Safe vcore for 6600K? Looking for tips!

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
it sounds like you have everything set right. Seems like the Gigabyte BIOS is a bit buggy when it comes to getting adaptive voltage working.

This is a bit of a long shot but worth trying. Try this specific sequence to see if this works. Boot to BIOS and put the settings above that worked in (44 on Auto) <save & exit>, let windows load. Verify freq and vcore. Shut down. Power on and boot to BIOS. Now just change AUTO to NORMAL <save & exit>, let windows load {but don't log in}. Shut down. Power on and let windows load. Hopefully, it will load.

While I was trying to get adaptive voltage to work, i had cases where BIOS gave me a pop-up stating settings weren't compatible (or similar) when I saved and exited so I had to re-enter BIOS and change vcore from NORMAL. Some cases it wouldn't boot at all. It was frustrating - but eventually it worked for me. Seemed the biggest issue was going directly from a manual vcore to NORMAL. The only things I did differently when I got it to work was (1) I went from AUTO to NORMAL instead of going from a fixed voltage to NORMAL (2) I Shut down as soon as windows booted after I made the change from AUTO to NORMAL.
 
Ok, so I did what you suggested, and I got windows to load with 44x and vCore set to normal. When I checked vCore at 44 set to auto, it was at 1.260. So, I then shut down and then changed multiplier to 47 and set DVID to +0.14. Everything booted fine, and vCore at idle was reported as minimum 0.036, but it kept bouncing around, generally sitting at about 0.8. However, when I began to run Aida 64, vCore jumped to 1.440, which is way higher than anything I've had since I started trying to oc. Temps didn't immediately jump nor did it crash, but I stopped the test anyhow.

So, I then changed DVID to 0.10, and also checked LLC, which was on auto. I changed that to high. Windows booted fine, vCore min 0.048, max 1.416, and during Aida 64, vCore stayed between 1.392 and 1.404, occasionally jumping to 1.416. Temps maxed out at mid 60s.

So it appears I have finally gotten what I was after, vCore to drop when cpu isn't at load. I may try to bring DVID down just a smidgen more and see where I can get to. I guess now the real question is which is worse, it running at almost 1.4 at idle, despite the core readings showing around 800mhz and DVID on the cores being around 0.8, or having the vCore jump so quickly when the cpu is loaded?

I should also add that I ran prime 95 briefly, and vCore never went above 1.356, and temps got into low 70s.
 
That is awesome. Congrats.

In your initial post you said you were stable at 4.7GHz with a Vcore of 1.290V. With your Vcore in NORMAL and DVID set to +0.10V you are a Vcore of approximately 1.4V. I wonder if you set DVID at +0.000V if you’ll get closer to your ideal 1.290V?

As to your question – I don’t think there is any issue with Vcore ramping up rapidly when the CPU goes from idle to a load. I think that is operating as it’s intended.

Did you happen to note any “magical” sequence that got adaptive voltage working for you? For me, I’m not sure if it was changing Vcore from AUTO to NORMAL instead of going from manual voltage to NORMAL, or if it was the complete shut down that got it working. This info may help the next person who has issues applying adaptive voltage on the gigabyte mobo.


 
Actually I was running 4.7 at manual 1.4. I actually had some issues with BF1 crashing after running normal and DVID at 0.10, so bumped it to 0.11. Has done fine on Aida 64 and prime 95 so far. I'm almost considering going back to fixed, due to the vCore jumping a bit over 1.4 in some instances. I'm going to play around with things a bit more and see where I'm at in terms of total stability.
As for sequence, I followed the sequence you recommended and it worked fine.
 
Ahh, I see.

I like running OCCT for stability testing and as a bonus it prints out all kinds of pretty charts when it’s done. One of the charts (below) plots vcore over the duration of the test so you can see if vcore jumped for just a bit – or if it was consistently high. Nice to see the trend over time.




 

The same thing happens to me. I can get "Normal" and +0.000 V working but only at stock 3.5 GHz (not great). The moment I put in the CPU Clock multiplier (46), even with absolutely nothing else changed, it no longer boots. I tried multiple combinations but even a complete stock reset (3.5 GHz) and "Relax OC" for the memory (also unchanged at stock 2133 MHz) offers no more booting. Just bad luck I suppose.

Here is the method I had in mind, wich I think is sound (the values are for illustrative purposes only of course) - I wrote it on the back of an envelope while sitting on the plane:
1. Go to the BIOS of your already stable overclocked CPU (e.g. 4.6 GHz OC and VCore value of 1.344 V)
2. Set CPU Vcore to “Normal”, the stock voltage recommended by Intel, based on the current settings
3. Keep “Dynamic Vcore(DVID)” (which lights up just under) unchanged at its default value of +0.000V (just enter 0.000)
4. Save (F10) then completely shut down (not just a restart) to make sure that “Normal” is applied to VCore
5. Back in Windows, read the “Normal” VCore value (e.g. 1.368) via CPUID HWMonitor and/or CPU-Z
6. Go back to the BIOS
7. To keep your existing OC running at 1.344 V, offset the Dynamic Vcore(DVID) by the right amount:
8. - Lower “Dynamic Vcore(DVID)” using an offset of -0.024 to reach 1.344 V (= 1.368 V – 0.024 V)
9. - Save (F10) then completely shut down (not just a restart) to make sure that the offset is fully applied
10. Back in Windows you will still be at 4.6 GHz, but with VCore “idle” at 0.804 V and VCore “load” at 1.344 V

I know that LLC (Load Line Calibration) can also be disabled if it interferes, but in my case it makes no difference - no boot.

What do you think with your "hands-on" experience with DVID Offset? Shouldn't this method work? Am I missing something?
 

philipew - while I don't think we have found a definite answer, there is one additional step that I recommend adding which may or may not be contributing to my success as well as ZEBuckeye81's. (basically going from AUTO to NORMAL instead of going from a manual vcore to NORMAL.)

1. Go to the BIOS of your already stable overclocked CPU (e.g. 4.6 GHz OC and VCore value of 1.344 V) and set Vcore to AUTO.

1a. Save (F10) then completely shut down (not just a restart) to make sure that “AUTO” is applied to VCore
 


The steps you listed are essentially what I did, and it seemed to be good to go initially, vCore set top normal, DVID set to +0.110, clock at 47. Results in hwinfo are as I listed above. Then, however then I decided yesterday afternoon to play some BF1. I started getting various crashes and bsod. Prior to that I had run various stress tests like I normally do, for just a shirt while each, with zero issue.

Now I did look in event viewer and found the error to be possibly related to a bad network driver, so I went through and reinstalled the driver, however I still had issues. Don't know if this is due to the overclock order not. I plan to tweak things a bit more tomorrow. FWIW, I had all settings the same as when I had 4.7 stable @1.4 vCore, 3.9 uncore on auto, all cstate settings on auto, and xmp profile 1, which for my memory is 3200.

Oh, and I switched LLC to high, to prevent such a drastic sudden increase in voltage to the cpu. Not sure if this is the correct line of thinking or not. I'm actually torn on whether to mess with using DVID, or just let my manual settings ride since core clocks were dropping under idle, as well as DVID voltages, just not vCore. I read somewhere that voltage and current both play a factor? And that even though vCore is at a higher level, it's not under load so not stressing the chip? I don't know, as this is my first go around at overclocking and am learning as I go.

I thought I had found my sweet spot, with vCore dropping quite low when the cpu wasn't over load. In fact, my idle temps were even dropping into the teens as opposed to with manual voltage where my idle temps were in the upper 20s.
 
Thanks for this info. Would like us to figure out this adaptive voltage.

With regards to LLC. LLC is to prevent rapid DROPS in voltage during a rapid increase in load. Not sure this can be monitored with accuracy when you have adaptive voltage implemented. That is why I recommend finding an appropriate level of LLC while you have a manual Vcore set (as well as turbo,and all enegy savers disabled; EIST/C-states). For me I took my stable OC with my Vcore manually set to 1.30V in BIOS. Then while monitoring Vcore in CPU-Z – I watched it at idle and it was 1.284, but as soon as I started OCCT Vcore would drop. When I enabled LLC the voltage would remain the same when idle as well as under load. Note: the reason I say I don’t think you can monitor when using adaptive voltage is that when you have this applied you want vcore to drop at idle and ramp up under load so there is no way of knowing if you have the proper amount of LLC. For me I only have AUTO, NORMAL, or HIGH settings for LLC but others have many levels of LLC. You want just enough LLC so that you don’t see a big drop in Vcore between idle and load (Vdroop) but you also don’t see a big increase in Vcore between idle and load (Vboost). You can google more info on LLC but this reddit thread has a good synopsis: https://www.reddit.com/r/overclocking/comments/28qlv4/guide_load_line_calibration_what_is_it_and_how/

Question: Above you state “I'm actually torn on whether to mess with using DVID, or just let my manual settings ride since core clocks were dropping under idle, as well as DVID voltages, just not vCore”. Where or how do you monitor DVID to know that was dropping under idle.

If I were you I think I would leave all my setting exactly how they are – but put Vcore at 1.40V instead of NORMAL, then play BF1 for a while to see if you BSOD. I would want to see if it is related to the adaptive voltage or if there is some other issue. Oh and note what Vcore/Freqs/Temps are doing with a manual Vcore as well as in adaptive, both at idle as well as when playing BF1 or stability testing. Keeping track of this in both settings as well as at idle and under load may help with figuring this out.

 
So at 47 clock on 1.4 manual vCore I had zero issues, stable stress testing and gaming.

Today I took a couple steps back, to the settings I had prior to overclocking. Stock clock, stock cache, auto vCore and ram on xmp 1 @3200. No stability issues, no crashes. So, I proceeded to set clock to 46, cache at 39, vCore normal, DVID +0.08. So, everything seemed fine, stable stress testing and in BF1. Plus, vCore dropped appropriately when cpu not loaded, and when loaded (stress testing) vCore stayed below 1.4 for the most part. So, I'm going to see if it stays stable and also will probably take DVID down a little further and see what happens.

So, I may have finally found my sweet spot. Yes, I can run 4.7, but I'm not sure how I feel about having to use manual voltage of 1.4 to do so for the slight gains. I may try to tweak bclk a little to see if that will give me a little more, but if not, I'll be happy to have a good stable overclock at 4.6.
 

Well, you could contact Gigabyte support with a so-called Product "information request”.... expect a few weeks for a reply, if at all ;-(.
 

I had the same, 4.7 GHz at 1.41V but with Uncore at 4.7 too (47/47), LLC (1.404V/1.416V), 16GB mem at 3200 (15 17 17 28 1T), and all C-states disabled. Great performance, unfortunately my single fan (budget) 120mm CPU cooler can't handle the high temp. My sweet spot is 100 MHz lower, 4.6 GHz (46/46). The jump in temp when going from 4.6 to 4.7 GHz is extremely high (shows the CPU's limits IMO). Better safe than sorry.
 
Yes, I definitely feel the same way Phil. 100 mhz isn't really a significant enough gain to warrant the extra voltage and heat.

Hm,you're able to run your cache at 46 also? Interesting, I'll have to play with mine. I hadn't changed it since backing off my core multiplier. May be able to increase it a bit.
 
Nice job!

yea, 4.6 is where mine likes to be as well. I had thought about tweaking BCLK a bit, but at this point the very small gains don’t really seem worth it. Going from 3.5/3.9 to 4.6 is a good sized gain. 4.6 to 4.7, not so much.


 

Here are my current, and ultra stable, BIOS settings for i5 6600K on Gigabyte GA-Z170XP-SLI:
[1] - CPU Core Ratio: 46
[2] - FCLK Frequency For Early Power: 1 GHz
[3] - Uncore Ratio: 46
[4] - CPU Flex Override: Disabled
[5] - Intel Turbo Boost Technology: Disabled
[6] - CPU Enhanced Halt (C1E): Disabled
[7] - C3 State Support: Disabled
[8] - C6/C7 State Support: Disabled
[9] - C8 State Support: Disabled
[10] - CPU Thermal Monitor: Enabled
[11]-CPU EIST Function: Enabled
[12]-Voltage Optimization: Enabled
[13]-Residency State Registration (RSR): Disabled
[14]-Hardware Prefetcher: Enabled
[15]-Adjacent Cache Line Prefetch: Enabled
[16]-Extreme Memory Profile (X.M.P.): Profile 1
[17]-System Memory Multiplier: 32
[18]-Memory Enhancement Settings: Relax OC
[19]-Channel Interleaving: Enabled
[20]-Rank Interleaving: Enabled
[21]-CAS Latency: 15
[22]-tRCD: 17
[23]-tRP: 17
[24]-tRAS: 28
[25]-Command Rate (tCMD): 1
[26]-CPU VCore Loadline Calibration (LLC): High
[27]-CPU VCore: 1.355V
[28]-CPU VCCIO: Normal
[29]-CPU System Agent Voltage: Normal
[30]-PCH Core: Normal
[31]-DRAM Voltage (CH A/B): 1.36 V
[32]-Internal Graphics: Disabled

Have a look here for the fans: https://postimg.org/image/8d0q9fv63/

Here is a revision
I noticed that I could push my CPU Core frequency from 4.6 GHz to 4.608 GHz (with Bclk = 128 MHz) without increasing VCore (no longer stable at 4610 MHz on VCore of 1.355 V for my chip), therefore without increasing temp. However, Uncore must be limited to 3968 MHz for stability (anything higher is not stable on my own system), and the idle frequency of the CPU goes from 800 MHz to 1024 MHz although my CPU usually spends very little time below 1024 MHz anyway. So it has no impact on temp. Interestingly, note that for the memory, 128 x 25 = 3200 exactly (this is important).

These settings give me up to 6 points more on the Cinebench R15 benchmark test and this is really "for jam" as VCore is unchanged. It is also stable (I tested on Prime95) with Fclk at x10 (1280 MHz) but I prefer to leave it low, although that could potentially provide additional (very slight) performance as shown in tests here: http://www.tweaktown.com/guides/7481/tweaktowns-ultimat...

Another advantage of a higher Bclk value, in this case of 128 MHz, that I also observed is the resulting CPU (Core) clock frequency while running (e.g. 4.598, 4.599, 4.600, 4.601) does not fluctuate as often (i.e. is more stable). The slightest fluctuation in Bclk frequency, e.g. 99.5, 100, 100.5, 100.2... GHz is multiplied by 46, if the multiplier is 46. As a higher Bclk value of 128 MHz reduces the multiplier from 46 to 36, the effect of the fluctuation of Bclk also seems to be less amplified (geared) which leads to a more stable resulting Core frequency, in this case centered around 4.608 GHz (128 x 36). And a more stable Core freq. works in favour of a more stable system altogether, with more consistent performance over time. However, this may not work favourably for everyone, just something to experiment with.

Summary
-----------
The only settings I personally applied manually in the BIOS of my INTEL i5 6600K are:

[1] - CPU Core Ratio:------> 46 = 4600 (Bclk 100) 36 = 4608 (Bclk 128)
[2] - CPU Base Clock (Bclk):------> 100 - 128 MHz
[3] - FCLK Frequency For Early Power:------> 1 GHz (x10) (Bclk 100) Normal (x8) (Bclk 128) <-------
[4] - Uncore Ratio:------> 46 = 4600 (Bclk 100) 31 = 3968 (Bclk 128)
[5] - CPU Flex Override:------> Disabled <--- 1/8
[6] - Intel Turbo Boost Technology:------> Disabled <--- 2/8
[7] - CPU Enhanced Halt (C1E):------> Disabled <--- 3/8
[8] - C3 State Support:------> Disabled <--- 4/8
[9 ]- C6/C7 State Support:------> Disabled <--- 5/8
[10]-C8 State Support:------> Disabled <--- 6/8
[11]-CPU Thermal Monitor:------> Enabled <----------- 1/7
[12]-CPU EIST Function:------> Enabled <----------- 2/7
[13]-Voltage Optimization:------> Enabled <----------- 3/7
[14]-Residency State Registration (RSR):------> Disabled <--- 7/8
[15]-Hardware Prefetcher:------> Enabled <----------- 4/7
[16]-Adjacent Cache Line Prefetch:------> Enabled <----------- 5/7
[17]-Extreme Memory Profile (X.M.P.):------> Profile 1
[18]-System Memory Multiplier:------> 32 = 3200 (Bclk 100) 25 = 3200 (Bclk 128)
[19]-Memory Enhancement Settings:------> Relax OC
[20]-Channel Interleaving:------> Enabled <----------- 6/7
[21]-Rank Interleaving:------> Enabled <----------- 7/7
[22]-CAS Latency:------> 15
[23]-tRCD:------> 17
[24]-tRP:------> 17
[25]-tRAS:------> 28
[26]-Command Rate (tCMD):------> 1
[27]-CPU VCore Loadline Calibration (LLC):------> High
[28]-CPU VCore:------> 1.355 V
[29]-CPU VCCIO:------> Normal <-------
[30]-CPU System Agent Voltage:------> Normal <-------
[31]-PCH Core:------> Normal <-------
[32]-DRAM Voltage (CH A/B):------> 1.360 V
[33]-Internal Graphics:------> Disabled <--- 8/8
================================================================================
Note that even at 4.6 GHz, using an Uncore (cache) multiplier of 39 instead of 46 (like the Core) is better for temp. (-5 C) and it only very (very) slightly affects performance (hard to detect even).

My road to 4.7 GHz is as follows:

Volt:-----1.344 -- [+12mV] -- 1.356 -- [+12mV] -- 1.368 -- [+12mV] -- 1.380 -- [+12mV] -- 1.392 -- [+12mV] -- 1.404
GHz:-----4.600------------------4.620------------------4.640-------------------4.660------------------4.680-------------------4.700
Vcore:---1.355------------------------------------------------------[+60 mV]-------------------------------------------------------1.41
 

TRENDING THREADS