'm making acoustic compromises. You can wrap Spire foam around duct-boxes, and you can thermally-control fans from the motherboard ports so the high-CFM/high-RPM performance is limited to the high thermal range. I discovered that improving airflow for some heatpipe coolers gets me a few C in temperature reduction. While I might use a 4,200 RPM Gentle Typhoon AP-30 as a ducted and muffled exhaust fan, I would limit the speed to around 3,500 -- which goes beyond providing the maximum performance for my heatpipe cooler under extreme stress-tests.

When it comes to performance in daily usage with gaming, the most you'll ever hear with my machines is occasional bursts in "white-noise" probably with lower dBA than your room's AC vent on a warm day.

 

[jhu]

Back in my day we didn't have an integrated heat spreader on the CPU. We put the heatsink directly on the die. And if the die cracked when we put on the heatsink, well tough nuts. And we liked!

 

[schmuckley]

1.22v and hitting 93c? yikes.
Up to you if you want to delid.
I probably would in that scenario.
Yes, I would.
Keep in mind it voids your warranty.
but heh..you can always reattach the IHS with back glue

 

[DrMrLordX]

I would say, only delid after you have eliminated every other possible cooling solution

*or*

do it because you are totally OCD and want the lowest temps possible no matter what.

I delidded my 7700k recently, and it bought me about 200 mhz of headroom at the top end (so far . . . !) which is really something for a Kaveri chip. It could do more for you. Your temps are quite bad, so delidding is something you should put on the table. Up to you if you want to relid like I did, or run bare die as jhu seems to be suggesting . . .

Also, wrt HSF, I know that budget dictates many things, so that TRUE you're wanting to use is looking a lot better than an nh-d15 or h240x or what have you, just from a price perspective. The ACX is cheap, but I remain somewhat skeptical.

Regardless, it's hard to beat free-as-in-already-paid-for-it.

If I recall, the TRUE can really shine with some nasty-loud fans on there. Set up a push/pull configuration with some Deltas or something, switch to CLU as your TIM-of-choice, then look at the delid. That is, if you're really going balls-out.

 

[hawtdawg]

depends. I delided my 3770K, but it was worth it to me, because I was hitting 90C at 4.3 and now I barely crack 70C at 4.6. been running at 1.35v for almost 2 years without issue.

I wouldn't recommend using a razor, that was the most nerve racking 20 minutes of my life, but the hammer/vice method looks easy-peasy, and I would say it's worth it if you're the type of person who likes to squeeze every last drop of performance out of things.

 

[fleshconsumed]

Not for me. I do not overclock because stock 4770K/4790K is plenty fast for me, and delidded CPUs are harder to sell. I even stopped lapping my CPUs because they're also harder to sell. But it might be worth it if you're into overclocking and don't care about resale value...

 

[BonzaiDuck]

. . . The ACX is cheap, but I remain somewhat skeptical. . . .

That's fine, and I should probably post photos and screenies, but it's true.

You know me (somewhat) -- I've been poking around here for years.

It is absolutely the weirdest discovery I ever made.

The comparison review was at Hardware Secrets -- jus'a-minute . . . Here it is . . .

http://www.hardwaresecrets.com/article/EVGA-Superclock-CPU-Cooler-Review/1336/6

Usually, I try to find a couple extensive comparison reviews on these suckers, but this may have been the only one I found this time. I can't even remember how I came across the ACX -- which was the "SuperClock" renamed, but otherwise identical. I'm fairly sure somebody posted something about it over the last few years at Anandtech. I remember reading some user's expression of satisfaction with it.

Notice the performance and data in comparison to the NH-D14. The fan on the ACX spins up a little higher, but there are two fans on the stock Noctua.

Things not to like about it: First, the LED fan, which is noisy at its top end. Second, the shape or "style" of the computer makes it difficult or impossible to mount a puller fan on the opposite side: It has rounded edges and doesn't seem in anyway equipped for such a fan. Third, the mounting hardware seems "OK" but I'd seen better.

What I did:

Both the NH_D14 in a 2600K system @ 4.6 Ghz, ~ 135W package power had my duct-mod with custom-chosen fans -- a GT AP-30 (ducted) exhaust, and an Akasa Viper 140R center-fan for the D14. The other system was a 2700K @ 4.7 Ghz ~140W package power with a ducted EVGA ACX cooler, same AP-30 exhaust with a ThermalRight rubber-accordion duct and a Noctua iPPC 120mm 3,000 RPM fan.

The fans in this test wouldn't make much of a difference if any of the fans were "different." The Akasa and Noctua iPPC fans were both spec'd with a rating around 110 CFM, I think. 10 iterations of Intel Burn Test "Maximum" on both systems showed a difference in average-maximum-of-cores that I calculated at 4.75C.

Both coolers used IC Diamond; both processors were lapped to bare copper.

It's just . . . that simple. Certainly it would have been better to use the same test-bed -- not two "equivalent" (or roughly equivalent) systems. But whether I used HW MOnitor or ASUS' AI Suite monitor, it didn't matter.

It's as clear as day: The ACX (aka "SuperClock") wins hands down, at least -- by that margin. Note that the package power load reading on the 2700K would even make for a likelihood of a smaller delta, and indeed my test falls about 1.25C short of the HW Secrets result.

The ACX is also less noisy or requires less attention to acoustic muffling. I account for this because the fins are thicker and less "tinny" than those of a Noctua cooler -- they're more rigid. So they don't transmit or amplify the fan noise as much.

Everything about the SuperClock/ACX would cause you to bet on the D14. The ACX is smaller with only a single tower. It seems designed to capture "bling" aspirations of impulse-buyers -- the cherry on the cake being the cheap red LED fan.

But like I said. Same computer cases (HAF 922's). Same intake fans (2x 200mm Bitfenix Spectre Pro's). Same exhaust fans. Same attention to blocking off unused vents. More or less equivalent CPU pusher fans. Same package power. Same load voltage -- less than a 10mV difference if there was any at all. And the clock setting was higher for the 2700K, as was the Package power -- with voltages identical.

The Hardware Secrets ranking is correct. Yet the EVGA cooler only costs $52+.

Go figure. Single tower beats double tower by at least 5C for (about) 140W of thermal power dissipation.