i5-6600K (pre-)overclocking advice

Will do! LLC is still at level 5.

I can't seem to get a consistent BIOS equals CPU-Z readout voltage yet, even before benching. I've went through this list up to now:


BIOS | CPU-Z

1.250 | 1.264
1.240 | 1.248
1.230 | 1.232 (jumped down from 1.248 shortly after OCCT started)
1.220 | 1.232 (unwavering)
1.210 | 1.216 (jumped down from 1.232 shortly after OCCT started)
1.200 | 1.216 (jumped down from 1.216 shortly after OCCT started) *
1.195 | 1.216 (not stressed due to BIOS/CPU-Z not matching instantly)
1.190 | 1.200 (unwavering)
1.180 | 1.184 (jumped down from 1.200 shortly after OCCT started)
1.170 | x.xxx (reboot upon opening OCCT)
1.175 | 1.184 (unwavering)
1.185 | 1.184 (jumped down from 1.200 shortly after OCCT started) *

* = only values that jumped back and forth between BIOS and CPU-Z read out values during OCCT stress testing.

Honestly, I don't know which voltage to pick from here on, as I assume I need to look for one where what I set in the BIOS is equal to the CPU-Z readout (at idle as soon as I have booted into Windows as well as during OCCT) and does not waver.
 
Last edited:
1.185 seems to be where you need to be. Try lowering LLC down a few notches and see what is reported then.
 
In Windows in idle 1.185v shows 1.200v in CPU-Z, and under stress in OCCT it drops to 1.185v, jumps back and forth between that and 1.200v, and returns to the latter after the stress test. Hence I thought that'd be wrong.

I'll lower the LLC. Would you advise me to lower it one level at a time?
 
would be interesting to see how much difference LLC makes at each level. So lower it 1 level at a time. This thread is proving to be an interesting look at OCing in z170
 
Yeah it is! Okay, both LLC levels 3 and 4, 1.185v BIOS is still 1.200v CPU-Z idle (and in the BIOS hardware monitor upon restart > BIOS). Don't think I need to run OCCT until in idle it matches what I set in the BIOS?

Edit 1:

All of the above, LLC level 3, OCCT, jumps down from 1.200v to 1.184v, then further down to 1.168v shortly after and stays there for a couple of minutes before Windows crashed and rebooted. Will see what LLC level 4 does.

Edit 2:

All of the above but LLC level 4, OCCT, jumps down from 1.200v to 1.184v and stays there for the remainder of the quick 10 minute test. Afterwards it goes back up to 1.200v while the core clock is down to 800MHz.
 
Thanks to both of you with regards to this thread. I'm keeping a close eye and will be using bits from here when OCing my 6600k as i'm seeing a lot of consistency between F34R's results and mine
 
My pleasure and I'd like to take this opportunity to thank all those who are invested in it and helping me out - much appreciated!

I'm hoping to have stabilised my overclock soon, it gnaws at me lol
 
No problem dude, it's what we are here for. If your anything like the rest of us, once you stabilise your OC it will niggle at the back of your head about wanting to see if you can get it a little further!

The voltage jumps I assume therefore are due to the multiplier changes. This is one of the reasons that most people disable state changes within the BIOS. Personally I would still be happy getting the voltage you currently have. Have you considered pushing the multiplier up a little more yet?
 
Last edited:
Exactly the same as you have already been doing. Increase multiplier and find a voltage that suits that clock speed. You will eventually get to a point that the voltage required will start exceeding the benefit. each to their own at that point, I personally like a good speed at as low a voltage as possible, for me that is 1.25v. The chip will happily take higher than that safely, but keep in mind that the more voltage you put through will add more heat and therefore reduce the longevity of the chip itself. I also aim for temps below 60c to keep my system quiet.

Push to see how far your happy taking it, then you can decide from there what sort of compromise you want to settle on.

I am trying not to give any set numbers or specific guide to what you should do. I will guide based on the theory behind what is happening but it ultimately is down to you to learn your chip. Your doing fine so far. Don't worry too much about getting something wrong, modern CPU's and Mobos are pretty idiot proof, just don't go looking for 5+Ghz @ 1.5v etc. Keep within a safe guideline and work slowly.
 
That makes sense, however what does not is that at low clocks it gets more volts than at high clocks. Also, I've been watching CPU-Z clock speeds while stress testing and the voltage differences don't occur with changing mp/clock speeds.

I can game fine for hours at the 1.185v (BIOS) | 1.184v / 1.200v (CPU-Z) without any hick-ups (in-game it seems to sit at 1.184v) so to speak so it would appear stable (well, it already was in OCCT which puts much more stress on the CPU than a game (AC: Syndicate in this case) would). Highest temp (measuring all cores separately) so far during the gaming session is 52 C.

I would like to keep the states enabled, since I equally use my system for gaming and normal use. However, if it helps finding that perfect voltage I could always disable the states (which ones then?) and re-enable them after?

No haven't really thought about it yet, mainly because I feel that 4.5GHz provides me with everything I need, and I don't bench.

I'm learning a lot here and it's real fun too! This hardware feels so much smoother to work with than my old setup.
 
Last edited:
Intel speedstep is what changes the core frequencies on the fly. Turn that off to see if your voltage stays stable. Also C1E state will drop your Vcore when the processor is idle.

You would be surprised how often a game will pick up a unstable OC more than OCCT. For example, I can have my chip stable at 4.6Ghz with quite a low voltage, it will run OCCT for as long as I want it to within good temps. However GTA V will blue screen every time at the same settings until I bump the voltage. It is all down to how games and synthetic benches use the CPU. Consider the fact that OCCT will use 100% for x amount of time then drop to allow a cooling period before working steadily to 100% again. Games and other software will use a random CPU core count and load all the time dependant on what the engine requires. That constant change will require a more stable voltage delivery than synthetic tests.

If you are happy @4.5Ghz then you can leave it as it is. The voltage variations are safe and will not effect the stability or increase your temps.

You can take this opportunity to look more into other features that skylake has to offer, BCLK can be changed without having to worry about other system ratios etc as it was decoupled. 20x multiplier with a 200 BCLK will get you 4Ghz as well as 40 x 100 will. The chips are remarkably robust and Intel have listened to the end user about what they wanted to see in the chips. BCLK however is yet another thing you can look at when you want to push further as it can be used to help stabilise high OC's. Past that there is your Uncore/cache, plus many other voltages that all start getting complex and bemusing.

Next comes GPU Overclocking, however this does come with a lot higher risk if pushed too far and I would highly recommend a good water block to keep everything cool and lets be honest, looking good :p.

Are you running XMP profile on your RAM? This is yet another area you can set manually and often push a little more speed from.

Overclocking is by far a really, really deep rabbit hole. New hardware has made it available to the masses in an easy to understand way, especially with UEFI BIOS allowing tips to what each setting does. For the majority of users, increased CPU speed is more than enough to feel they have received a good deal on their hardware. I mean who doesn't appreciate easily attainable "free" upgrades.
 
Intel speedstep is what changes the core frequencies on the fly. Turn that off to see if your voltage stays stable. Also C1E state will drop your Vcore when the processor is idle.

I will give that a go in a minute, and turn off SpeedStep and C1E, cheers! Weird thing is that during OCCT is uses less voltage than in idle; I would've expected it to be the other way around. I will set the LLC back from 4 to 5; less Vdroop and for the voltage it didn't matter.

You would be surprised how often a game will pick up a unstable OC more than OCCT. For example, I can have my chip stable at 4.6Ghz with quite a low voltage, it will run OCCT for as long as I want it to within good temps. However GTA V will blue screen every time at the same settings until I bump the voltage. It is all down to how games and synthetic benches use the CPU. Consider the fact that OCCT will use 100% for x amount of time then drop to allow a cooling period before working steadily to 100% again. Games and other software will use a random CPU core count and load all the time dependant on what the engine requires. That constant change will require a more stable voltage delivery than synthetic tests.

Thanks for explaining that, I get it now.

If you are happy @4.5Ghz then you can leave it as it is. The voltage variations are safe and will not effect the stability or increase your temps.

Awesome! Yeah I'm dead happy with 4.5GHz / 4.1GHz (Core / Cache)!

I entered the BIOS upon a cold boot this morning, and there it said 1.200v too in the hardware monitor tab on the right.

So, I'll leave things as they are then, and will make a post later which lists all the current settings, perhaps they can be useful guidance for someone else (I know all parts are individually different).

You can take this opportunity to look more into other features that skylake has to offer, BCLK can be changed without having to worry about other system ratios etc as it was decoupled. 20x multiplier with a 200 BCLK will get you 4Ghz as well as 40 x 100 will. The chips are remarkably robust and Intel have listened to the end user about what they wanted to see in the chips. BCLK however is yet another thing you can look at when you want to push further as it can be used to help stabilise high OC's. Past that there is your Uncore/cache, plus many other voltages that all start getting complex and bemusing.

Next comes GPU Overclocking, however this does come with a lot higher risk if pushed too far and I would highly recommend a good water block to keep everything cool and lets be honest, looking good :p.

Yeah earlier I wanted to try a 190 BCLK with a 24 MP but I might keep that for a later time.

GPU overclocking I've done a lot, but in all honesty I don't need it now; my Strix is mighty fast! I usually do it when I need the extra speed.

Are you running XMP profile on your RAM? This is yet another area you can set manually and often push a little more speed from.

Yeah, XMP for 3GHz. I'll get a little extra (which for gaming makes little no difference, so I'm good) if I do the BCLK OC.

Overclocking is by far a really, really deep rabbit hole. New hardware has made it available to the masses in an easy to understand way, especially with UEFI BIOS allowing tips to what each setting does. For the majority of users, increased CPU speed is more than enough to feel they have received a good deal on their hardware. I mean who doesn't appreciate easily attainable "free" upgrades.

Right you are! And there's so much difference in accessibility and options as compared to my previous setup; things are definitely different and the better for it.

Edit 1:

Disabled SpeedStep and all C states (can't do them separately) and ran OCCT to look at voltages. It was very jumpy between 1.184 and 1.200v - in fact, it seemed better with the C states enabled (on auto, to be precise).

Edit 2 (2 December):

Thought I might display voltage in-game and see how it behaves. It sits at a solid, unwavering 1.200v, which is fine by me.
 
Bit of a thread revival since i've been playing with my overclock and thought i'd report here since there's already quite a lot of good content.

my BIOS settings are as follows...

XMP: Enabled

Core ratio: 46 (4.6GHz)

FCLK Frequency: Upped to 1GHz

Speedstep: Disabled

Min/Max CPU Cache ratio: 43 (4.3GHz)

Core Voltage: 1.36v


Currently done a run of RoG Real Bench which was stable and in the process of doing 7 Hours of x264. few hours in now and running stable.

Temps are maxing out at around 65 Degrees on these benchmarks

Thought? Comments?
 
Bit of a thread revival since i've been playing with my overclock and thought i'd report here since there's already quite a lot of good content.

my BIOS settings are as follows...

XMP: Enabled

Core ratio: 46 (4.6GHz)

FCLK Frequency: Upped to 1GHz

Speedstep: Disabled

Min/Max CPU Cache ratio: 43 (4.3GHz)

Core Voltage: 1.36v


Currently done a run of RoG Real Bench which was stable and in the process of doing 7 Hours of x264. few hours in now and running stable.

Temps are maxing out at around 65 Degrees on these benchmarks

Thought? Comments?

Certainly a good OC dude. Temps are good. Have you tried to bring the Volts down a little? Obviously silicon lottery so could go down by a lot or nothing.
 
Certainly a good OC dude. Temps are good. Have you tried to bring the Volts down a little? Obviously silicon lottery so could go down by a lot or nothing.

well my bios setting is 1.365 at the moment with cpu-z and HWMonitor reading out 1.36 in OS. Initially I tried 4.6 at 1.35 volts but couldn't get through a full 15 minutes of real bench so I think 1.36 is prob gonna be right on the money. We'll see how the 7 hour run goes and then I may try reduce BIOS voltage down to 1.36 from 1.365.

Also debating whether to try putting the cache clocks to 4.5 or even 4.6 to match the core clock??

Cheers for the initial feedback though
 
Cache clock is a very sticky point with OC's. Real world performance boost is neither here nor there. However it does become very useful when pushing for memory speeds and stabilising higher overclocks.
 
Cache clock is a very sticky point with OC's. Real world performance boost is neither here nor there. However it does become very useful when pushing for memory speeds and stabilising higher overclocks.

I guess as mine is already at 4.3GHz it's not that urgent. I too read that there's v lil real world benefit to it but I upped it a little just to be sure
 
Hello fellas, gonna join the thread with my first post.

I want to ask how good would be to get the NB from 3500 to 4400 ish if the other speeds are the same?

I am able to run this 6600k at 4,5 ghz with 3ghz ram set at either 45x100 or 25x180 at 1.35v, the question is, should I prefeer the bclk change?

Ram set to 3000mhz vs 2880@180bclk*16

Another small newbie question, when I saw the results from the oc3d post about oc 6600 he was able to get his ram to 3.2ghz~ and that did show in the cpuz with the other info, why in my case cpuz it shows only half of the speed?

Tyvm!!!
 
In which tab, maybe you can link the pic? It should show halve the actual speed.

I would go for the MP only OC; you get the same clocks but the higher RAM speed.

I would like to see an in-depth article on MP vs MP + BCLK OC'ing, could be interesting!

Additionally I am also interested in reading an in-depth article on Skylake cache overclocking. When what comes into play / is beneficial, the exact numbers, etc. I haven't found one yet. Right now I have min 35, max 41.

I did find this (someone's findings):

Regarding that cache matter. Cache clock directly affects memory performance. Simply when you set higher cache clock then memory performance will be higher but there is some point at which performance is not scalling good ( usually above 5000MHz is lower than expected, or maybe only for me ).
If you are not overclocking cache then memory performance using DDR4-3000 and DDR4-3733 kits will be almost the same. When you set higher cache frequency then you notice that memory bandwidth will go up but at 3000 it will stop scalling faster than on 3733 memory clock. However for CPU/cache clocks up to ~4.7-4.8GHz ( what is usually limit on water cooling ) you won't need much more than DDR4-3000.

When you are overclocking CPU on Skylake boards then CPU and cache has linked voltage. In this case best is to set CPU clock = cache clock for optimal performance. At clocks above 5GHz it can be hard but on air/water I doubt you will stabilize it at 5GHz or above.

I don't think that core equals cache clock is necessary perse, the benefits might be very few depending on the component speeds obviously and scenarios.
 
Last edited:
In which tab, maybe you can link the pic? It should show halve the actual speed.

I would go for the MP only OC; you get the same clocks but the higher RAM speed.

I would like to see an in-depth article on MP vs MP + BCLK OC'ing, could be interesting!

Additionally I am also interested in reading an in-depth article on Skylake cache overclocking. When what comes into play / is beneficial, the exact numbers, etc. I haven't found one yet. Right now I have min 35, max 41.

I did find this (someone's findings):



I don't think that core equals cache clock is necessary perse, the benefits might be very few depending on the component speeds obviously and scenarios.


Right now I am very stable with this setup, note how the ram speed @2666 shows only half (should be normal like that I guess?) but the weird thing is how other folks show reports with full ram speed on cpuz:

http://1drv.ms/1ZbxDdS

I can go up to 4.5-4.6 with just MP, but on some benchmarks I get just a little better results with this config even the cpu speed being a bit lower. Can't notice a thing when using it for games but the benchmarks said it so yea...
 
Back
Top