I have finished overclocking tests on my new Haswell 4700k setup and I wanted to get some opinions on the test results and overclock configuration.

I have settled with this following multipler/voltages setup, which is fully stable using all 3 tests from the NickN Haswell Overclocking guide which tests for:

 

1. Temp and load (AIDA64 1 hour)

2. Stability (OCCT large 64bit 1 hour)

3. Cache stability (OCCT large 64bit 1 hour)

 

Core/Cache CPU Core Voltage Override CPU Cache Voltage Override CPU System Agent Voltage Override VCCSA Analog I/O Digital I/O Test

Max

Temp

4.4/4.4 1.300 1.300 0.290 0.050 0.100 Phase1 (AIDA max temp) 77

 

This extract is from a spreadsheet I made for testing purposes to keep track of all the configs/temps.

I worked my way up from 4.0/3.9 to where I'm at above.

 

Please let me know if these numbers are reasonable.  The hardware config is in my profile on the left, and the HAF 932 case is enormous, much more roomier than the midtowers I am used to, but I specifically bought it because NickN recommended it for it's airflow design/fan setup.   I'm pretty sure I put on the thermal paste correctly, but I have no reservations about reapplying if it can shave off some more heat from the cpu.   But all told, I am happy to have a 1:1 core/cache ratio, and I shouldn't be greedy, but is there any point to going for 4.5 by further tweaking the voltages, or should I consider myself lucky and just accept what I have here?

 

 

Is that with all options in the Aida64 stress test checked or just FPU?

Vcore seems a little high to me, but that's purely based on my chip. I'm at 4.5Ghz 1.225v

 

These are my settings I posted over at overclock.net

 

Core Multiplier:45
Vcore: 1.225
Uncore Multiplier: 42
Uncore Voltage: 1.18
VCCin:1.82
Cooling Solution: Corsair H110
Stability Test: Prime 95 (24 Hours), IBT, AIDA64, Cinebench, Handbrake, General Use
Batch Number: L307B246
Ram Speed: 2400Mhz XMP

Is that with all options in the Aida64 stress test checked or just FPU?

Just FPU, per NickN's exact instructions.

 

Vcore seems a little high to me, but that's purely based on my chip. I'm at 4.5Ghz 1.225v

 

These are my settings I posted over at overclock.net

 

Core Multiplier:45

Vcore: 1.225

Uncore Multiplier: 42

Uncore Voltage: 1.18

VCCin:1.82

Cooling Solution: Corsair H110

Stability Test: Prime 95 (24 Hours), IBT, AIDA64, Cinebench, Handbrake, General Use

Batch Number: L307B246

Ram Speed: 2400Mhz XMP

wow, I had a feeling 1.30 is a bit high.  And you basically have the same setup as me, except for the Mobo, CPU heatsink, and the case (not sure what you have there), but I presume my case is more than capable of allowing proper airflow for a 4.4/4.5 clock.   Cable management is excellent, so I can count any hindrance to airflow out, and the heatsink is powerful with two fans.  What steps should I take to reduce the Vcore?  Should I back down on the Uncore multiplier?

Just FPU, per NickN's exact instructions.

 

wow, I had a feeling 1.30 is a bit high.  And you basically have the same setup as me, except for the Mobo, CPU heatsink, and the case (not sure what you have there), but I presume my case is more than capable of allowing proper airflow for a 4.4/4.5 clock.   Cable management is excellent, so I can count any hindrance to airflow out, and the heatsink is powerful with two fans.  What steps should I take to reduce the Vcore?  Should I back down on the Uncore multiplier?

 

 

 

Don't forget, we are all subject to the silicone lottery.

 

My 3770K for example requires higher voltage than most.

 

By all means investigate overclocking fine tuning, in an attempt to reduce voltage...

 

but it  is possible that you simply have a chip that likes volts.

My case is a Corsair Carbide 500R.

 

The process I followed was leave uncore at default multi and push the core multiplier first. Get that Stable.

 

Then push up the uncore multi and get that stable.

 

It may be more difficult or more vcore may be required if you are trying to set uncore multi to be the same as the core multi. The performance benefit is also virtually negligible.

 

You're best to set your uncore multi 300-500mhz lower than your core multi and hopefully reap the benefits of lower vcore, uncore voltage and VCCin and thus lower temps.

 

It is a bit of a chip lottery though so you just need to play around and see what your chip is capable of.

 

 

But all told, I am happy to have a 1:1 core/cache ratio, and I shouldn't be greedy, but is there any point to going for 4.5 by further tweaking the voltages, or should I consider myself lucky and just accept what I have here?

 

My friend, do yourself a giant favor and go w/ the latter-consider yourself lucky.  Everyone w/ modern hardware overclocked anywhere in the range you are already in have found the promised land, for all practical purposes.  Going from 4.4 to 4.5 is absolutely positively much ado about almost nothing, and almost nothing is rarely ever relevant, FSX 'performance' notwithstanding.  Where it does matter to some is in the domain of overclocking milestones, but don't be fooled into thinking getting to that next .1Ghz higher will make or break nirvana in FSX--it just doesn't matter nearly so much as customizing your specific configuration for the simulator.  If it's FSX 'performance' you are really doing this for the place to spend time is making sure you have the simulator set up for results emphasizing what you care most about, and that changes according to what you fly in and where.  Everyone here, even those w/ overclocked rigs to >5.0Ghz must still decide what to emphasize in their simulator setup, else it would be a simple matter of moving ALL sliders fully right--that's right, road and air and airport traffic at 100%, every single variable set to the absolute max, in any kind of scenery addons available, in every place and in every plane.  No one maintains the holy grail of 30 frames locked in this scenario, and if they do which they can't, I guarantee you someone else will demand 60 frames locked is where true nirvana happens, and then you're right back to settling on settings you can live with.

 

Be happy with that new build as it is guaranteed to do well provided you emphasize the software configuration refinement piece, OS & simulator.  

Getting a max temp of 77c in AIDA64 100% FPU with an air cooled 4770k running 4.4 @ 1.30v is pretty amazing.  Unless you've either done a delid (IHS mod) or have goofed up your runtime settings, that max temp with your system seemingly defies explanation.

What is you max temp when running OCCT Linpack AVX 64 bits.  This is the temp that you want each core under 92C as shown in HWMonitor.  This is the test to use to make sure that your OC will not run to hot if it ever gets slammed by a program that runs AVX instructions.

 

As far as your clock voltages.  To find you lowest vcore reduce you vcore by 0.01 (1.30 > 1.29) and run OCCT CPU Large Data Set 64 bits. If it passed reduce it by 0.01 and test again.  Do this until it fails and then raise your vcore back to the last value it passed.  Then run OCCT Linpack AVX 64bits and confirm it will pass.  I have never had OCCT Linpack AVX 64bits fail if OCCT CPU Large Data Set 64bits has already passed, but you need to check just the same.

 

After you find your lowest vcore do the exact same thing for CPU Cache and then System Agent Offset voltage.

 

For 44/44 a vcore of 1.30 is kind of high, but all chips are different and your chip may just need that much.

This is the latest official LINPACK from Intel built just a few weeks ago. It will melt your 4770k: I can 'safely' run OCCT AVX linpack with my OC, but running the .bat file in this version of LINPACK instantantly sends my cores to 100c. Runing this non-OC I hit over 180Gflop, which is amazing and probably explains the radical heat.

http://registrationcenter.intel.com/irc_nas/3669/w_lpk_p_11.1.1.003.zip

I have settled with this following multipler/voltages setup, which is fully stable using all 3 tests from the NickN Haswell Overclocking guide

I question the value of using artificial benchmark routines to establish stability unless the real world application puts the same stress as the the artificial benchmark routine will. I learned this from listening to a video on youtube by JJ of ASUS who was describing overclocking SB-E platforms. I think the logic is valid, and here's why it may not be best to base your evaluation of stability on AIDA64, Prime95 or other: if you really only need, for example 1.28v to get to 4.4Ghz stable in FSX, not necessarily AIDA64 etc, you can end up putting your processor thru more volts/heat than in truly needs. I've used a simpler approach: I run Prime95 for 15 minutes at the lowest vCore that gets no errors for 15m, then use that for FSX, dialing back .05v. I am able to run my SB-E hexacore at 4.3Ghz w/ HT enabled at vCore of 1.29. Temps stay under 60C always. I have yet to see a BSOD w/ this approach. I've dabbled in 4.4Ghz, but it's just so little difference I can assure you in a double-blinded test you couldn't guess whether FSX was running at 4.4 or 4.3. I'm guessing this guy will go up to 4.6 or higher, but I just don't see the value from an FSX performance viewpoint, nor the steep ramp up in temps/watts that happens when you o/c to the bleeding edge. Also, I wanted to see how IB-E did before risking shortening this proc's life as it's really a nice proc. Looks like IB-E will be a suitable replacement should I decide to go for a more stress-inducing o/c for my SB-E chip, which I may do sometime just for kicks. I'd want to see at least a 7% bump in performance before subjecting this proc to undo stress/heat/watts. If I put a 2nd Titan in my machine I would do this after changing to IB-E which is a lower power user than SB-E and so my 850W PSU should be able to handle SLI'd Titan by every PSU calculator I've tried this on.

Just FPU, per NickN's exact instructions.

 

wow, I had a feeling 1.30 is a bit high.  And you basically have the same setup as me, except for the Mobo, CPU heatsink, and the case (not sure what you have there), but I presume my case is more than capable of allowing proper airflow for a 4.4/4.5 clock.   Cable management is excellent, so I can count any hindrance to airflow out, and the heatsink is powerful with two fans.  What steps should I take to reduce the Vcore?  Should I back down on the Uncore multiplier?

 

 

Vcore seems a little high to me, but that's purely based on my chip. I'm at 4.5Ghz 1.225v

 

These are my settings I posted over at overclock.net

 

Core Multiplier:45

Vcore: 1.225

Uncore Multiplier: 42

Uncore Voltage: 1.18

VCCin:1.82

Cooling Solution: Corsair H110

Stability Test: Prime 95 (24 Hours), IBT, AIDA64, Cinebench, Handbrake, General Use

Batch Number: L307B246

Ram Speed: 2400Mhz XMP

 

The Corsair H110 is a liquid cooler isn't it?

 

I think it's a bit easier to achieve higher clock speeds with liquid cooling as opposed to air cooling like the OP is using.

Just a follow up here, I understand that my chip may just need high voltage to maintain 4.4 core, it's the luck of the draw right?.   But I will try tests to lower the vcore, such as Idahosurge's suggestion to use OCCT cpu and linpack tests for this testing.  1:1 ratio of core/cache (or uncore) is really not necessary, as 300-500mhz lower as mentioned above and by NickN is fine.  So, I lowered uncore to 4.0ghz, dropped vcore from 1.30 to 1.29, and I'm presently running NickN's 3 tests to verify temp and stability, followed by Idahosurge's OCCT test to tweak vcore lower if possible.

 

I'm perfectly happy with 4.4 core, I could care less about hitting 4.6 or above as I'm not an overclock enthusiast, and I have no plans on delidding the Haswell or getting water cooling which is a prerequisite for hitting that speed mark.  My goal now is just to bring down the vcore as much as possible while remaining stable. 

I'm currently testing with lower vcores, I went from 1.300 to 1.280 stable, and I'm running 1.27 now.

I wanted to know, what is the value in trying to have the uncore(cache) multiplier match or closely match the core multiplier?   Is there any performance benefit whatsoever, is it worth my time trying to bump up from 4.4/4.0 to 4.4/4.1 or greater after I find the lowest voltages for vcore/CPU Cache/System Agent Offset?

I'm currently testing with lower vcores, I went from 1.300 to 1.280 stable, and I'm running 1.27 now.

I wanted to know, what is the value in trying to have the uncore(cache) multiplier match or closely match the core multiplier?   Is there any performance benefit whatsoever, is it worth my time trying to bump up from 4.4/4.0 to 4.4/4.1 or greater after I find the lowest voltages for vcore/CPU Cache/System Agent Offset?

 

 

Core clock is king so far as performance goes, so you could have one chip running 4.5Ghz cpu multi/4.2Ghz uncore multi and it will outperform one set to 4.4Ghz/4.4Ghz but probably at lower voltages.

 

Running the uncore equal to the core multi offers negligible performance improvements (you won't know the difference unless you're looking at a benchmark result and even then it will be minimal). It can also potentially result in the need for increased voltages and thus leading to increased temps or even limit your ability to run fast RAM at it's rated speed.

 

Keeping them within 300-500Mhz of each other means that you will hopefully be able to benefit from being able to run lower voltages and lower temps, so it's definitely the way to go. As far as bumping uncore from 40 to 41, that's up to you but it probably isn't necessary since you're within the 300-500mhz window. I'm running mine at 4.5Ghz/4.2Ghz purely because I wanted to be at top end of that recommended window.

 

 

 

The Corsair H110 is a liquid cooler isn't it?

 

I think it's a bit easier to achieve higher clock speeds with liquid cooling as opposed to air cooling like the OP is using.

 

H110 is an AIO liquid cooler, but based on temps he's getting he's not doing badly at all, but there is probably room for fine tuning which is what he's looking at now.