Intel Haswell 4670k + 4770K Overclocking guide

[AngryGoldfish]

I recognize the Uncore now. It's showing the value in watts. Any way to change it to Volts?

 

Oh sorry, I misread your comment. I thought you were talking about the Uncore frequency. I've never needed to keep an eye on Uncore voltages so I haven't looked where to find it outside of the BIOS.

[BigDay]

So far everything is going well. I'm chasing after a 4.4GHz overclock first and then I'll go after higher clocks. At the moment, I've got a 4.4GHz core clock with 1.212 volts on the core. Started @ 1.25V.

 

I'm in the process of trying to find the lowest core voltage for the core clock. I've been stress testing using IETU, I use an 8 hour test.

[BigDay]

When stress testing my CPU, should I disable "Turbo Boost" in the BIOS? If so, why?

[arniks8]

When stress testing my CPU, should I disable "Turbo Boost" in the BIOS? If so, why?

Turbo boost allows your CPU to operate on higher frequency than stock. Basically you can't OC if turbo boost is disabled.

[BigDay]

Is it OKAY to have my VRING (Cache Voltage) higher than my VCORE? (Core Voltage)?

[AngryGoldfish]

Is it OKAY to have my VRING (Cache Voltage) higher than my VCORE? (Core Voltage)?

 

Yes, it is, but it's not worth it. The added heat and potential for instability later on due to voltage leakage over time is not worth the performance gain. Core is king. Getting the cache ratio up is great for an all-round high performing system, but it's not worth piling on voltage to get it 1:1. If you want to do it for fun, to squeeze that little bit extra performance out for a few benchmarks, go for it. It won't harm the CPU (AFAIK). But I'd dial it back for 24/7 personally since the performance gain is not worth the added heat and potential degradation.

[BigDay]

Yes, it is, but it's not worth it. The added heat and potential for instability later on due to voltage leakage over time is not worth the performance gain. Core is king. Getting the cache ratio up is great for an all-round high performing system, but it's not worth piling on voltage to get it 1:1. If you want to do it for fun, to squeeze that little bit extra performance out for a few benchmarks, go for it. It won't harm the CPU (AFAIK). But I'd dial it back for 24/7 personally since the performance gain is not worth the added heat and potential degradation.

 

Thanks for the input bro.

 

Here's where I'm at now. I've found stability with the following settings:

 

4.4GHz Core Frequency @ 1.20V (Core Voltage)

3.6GHz Uncore Frequency @ 1.20V (Cache Voltage)

VRIN @ 1.90V

 

LLC = Extreme

HTT = Enabled

Intel Turbo Boost = Auto

 

C1E = Disabled

C3 = Disabled

C6/C7 = Disabled

CPU Thermal Monitor = Auto

EIST = Disabled

 

XMP Profile = Off

RAM = 1333MHz (down-clocked to remove it as a potential cause for instability)

Memory Voltage = 1.50V

 

Basically, my voltages are set to STATIC for stress testing. Temperatures are just fine. Nothing over 66 degrees Celcius during tests. I have an AIO Cooler. Average temperature is 57 degrees celcius.

 

THE PROBLEM:

 

I'm finding instability when I increase the Uncore Clock to 4.4GHz @ 1.20V (VRING). I've already tried to increase the VRIN to 2.1V and 2.0V, but I'm still unstable.

 

I'm trying to bring the Cache ratio up to a 1:1 ratio with the Core along with the voltages. Where should I go from here? I asked about increasing the VRING because I thought that would potentially bring about stability.

 

Scored 431 marks with my overclock (as noted above) on IETU and 500 marks on default. *sigh* -___-

 

Thank you!

[AngryGoldfish]

I'm concerned why you scored lower (much lower) when overclocked compared to stock. Was your 431 mark a typo? I just quickly ran the test on my 24/7 4.4Ghz overclock and scored 900.

[BigDay]

I'm concerned why you scored lower (much lower) when overclocked compared to stock. Was your 431 mark a typo? I just quickly ran the test on my 24/7 4.4Ghz overclock and scored 900.

 

Not sure why IETU scored me at 431 with my OC, but 500 @ stock. Forgot to mention that I had my ram dialed down to 1333 mhz when I was benchmarking on IETU at my overclock settings. When I benchmarked using stock settings, it was up to 1600 mhz.

 

I tested using CINEBENCH to try out a new utility benchmark and i got the following results:

 

4.4GHz Core Frequency @ 1.20V

3.6GHz Uncore Frequency @ 1.20V

VRIN = 1.90V

 

Cinebench Score: 608

 

Loaded optimized defaults on my board:

 

3.6GHz Core Frequency @ Auto

3.6GHz Uncore Frequency @ Auto

VRIN = Auto

 

Cinebench Score = 491

 

Still concerned?

[AngryGoldfish]

Not really. IETU isn't the most stable program out there. It reads things wrong from time to time. Just make sure you're not throttling (though I see no reason for it) and you should be fine.

 

 

Honestly, 4.4Ghz at 1.2V is not that much. You should be able to hit 4.5Ghz with 1.3V easily. That will net a bigger performance increase over Ring Ratio. You may even be able to hit 4.6Ghz at something like 1.32V, which would be a stellar overclock without custom water cooling. Give it a try. 'Dirty' overclock your CPU to 4.6 @1.3V (stock ring ratio) and see if you can pass the stress tests. If you don't, either increase the voltages as per normal or back down to 4.5Ghz. 4.5Ghz is undoubtedly the sweet spot for the 4670K. That's where I've settled for my max overclock at 1.3V (cache ratio at stock currently).

 

But if you're happy with 4.4Ghz and want to eke out more performance, decrease your cache ratio 4.0Ghz at 1.2V. Try that. Very few Haswell chips can hit 1:1. Bear that in mind.

[arniks8]

Not sure why IETU scored me at 431 with my OC, but 500 @ stock. Forgot to mention that I had my ram dialed down to 1333 mhz when I was benchmarking on IETU at my overclock settings. When I benchmarked using stock settings, it was up to 1600 mhz.

I tested using CINEBENCH to try out a new utility benchmark and i got the following results:

4.4GHz Core Frequency @ 1.20V

3.6GHz Uncore Frequency @ 1.20V

VRIN = 1.90V

Cinebench Score: 608

Loaded optimized defaults on my board:

3.6GHz Core Frequency @ Auto

3.6GHz Uncore Frequency @ Auto

VRIN = Auto

Cinebench Score = 491

Still concerned?

IETU scores higher with RAM clock and timings set to their maxium.

[BigDay]

Not really. IETU isn't the most stable program out there. It reads things wrong from time to time. Just make sure you're not throttling (though I see no reason for it) and you should be fine.

 

 

Honestly, 4.4Ghz at 1.2V is not that much. You should be able to hit 4.5Ghz with 1.3V easily. That will net a bigger performance increase over Ring Ratio. You may even be able to hit 4.6Ghz at something like 1.32V, which would be a stellar overclock without custom water cooling. Give it a try. 'Dirty' overclock your CPU to 4.6 @1.3V (stock ring ratio) and see if you can pass the stress tests. If you don't, either increase the voltages as per normal or back down to 4.5Ghz. 4.5Ghz is undoubtedly the sweet spot for the 4670K. That's where I've settled for my max overclock at 1.3V (cache ratio at stock currently).

 

But if you're happy with 4.4Ghz and want to eke out more performance, decrease your cache ratio 4.0Ghz at 1.2V. Try that. Very few Haswell chips can hit 1:1. Bear that in mind.

 

Ok sounds good dude. I'll give this a shot now.

[BigDay]

Do I really have to keep stress testing for a minimum of 8 hours on IETU? This is taking forever...

[AngryGoldfish]

Do I really have to keep stress testing for a minimum of 8 hours on IETU? This is taking forever...

 

I personally don't think so. I think 3-4 hours is sufficient when fine-tuning little details. Then when you feel like it's stable and you're happy, run a 8-24hr test to finalise it. Running a new 8hr test after one small change is silly, in my opinion, when even that won't reveal all instabilities. I've noted people crashing 18 hours into a 24hr test. That could be attributed to something else, but it's interesting. Also, artificial stress tests are not the final tests. Using the computer normally is.

 

For your initial dirty overclocks, stress test for a minimum of 8 hours. Once you've found a stable frequency you like, then revert to 4hr tests. Then when you've finalised your entire overclock, run a 24hr test, benchmark multiple times, game, render a random video, etc.

 

That's my 2c. Others may disagree.

 

[AngryGoldfish]

Also, don't think of it as a task that you HAVE to do by a certain date. Continuously tweak it over time. That's what I do. Last night I ran a new test for my 4.5Ghz overclock. It passed, but it's not 100% stable in my opinion. I will game, run benchmarks, render an image repeatedly, fine-tune the voltages, etc. You can never get it 100% stable, so don't worry too much about doing it all in one go. Take a month to finish your overclock. You may have to correct it down the line anyway when a new game comes out, you update your motherboard's BIOS or your CPU begins to leak voltage slightly and becomes unstable.

[BigDay]

Thanks for the input bro

 

I got 4.6 GHz @ 1.30V to pass an 8 hour stress test with IETU. Uncore ratio was 34 with a 1.20V Vring and a 1.90 VRIN

 

Currently, I'm testing 4.6 GHz Core @ 1.29 Vcore. Just had 1.28 Vcore fail after 4 hours

[AngryGoldfish]

Thanks for the input bro

 

I got 4.6 GHz @ 1.30V to pass an 8 hour stress test with IETU. Uncore ratio was 34 with a 1.20V Vring and a 1.90 VRIN

 

Currently, I'm testing 4.6 GHz Core @ 1.29 Vcore. Just had 1.28 Vcore fail after 4 hours

 

4.6Ghz @1.28V would have put you in the top 20-30%. The average for the chip is 4.5Ghz @1.3V, so hitting 4.6Ghz is in the top 40-50%, I'd say. Nice one. Remember to test your overclock with benchmarks (Cinebench is a good one) as well.

[BigDay]

Thanks for all the help so far guys!

 

Update: June 8, 2015 @ 7:30 AM:

 

Stable at:

 

4.6 GHz Core Frequency @ 1.29V Vcore

3.4 GHz Cache Frequency @ 1.20V Vring

1.90V VRIN

Maximum Temperature = 76 °C

Average Temperature = 69 °C

 

Currently testing:

 

4.6 GHz Core Frequency @ 1.29V Vcore
3.9 GHz Cache Frequency @ 1.29V Vring

2.10V VRIN

Recent Maximum Temperature = 77 °C

Recent Average Temperature = 69 °C

 

I've had to lower the Cache Ratio from 46 to 39 over several tests in order to find stability. During that time, I ran a VRIN voltage of 2.10V and a Vring of 1.29V, but I still couldn't find stability with a higher Cache Ratio.

 

If I pass the above, I will start lowering the VRIN and the Vring voltages. That being said, I'm not sure which one I should lower first... suggestions?

[AngryGoldfish]

Although I have no personal experience with that particular question, the guide in this thread suggests reducing the Input Voltage last. So lower your Vring first, find instability, go back up one and retest. Then begin lowering your input voltage as a final note.

 

Your temperatures look very good. The cache ratio being lowered to 3.9Ghz is not a big deal at all. That is very common.

[BigDay]

Although I have no personal experience with that particular question, the guide in this thread suggests reducing the Input Voltage last. So lower your Vring first, find instability, go back up one and retest. Then begin lowering your input voltage as a final note.

 

Your temperatures look very good. The cache ratio being lowered to 3.9Ghz is not a big deal at all. That is very common.

 

Thanks for the quick response AngryGoldfish

 

I agree, lowering the Vring first seems like the best course of action. I will lower the VRIN afterwards

 

After I find stable settings for 4.6 GHz on the core, I may push up to 4.7 GHz on the core and see if I can find stability with safe temperatures

 

I ran some tests with 4.7 GHz on the core and I was unable to find stability with a 1.33V Vcore, a Cache Ratio of 34, Vring of 1.20V, and a VRIN of 1.90V. Looks like I'll have to keep pushing

[AngryGoldfish]

4.7Ghz is pretty illusive for most Haswell users. Haswell refresh allowed for higher overclocks due to improved yields in the production and a better heat spreader underneath the lid. For many, though, the only way to achieve those kind of numbers was with exotic cooling or dealing with 80-90°C. I don't like seeing those numbers personally so I never tried it. I'm happy with 4.5Ghz. Piling on voltage will only reduce the lifespan anyway.

[BigDay]

4.7Ghz is pretty illusive for most Haswell users. Haswell refresh allowed for higher overclocks due to improved yields in the production and a better heat spreader underneath the lid. For many, though, the only way to achieve those kind of numbers was with exotic cooling or dealing with 80-90°C. I don't like seeing those numbers personally so I never tried it. I'm happy with 4.5Ghz. Piling on voltage will only reduce the lifespan anyway.

 

Very interesting. I may just stick with 4.6 GHz if the temperatures get too crazy at 4.7GHz. I'm not too keen on running my CPU with voltages past 1.30V on the Vcore and Vring

 

On another note, I just ran some benchmarks as I'm taking a break from testing. Here's what I got with CineBench R15.0:

 

Haswell i5-4670k stock settings; score: 491

 

Overclocked to:

 

4.6GHz Core Frequency @ 1.29V Vcore

Auto (4.0GHz) Cache Frequency @ Auto Vring

1.90V VRIN

1600MHz ram

 

Core settings are stable. Not sure if the Cache settings are stable though. I'll find out soon enough

 

Score: 682

 

An increase in performance of 38.9% from stock

[AngryGoldfish]

Nice one. I just ran Cinebench and scored 655, so your scores are on level with mine.

 

What XMP is your RAM? Are you interested in overclocking your memory? I might give it a try soon. I doubt it'll get much higher than 1866Mhz, though.

[BigDay]

Nice one. I just ran Cinebench and scored 655, so your scores are on level with mine.

 

What XMP is your RAM? Are you interested in overclocking your memory? I might give it a try soon. I doubt it'll get much higher than 1866Mhz, though.

 

Cool! Yeah our clocks are close

 

I haven't touched my RAM. During testing, I leave I downclock it to 1333MHz. Outside of testing, it goes back to "Auto", which is @ 1600MHz @ 1.5V. I need to study up on memory. I don't know much about it. I know how to overclock it, but that's about it.

 

I heard there is not much of a performance increase when you OC your RAM. Is this true?

[AngryGoldfish]

Cool! Yeah our clocks are close

 

I haven't touched my RAM. During testing, I leave I downclock it to 1333MHz. Outside of testing, it goes back to "Auto", which is @ 1600MHz @ 1.5V. I need to study up on memory. I don't know much about it. I know how to overclock it, but that's about it.

 

I heard there is not much of a performance increase when you OC your RAM. Is this true?

 

It depends on the title or application. 1600Mhz is the standard, but higher memory clocks can yield benefits. I've seen 3000Mhz DDR4 take 5 frames or more over 1600Mhz DDR3, but this could be due to the architecture. It's probably on par with Cache Ratio overclocks. Some applications love it, some couldn't care less, others are amicable. I might try it for fun.