Overclocking 5820K. A personal experience.

[avatar_raq]

Hello everyone. This is my first time with a Haswell-E CPU and I am almost done overclocking my 5820K. I thought I share my personal experience and settings so others may give me feedback and perhaps benefit from my steps and mistakes (if any), as there is lack of overclocking guides that suit my goal, that is to achieve a stable 4Ghz OC for 24/7 operation with as minimal increase in voltages as possible, to maximize the life span of the CPU while maintaining quiet operation and decent performance.

First of all my specs:
Intel Core i7 5820K
Asus X99 Deluxe U3.1 (BIOS 3004)
Corsair LPX 2666 mhz C16 4x8Gb
Corsair H110i GTX
Corsair Graphite 760T
Corsair RM 1000i

To keep the AIO CPU water cooler running quietly I connected the stock fans to the CPU block, the way Corsair recommends, and used the Corsair Link software to keep the fans at ~800 rpm most of the time by setting a custom fan profile, I also set the pump to quiet mode. (P.S I mounted the CPU cooler at the top panel and set its 2 fans as exhaust as I believe it achieves the best overall cooling of the case and the hardware inside it.)

I did not use the RAM's XMP profile as it sets the CPU strap and Bclock to 125 which I do not like. Instead I dialed in the nominal clock and timings (2666, 16, 18, 18, 36) and left the voltage to auto as it was correct (1.2v).

Using CPU-Z I found that the actual Bclock was 99.9 if set to auto or 100 in the bios so I changed it to 100.1 to have nice rounded clock numbers on the CPU and RAM. I left speed step technology and turbo mode enabled in bios, I also changed the minimum CPU power state setting in windows to 1% to ensure the CPU clocks down in idle. I changed the maximum CPU multiplier to 40 in bios and the Core voltage to adaptive mode (+0.001 as it is the minimum) and set the turbo boost voltage to 1.000, as I read in another thread @cadaveca recommends this when the Bclock is not overclocked, which makes sense. I left the cache and CPU input voltages to auto as I did not OC the cache, I disabled LLC (by setting it to level 1) to ensure minimal input voltage spikes that may shorten the life span of the CPU. I also set system agent voltage offset to (+0.001) to ensure the mobo does not increase the voltage if left auto.

Using ASUS AI suite and CPU-Z the voltages are:
CPU Core 0.769 v (Idle, multiplier x12) - 1.037 (Load, multiplier x40)
Cache 0.983 v
System agent 0.816 v
CPU Input 1.760 v

I used LinX, ASUS Realbench and prime95 (latest versions) to test for stability and temperatures. I read a lot of forums and guides (including ASUS's official ROG guide) recommending against using the new prime95 as it can harm the CPU when highly overclocked but I used it anyway as I found it was the fastest test to detect instabilities and the one which produced the highest temps, besides my OC and voltages are far from high or extreme, so it should be safe. Maximum temps with prime95 were 67 on the cores, 72 on CPU package and 70 on VRM, which I believe to be good considering how quiet the case is. Other stress tests did not come near these temps. At idle the temps are 35, 38, 40 respectively.

In spite of all my efforts, the CPU clock goes up and down even when sitting in windows doing nothing but browsing and it never stays at the x12 multiplier for more than few seconds. I think this is normal behavior but I wish I can change this.

So what do you think? Did I achieve my goal? Should I change anything? Spend more time fine-tuning things? Can any of the voltages be lowered further without affecting the stability? What would you do differently?

 

[basco]

every cpu is different but i think with 1,037volt it should be lower then 67° - but it will not harm anything.
maybe too much thermal paste or too tight screws?

and llc level 1 is minimum\lowest? normally i have 6 or 7
you can test this if your voltage goes down under load

ps: nice clean build !

 

[avatar_raq]

every cpu is different but i think with 1,037volt it should be lower then 67° - but it will not harm anything.
maybe too much thermal paste or too tight screws?

The summer is hot here but I did use the thermal paste which came pre-applied on the AIO cooler and it is thick. I may try cleaning it and applying a thin film of a Zalman paste I have left and see if it helps temps, but honestly I am satisfied as these temps are only achieved with prime95 small FTTs test.

and llc level 1 is minimum\lowest? normally i have 6 or 7
you can test this if your voltage goes down under load

Yes level one in my bios is stated as zero LLC. CPU-Z and AI suite reads the V core as 1.037 v with x40 multiplier and it sometimes drops under full load to 1.27 or 1.22 v. So is this the default Vdroop?

ps: nice clean build !

Thanks. The picture was taken at day one, I later did some more cable management!

 

[basco]

ok yeah outside temps will play in-so like you say only during prime is ok.

yes this should be default vdroop- i am using the higher ones since a long time- so just set it how you like it.

 

[cdawall]

Temps seems extremely high? I was pulling 70's at 4.5 on my H100i with push/pull corsair SP fans. Currently running 4.7 on a custom loop which is just a single 240mm rad

 

[avatar_raq]

Temps seems extremely high? I was pulling 70's at 4.5 on my H100i with push/pull corsair SP fans. Currently running 4.7 on a custom loop which is just a single 240mm rad

Hmmm. All fans (except the 2 front intake ones) are running near their lowest RPM, so does the pump. This is how I explain the higher than expected temps. But now I am tempted to reapply thermal paste and see if it helps.

 

[cdawall]

Hmmm. All fans (except the 2 front intake ones) are running near their lowest RPM, so does the pump. This is how I explain the higher than expected temps. But now I am tempted to reapply thermal paste and see if it helps.

I wouldn't turn down the pump RPM...

 

[RejZoR]

You can use XMP with 100 CPU strap. Just set it manually afterwards... CPU voltage, just gradually decrease it. Use ASUS RealBench H.264 test and run it in a loop. It's using AVX instruction and is from my experience best for revealing instabilities. Even if system doesn't crash, it will show encoding errors. This means it's sort of stable, but you need to bump up voltage a bit. At 4 GHz, I'd just use 1.0 V directly and see how it goes. Should go fairly easily till 4.5 GHz. Above that it's all luck.

 

[xkm1948]

Check if you can POST with 1.3V at 4.5GHz. If yes try to fine tune it for stability. If not up the vcore or down the OC.

HWE overclocks really well comparing with BWE. So make the best use out of it.

Also I would highly recommend you read the ASUS HWE overclocking guide. You need to tune up your cache frequency as well.

 

[avatar_raq]

I wouldn't turn down the pump RPM...

Why? I am using their own software, so it is not operating out of spec.

You can use XMP with 100 CPU strap. Just set it manually afterwards... CPU voltage, just gradually decrease it. Use ASUS RealBench H.264 test and run it in a loop. It's using AVX instruction and is from my experience best for revealing instabilities. Even if system doesn't crash, it will show encoding errors. This means it's sort of stable, but you need to bump up voltage a bit. At 4 GHz, I'd just use 1.0 V directly and see how it goes. Should go fairly easily till 4.5 GHz. Above that it's all luck.

As I mentioned, the Turbo V core was set to 1.000 volt in the bios with an adaptive offset of 0.001 but in reality it registered arounf 1.47 v. I tried decreasing the setting to~ 0.970 v in bios to hit 1 v actual, but it was not stable.

Check if you can POST with 1.3V at 4.5GHz. If yes try to fine tune it for stability. If not up the vcore or down the OC.

HWE overclocks really well comparing with BWE. So make the best use out of it.

Also I would highly recommend you read the ASUS HWE overclocking guide. You need to tune up your cache frequency as well.

As I said I do not want to push the CPU as high as I can, I want a reasonable OC to last some 5 to 6 years as my last system (x58) did, I cannot afford a new system every year or two. I did read ASUS guide as well as others, and the general consensus is that cache clock is only useful in certain synthetic metrics and has no substantial effect in gaming and my everyday's tasks. This is the same reason I am keeping the RAM at its rated clock and timings for now as the performance benefits are not worth the headache.

 

[xkm1948]

As I said I do not want to push the CPU as high as I can, I want a reasonable OC to last some 5 to 6 years as my last system (x58) did, I cannot afford a new system every year or two. I did read ASUS guide as well as others, and the general consensus is that cache clock is only useful in certain synthetic metrics and has no substantial effect in gaming and my everyday's tasks. This is the same reason I am keeping the RAM at its rated clock and timings for now as the performance benefits are not worth the headache.

I plan on to use my current X99 system for a while as well. HWE can easily handle 4.2~4.3GHz. Anything onward you will be playing with Silicon lottery.

Moderate cache OC does have some effect. For example I use ~3.5GHz Cache. This will avoid getting cache degradation.

It is about finding your rig's optimal spot. There is limit pushing OC and moderate optional OC. I believe your system can be fine tuned a little bit more for better performance without sacrificing longevity.

 

[avatar_raq]

I plan on to use my current X99 system for a while as well. HWE can easily handle 4.2~4.3GHz. Anything onward you will be playing with Silicon lottery.....It is about finding your rig's optimal spot. There is limit pushing OC and moderate optional OC. I believe your system can be fine tuned a little bit more for better performance without sacrificing longevity.

Hmmm. I am sold. I will try to see the max OC I can get with an actual 1.1 v core voltage when I have the time. I will share my findings here.

 

[basco]

if its nearly 40° outside temp in your country then i think loadcputemp is ok.

 

[avatar_raq]

if its nearly 40° outside temp in your country then i think loadcputemp is ok.

Right now it is 44°C outside but the room I am sitting in is ~22-25°. Once my demciflex dust filters kit reach me, I am gonna add a bottom fan, increase the RPM of my top fans (CPU cooler fans) and try to optimize the air flow for a positive pressure inside the case and see if this helps.

 

[avatar_raq]

I found a solution for the jumping multiplier in another forum. Someone was having the same issue and found the culprit!

I figured this out, can't believe what it was that was causing it. It seems if I use a custom power plan in windows that starts from the "High Performance" template it causes the CPU to do this. If I switch the power plan to the Balanced setting, it works fine. If I make my own custom power plan from the balanced template, it also works fine. Seems odd because I use the same settings for the CPU in each plan but it must be some glitch with the power plans

I tested it and it works. Now my multiplier sits at 12 when idle.

 

[RejZoR]

You should update your profile specs, they are a bit outdated now...

 

[avatar_raq]

You should update your profile specs, they are a bit outdated now...

Done.

I believe your system can be fine tuned a little bit more for better performance without sacrificing longevity.

I tried fine tuning my overclock a bit and the initial results are not encouraging. My chip looks to be a bad overclocker or I am just missing something. I will post updates when I have solid proof.

 

[RejZoR]

4GHz at 1.0 V is far from bad considering 5820K runs 1.050 V at stock 3.6 GHz turbo. I'm running mine 1.175V at 4.5 GHz. 1.37 V you use for 4GHz sounds a lot. Do you really need such voltage?

CPU Cache degrades the fastest out of everything so keep that at stock voltage and clocks. The gain is really not needed if you ask me because this platform already has tons of bandwidth on all ends. Keep it safe for now and utilize it when CPU will start losing steam at the end of its life cycle and you boost it then. That's how I decided. I'm keeping the standard overclock though

 

[avatar_raq]

4GHz at 1.0 V is far from bad considering 5820K runs 1.050 V at stock 3.6 GHz turbo. I'm running mine 1.175V at 4.5 GHz. 1.37 V you use for 4GHz sounds a lot. Do you really need such voltage?

CPU Cache degrades the fastest out of everything so keep that at stock voltage and clocks. The gain is really not needed if you ask me because this platform already has tons of bandwidth on all ends. Keep it safe for now and utilize it when CPU will start losing steam at the end of its life cycle and you boost it then. That's how I decided. I'm keeping the standard overclock though

I set vcore to adaptive voltage (+0.01) and turbo voltage (0.99) to make the total 1.0v. The actual Vcore read by CPU-Z however is 1.037 @ 4Ghz which drops to ~ 1.024v under full prime95 load. The actual vcore reading is higher than what I set in bios in spite of disabling LLC which is weird, although I read somewhere that LLC only affects input voltage on this platform. I tried lowering the bios value to 0.7v to achieve an actual 1.0v but that proved to be unstable under prime95. So again either I have an unlucky chip or it is my mistake that I am sticking to prime95 or Linx which are rarely recommended by Haswell-e overclocking guides. Using ASUS realbench yields far better results and I can pass it with higher OCs @ lower voltages.

Edit: Corrected the voltages which caused some confusion to @RejZoR. It is 1.037 instead of 1.37! Sorry!

 

[RejZoR]

For me, not much difference if I set offset to 0.001V or leave it on Auto. Idle voltage is the same and turbo voltage is defined as maximum anyway for what you set it. I've used 1.0 V for 4GHz and in both cases idle and load voltages were basically the same with tiny deviation.

 

[avatar_raq]

For me, not much difference if I set offset to 0.001V or leave it on Auto. Idle voltage is the same and turbo voltage is defined as maximum anyway for what you set it. I've used 1.0 V for 4GHz and in both cases idle and load voltages were basically the same with tiny deviation.

I just use the lowest value offset of 0.001 to ensure no more voltage is added at multipliers lower than 33. I minimize the use of auto as my past experience with ASUS boards, they tend to use high voltages under auto.

 

[avatar_raq]

I edited my previous post and system specs to correct the voltage values. I do not know what was wrong with me at the time I wrote them. Oh wait I was fasting!

 

[RejZoR]

People seem to still think AUTO mode applies too much voltage from past experience 10 or 15 years ago. From what I see it's nothing like that anymore. Besides, in most cases it's either full load or no load. My voltage fluctuates between 0.760 and 1.032 in both cases when using mild load of 2 threads using Ultra preset in 7zip. Clock jumping between 3.2 and 4 GHz.

 

[basco]

my chip needs 0,050volt moar with adaptive as to fixed volts

and keep on going just 1 more week fasting

 

[heky]

@RejZoR
5820K at 4.5Ghz with 1.17v under full load? Pics or it didnt happen!