Haswell Overclock Thread

shadow_boxer · 13 Jun 2013 at 10:58

lettuce said:
Damn really!, now I'm happy with 4.7 @ 2.585v

andybird123 · 13 Jun 2013 at 11:07

Sal0hcine said:
Hmm Mine is running 4.5 @ 1.25 volts but seems to be hitting 77c. It idles around 30c.

I am running my first custom water cooling loop so was expecting better temps. I wonder if I need to re-seat my CPU cooler as my GTX 470s are running at 30c idle and 45c load which is a lot more acceptable.

I did the Asus test on my Gigabyte OC board, booted into Windows but crashed on Prime. So 4.6 might be possible.

this is haswell... using TIM under the heat spreader, there is only so much that cooling can do - to get good temps you need to delid the CPU and cool the chip directly

many people are reporting that you get no real benefit from a £30 air cooler through to £££ watercooling, as you add volts to haswell the temps go up... 77C is actually a pretty decent temp, some people are hitting high 80's / low 90's at 4.5ghz

Shiari · 13 Jun 2013 at 11:16

andybird123 said:
this is haswell... using TIM under the heat spreader, there is only so much that cooling can do - to get good temps you need to delid the CPU and cool the chip directly

It's not so much as the TIM as it is the glue between the circuit board and the heatspreader which creates a (small) gap between the silicon and the heatspreader, which makes it much harder for the TIM to pass the heat on. Other than that the small size of the chip isn't helping, with a lot of heat focused in a very small area.

Delidding should certainly help, if only to get rid of that glue and make the gap smaller. Not going to do that just yet though.

Steampunk · 13 Jun 2013 at 11:42

I think I've been pretty lucky with my 4770K. I didn't de-lid because I can't afford to trash a CPU this expensive. I was lazy and just allowed the Asus utilities find me a rock solid overclock, and the CPU is now doing 1-2 cores at 4.8, and 3-4 cores at 4.7 @ 1.264 volts.

Doing MP4 rendering is only maxing CPU temps to 83 degrees with all four cores working. I'm pleasantly surprised to get a 30 percent overclock for very little effort. I'm not greedy enough to try and squeeze any more at the expense of stability.

At those overclock settings Intelburntest/Linpack is a bit scary though. It the space of a few seconds it's banging on the 100 C limit. At least I know the thermal throttling is working.

I'm actually running on air with a Noctua NH-U14S, and I'm wondering if something that big with it's own backplate actually squeezes the CPU slightly and makes especially good contact by squashing the glue and putting pressure on the centre of the heat spreader.

Shiari · 13 Jun 2013 at 11:49

Steampunk said:
I'm actually running on air with a Noctua NH-U14S, and I'm wondering if something that big with it's own backplate actually squeezes the CPU slightly and makes especially good contact by squashing the glue and putting pressure on the centre of the heat spreader.

Nah ... my Noctua NH-D14 is huge too, but so is the plate pushing on the heatspreader (and it's much larger than the heatspreader). The pressure on the heatspreader will be taken by the vertical edges of the heatspreader.

lettuce · 13 Jun 2013 at 12:54

shadow_boxer said:

Yeah stupid iPad virtual keyboard

What i ment was, 1.255v.

To be fair these boards seem to be a lot easier to get an overclock from than my previous x58 mobo. The only settings I need to alter was....

CPU Vcore
CPU Base Clock
CPU Clock Ratio
Uncore Freq
Turbo ratio clock
System Memory Multiplier
Memory Channel Timings
Memory voltage
Energy saving states

Everything else can be left as Auto

Zeed · 13 Jun 2013 at 13:02

i would love haswell just to have a play with overclocking it

I must have invested well over 50 hours in tweeking my sandybridge and i still sit and play with it every second day.

Since i got my Samsung greens i play BIOS more than Games

dim_pwn · 13 Jun 2013 at 13:04

*crosses fingers for a good chip*

Weazle · 13 Jun 2013 at 13:19

Silly question, what kind of UNCORE settigns should I be settign in relation to the normal CPU settings?

For example: Basic overclocking is to ramp up he multiplier say 40 for i5-4670K, should the UNCORE multiplier be 34 (stock), 39 or 40?

Duff-Man · 13 Jun 2013 at 13:22

Weazle said:
Silly question, what kind of UNCORE settigns should I be settign in relation to the normal CPU settings?

For example: Basic overclocking is to ramp up he multiplier say 40 for i5-4670K, should the UNCORE multiplier be 34 (stock), 39 or 40?

What what I've read, you ideally want Uncore to be one multiplier below the CPU multiplier; so for your 4Ghz clock you would want uncore multiplier at x39.

That being said, in my testing I've noticed little to no difference between the two settings.

Shiari · 13 Jun 2013 at 13:40

Weazle said:
Silly question, what kind of UNCORE settigns should I be settign in relation to the normal CPU settings?

For example: Basic overclocking is to ramp up he multiplier say 40 for i5-4670K, should the UNCORE multiplier be 34 (stock), 39 or 40?

I've read all kinds of advice. Some say two lower. Some say 40. I can't give you a definitive answer.

It does seem that lowering it a bit can help getting the overclock stable.

SkyTerran · 13 Jun 2013 at 21:55

The four bin overclocking for non-K i5 and i7 is gone: http://techreport.com/news/24950/intel-removes-modest-free-overclocking-from-standard-haswell-cpus :rolleyes:

Yesterday I had a chat with a big corp sys admin. They have acentralized monitoring application and the CPU temperatures of their Ivy Bridge PCs are increasing by half a degree or so every month. No such problem on their Sandy Bridge boxes. Makes me wonder if Intel's plan with the TIM isn't to have people replacing their computers every 3-4 years because of the overheat throttling.

8 Pack · 13 Jun 2013 at 22:24

For 4.2-4.4ghz oc 2 multi down for uncore is fine 4.6-5.0 3-4 fine. Its diminishing returns. At 6.7ghz, uncore of 5.8 is enough for max bench scores.

tipes · 13 Jun 2013 at 22:29

8 Pack said:
For 4.2-4.4ghz oc 2 multi down for uncore is fine 4.6-5.0 3-4 fine. Its diminishing returns. At 6.7ghz, uncore of 5.8 is enough for max bench scores.

6.7Ghz indeed - you showoff you. Lol

Mark

Jokester · 13 Jun 2013 at 22:31

I need more willpower, looks like 4.4GHz is as good as mine will do. Back to my IB I think.

lettuce · 14 Jun 2013 at 01:00

8 Pack said:
For 4.2-4.4ghz oc 2 multi down for uncore is fine 4.6-5.0 3-4 fine. Its diminishing returns. At 6.7ghz, uncore of 5.8 is enough for max bench scores.

That will make the system more stable right, as at the moment ive only been setting uncore 1 down from my clock??...might be the reason why i cant get 4.8Ghz to be stable at stress, ill try 4.5-4.6 for 4.8 clock speed then

Zeed · 14 Jun 2013 at 07:57

Jokester that was reason why i have not ordered one with maximus VI gene ... What if it cant pass 4.6 ?? Its useless for me as i dont care about power consumption at all

And it would be bit **** to buy one and send it back 3 days later cause i am not happy with speeds im getting :/

smilertoo · 14 Jun 2013 at 08:01

I seem to be getting software startup errors on the MSI x87 g65 gaming; the 4.3oc passes all stress tests and runs rendering apps...but the killer lan software crashes on startup and causes a runaway cpu thread until shut down in task manager.

illuz · 14 Jun 2013 at 09:18

Quite disappointing all these Haswell clocks, even the Engineers AMA on Reddit said it would be an overclocking beast. As for the TIM, delid. If you have a good chip - it's worth it. 3770k @ 4.8 v1.4 here under water max temps 65c direct-die cooling.

game4it · 14 Jun 2013 at 09:22

smilertoo said:
I seem to be getting software startup errors on the MSI x87 g65 gaming; the 4.3oc passes all stress tests and runs rendering apps...but the killer lan software crashes on startup and causes a runaway cpu thread until shut down in task manager.

I've the same board and sometimes when i boot up the lan goes flakey.It takes about 5 mins for it to open a web page.Not a fan of it tbh.