i5-3570k overclock temps

[atramel]

Yeah, I'm not sure what to make of the overclocking situation with this processor. Seems like it really comes down to luck of the draw. I just setup a new gaming system about two weeks ago, and went with the 3570k for this build.
I did a lot of reading both before and after, and it was a little disappointing to see that people weren't getting the kind of overclock that you could with the Sandy Bridge parts.
Somehow I got extremely lucky with this processor - first time that's ever happened to me! I was worried I'd be stuck at near stock clocks, but this one is prime stable at 4.7GHz and 1.264v just using a Hyper 212 cooler with peak temps in the 81-82 degree range.
It may even make it to 4.8-4.9, which would be pretty darn close to what I was able to achieve with Sandy Bridge, though my Sandy Bridge system is cooled with a custom water loop, so I was able to push it a little harder than this.

 

[BrksEverything]

My 3570k was 4.6 stable at ~1.248 (offset of ~.055), and temps maxed at 82-83 under IBT load. (arctic freezer i30 air cooler, MX-4 paste).

I wanted to go up to 4.8 but knew I would quickly hit a thermal ceiling. So, I delidded, used CLU under the IHS, and NT-H1 between the IHS and heatsink. IBT load temps dropped ~19 degrees(@4.6), and max out around 72 degrees while clocked up to 4.8.

Where I'm going with all this, is that what really amazed me was even with all the thermal head-room delidding provided, the jump in voltage needed to stabilize 4.8 from 4.6 was unreal.

The lowest offset voltage I can get 4.6 stable is .055. 4.8 took .155, corresponding to a load Vcore of 1.365! I know the general wisdom is this voltage, and temps in the mid 70s isn't too dangerous, but my point is the exponential increase in voltage needed, even when temps aren't a factor indicates to my uneducated, non-engineering mind that this chip just isn't made to do what we want.

To use a horrible, horrible analogy it's like getting mad at a VW Bug for not hitting 140 mph when injecting NOS into it. Even if you can cool the engine enough, and keep it from shooting a piston through the hood, it just wasn't built for it.

 

[Yuriman]

Would surprise me, I thought Mini's were actually pretty quick. =P

I too find that above 4.6ghz the voltage requirements increase very rapidly for my chip.

 

[BrksEverything]

Would surprise me, I thought Mini's were actually pretty quick. =P

I think you're right - fixed.

 

[dave_the_nerd]

So how much does it hurt the chip (if any) to be running it at 85C (OC'd) vs. 65C (stock) (load temps) Rule of thumb back in the day (Athlons, Core 2s) was to keep it under, like 60C if you could.

If that TIM is made from asbestos or leftover space shuttle tiles, that'd actually explain some things.

 

[BrksEverything]

So how much does it hurt the chip (if any) to be running it at 85C (OC'd) vs. 65C (stock) (load temps) Rule of thumb back in the day (Athlons, Core 2s) was to keep it under, like 60C if you could.

If that TIM is made from asbestos or leftover space shuttle tiles, that'd actually explain some things.

I think 'idontcare' demonstrated that it's not that Intel used a sub-par thermal paste, per se, but rather that they used a thermal paste at all rather than solder, combined with the increased gap between the silicon die and IHS that led to higher temps. That's why delidding and using liquid gallium based TIM seems to provide the greatest reduction in temps; the gap is reduced, and the liquid metal TIM approximates the heat transfer ability of soldering.

I honestly have no clue about the rest.

 

[Idontcare]

The lowest offset voltage I can get 4.6 stable is .055. 4.8 took .155, corresponding to a load Vcore of 1.365! I know the general wisdom is this voltage, and temps in the mid 70s isn't too dangerous, but my point is the exponential increase in voltage needed, even when temps aren't a factor indicates to my uneducated, non-engineering mind that this chip just isn't made to do what we want.

Its more an issue of the process tech than chip itself per se. The 22nm finfets and metal interconnect just aren't designed to enable the kinds of transistor switching speeds that are necessary to push the clockspeeds any better than those of 32nm HKMG.

Sure you can get brutal in forcing the issue, but with drastically diminishing returns as you noted.

To use a horrible, horrible analogy it's like getting mad at a VW Bug for not hitting 140 mph when injecting NOS into it. Even if you can cool the engine enough, and keep it from shooting a piston through the hood, it just wasn't built for it.

You can also think of it from a transmission-limited standpoint or even a drag-limited perspective.

If you are transmission limited then putting in a bigger engine isn't going to get any higher top-speed because the transmission can't actually get the extra torque to the wheels, just burns out the clutch-plate on the existing tranny all the sooner.

Alternatively if you are drag limited, putting in a beefier engine will incrementally boost your top speed but the already excessive drag is only going to become even more excessive, resulting it silly levels of diminishing returns for your higher horses and torque. Formula 1 race cars are not pickup trucks bodies for a reason

 

[Idontcare]

So how much does it hurt the chip (if any) to be running it at 85C (OC'd) vs. 65C (stock) (load temps) Rule of thumb back in the day (Athlons, Core 2s) was to keep it under, like 60C if you could.

If that TIM is made from asbestos or leftover space shuttle tiles, that'd actually explain some things.

Best I could determine the CPU TIM has a comparable (if not slightly better) thermal conductivity as NT-H1.

The problem is how thick the CPU TIM is. Take the best TIM in the world but make it be an inch thick and you've created yourself a thermal barrier.

Temps are still a concern, particularly when you are dealing with both an overclocked and over-volted processor.

The TJmax value is only a "safe" max operating temperature provided you are operating the CPU at stock clocks and at stock voltages.

The gray zone is when you overclock but under-bolt, the tradeoffs there can result in basically no compromise in operating lifespan for any given operating temperature in comparison to a stock configuration.

 

[dave_the_nerd]

I think 'idontcare' demonstrated that it's not that Intel used a sub-par thermal paste, per se, but rather that they used a thermal paste at all rather than solder, combined with the increased gap between the silicon die and IHS that led to higher temps. That's why delidding and using liquid gallium based TIM seems to provide the greatest reduction in temps; the gap is reduced, and the liquid metal TIM approximates the heat transfer ability of soldering.

I honestly have no clue about the rest.

Sorry, I was trying to make a joke about my CPU catching on fire.

Because even stock, it runs much hotter than the old ones did.

 

[coffeejunkee]

Because even stock, it runs much hotter than the old ones did.

Not really:

i5 750, 20 runs Linx: 54-56c (lowest/highest core)
3570K, 20 runs Linx: 48-53c

Not completely comparable because of (slightly) better cooler used on 3570K but also 8GB vs 4GB with the i5 750 (more ram means higher temp in Linx).

 

[fourfist]

i can only keep my i5 3750 stable at 1.38V with pll set to 1.88 and vtt at 1.14
(for an OC of 4.4)

temps are between 80-88.

should i try anything else

 

[fourfist]

i can only keep my i5 3750 stable at 1.38V with pll set to 1.88 and vtt at 1.14
(for an OC of 4.4)

temps are between 80-88.

should i try anything else[/Q
I had my intake fan facing the wrong way, by flipping it reduced my temp 10 degrees