i5-2500k - Multi LLC - Vcore - Wattage

Enthusiatic as I am as a member of ours proud OCF benching team, I'm also on the "green side".
I like to overclock and maximize energy as the same time.

I have a new processor and done toying around a bit. Too bad, it couldn't do any higher than 51x
So now it's going to be a 24//7 machine replacing my current i5-750.
Interestingly, while I was looking for an optimal setting for this chip and learning more about Multi Load Line Calibration functionality. I also found various energy consumption associated with it.

Previously, I've tested at various ranges of 4.40GHz~4.70GHz which I found to be the best clocks for i5-2500k and i7-2600k as far as speed, energy consumption and temperature ideally for my usage. I would like to note that every chip is difference. Your milage may vary.

===========
===========
===========

I'm using underload wPrime1024M and Kill-A-Watt EZ as a method to measure energy used for a whole system.
Passing wPrime1024M, Prime95 sFFTs for 20 minutes and Cinebench R11.5 considered stable of this testing.
(notice Prime95 pulls the highest wattage)

CPUz and CoreTemp, HWM Pro do not report voltages correctly. Gigabyte's Touch BIOS pc status monitor works fine.

SYSTEM hardwares with peripherals connected:
Win7 x64 - NO TWEAKS -Fully updated and loaded with programs including antivius security.
Gigabyte Z68X UD3H B3
8GB (2x4GB) G.Skill 1600MHz
Frio with dual Cooler Master PWM 120mm LED rifle bearing fans
(original Frio's fans removed due noisiness even at its lowest speed adjusted)
1- 92mm PWM fan for chipset cooling
1- Seagate 1.5TB Sata III / AHCI mode
1- DVD drive
1- USB multi-card reader
1- PCI Wifi
1- 18 LEDs light bar
(using IGP Intel's HD 3000 - without discreet GPU)

STEP 1:
I've choosen 4.5GHz as a final 24/7 clock as follows (so I thought...)

ENABLED 4 energy saving features (C1E, C3/C6, EIST and Intel Thermal Monitor)
DISABLED Multi Load Line Calibration
1.360 Vcore (this much needed to pass wPrime 1024M)
1.150 QPI/VTT
1.100 Graphic Core
1.815 CPU PLL
1.510 DRAM
888-24 1T Turbo Performance
28-27-29-29C IDLE 65 watts
54-56-57-56C LOAD 149 watts
It passed wPrime1024M with flying colors....but fail Prime95

STEP 1 1/2:
Level 10 LLC ENABLED......Pass and stable...but I forgot to take notes.

STEP 2:
Next, I've dropped vcore down to 1.330 but kept LLC at level 10 (other settings were kept the same)
Voila! It was still good and stable....but I've noticed it was hogging 191 watts on the register.
That's an extra 42 watts for level 10 LLC enabled with lower vcore was used. That's too much.

STEP 3:
Next, kept vcore at 1.330 and dropped LLC at level 5 (other settings were kept the same)
Voila! It was still good and stable. It now consumes only 169 watts which reduced by 22 watts.
It's in fact comsumes less energy with LLC reduction.

STEP 4:
Further reduction of vcore to 1.300 and kept level 5 for LLC = Pass everything OK
It registered 162 watts. That's only 7 watts less than STEP 3 above. (Prime95 sFFTs pulls 185 watts)
So, LLC plays a bigger role of energy consumption that I've notice here.
(Any lower vcore would fail. A combination of lower than 1.30 vcore with level 10 LLC did not help. BSOD)

OK, what's all this tell me?
1- Lower VCORE can be used if LLC was enabled. It helps with stability.
2- Lower VCORE setting does not neccessarily mean lower energy consumption if LLC was enabled.
2- Too much of higher level of LLC applied can be energy wasteful.

For now I'm enjoying a whisper quiet open air system at 4.5Ghz energy efficient.
Dual PWM fan on the Frio lugs around 1100~1200rpm surfing until duty calls.
92mm PWM fan lugs around 800~900rpm noiselessly.

=========
=========

Finalized settings for 4500MHz
45x
ENABLED 4 energy saving features (C1E, C3/C6, EIST and Intel Thermal Monitor)
ENABLED Multi Load Line Calibration at Level 5
1.300 Vcore
1.150 QPI/VTT
1.100 Graphic Core
1.815 CPU PLL
1.510 DRAM
888-24 1T Turbo Performance

wPrime 1024M
30-30-32-32 IDLE 65 watts
56-59-59-60 Peak - LOAD 162 watts

Prime95 sFFTs
29-29-31-31C IDLE 65 watts
61-65-66-66C Peak - LOAD 185 watts

You'll be amazed how much heat and energy consumed when comparing this with a Q6600 clocked at 3.80GHz.





==========
==========
==========

Also, I found this setting to be optimal for my chip at 4700MHz saved into user's profile setting just incase I ever needed with a few mouse flicks.
47x
ENABLED 4 energy saving features (C1E, C3/C6, EIST and Intel Thermal Monitor)
ENABLED Multi Load Line Calibration at Level 5
1.380 Vcore
1.270 QPI/VTT
1.130 Graphic Core
1.830 CPU PLL
1.510 DRAM
888-24 1T Standard Performance

wPrime 1024M
31-30-32-32C IDLE 68 watts
61-64-64-63C Peak - LOAD 185 watts

Prime95 sFFTs
31-29-31-31C IDLE 68 watts
67-72-72-72C Peak - LOAD 214 watts





===========
===========
===========








At the moment, I can not recommend this Gigabyte Z68X UD3H B3 motherboard to anyone.
It's having a problem with woken up from sleep mode to have the overclocked turbo mulitplier automacially resetted back to 34 or 35x as reported by a few members here on OCF and on Gigabyte's Forum as well.

LLC affects the loaded vcore, so you have to account for that. LLC doesn't directly affect power usage, but it's effect on vcore under load does.

Good work on the quest for a cool and quiet system that still flies! I love my little i3 550 system; fast as hell, but still super quiet w/ very little energy consumption!

jason4207 said:

LLC affects the loaded vcore, so you have to account for that. LLC doesn't directly affect power usage, but it's effect on vcore under load does.

Good work on the quest for a cool and quiet system that still flies! I love my little i3 550 system; fast as hell, but still super quiet w/ very little energy consumption!

Click to expand...

That's so true and thanks for pointing that out.

While LLC disabled, the load voltage dropped so much thus less energy usage. With not enough voltage being loaded, that caused Prime95 to crash as shown from STEP 1 OP above.

With LLC enabled, the load voltage didn't drop so much thus more energy consumed. Therefore, Prime95 did not crash with sufficient voltage being loaded.

4700MHz ___ IDLE (1.380vcore - LLC Enabled Level 5 = 1.390v shown on Giga's Touch Bios)


4700MHz ___ LOAD

Seriously? You got 4.7Ghz significantly below 1.3 load?
That's awesome dude.

Theocnoob said:

Seriously? You got 4.7Ghz significantly below 1.3 load?
That's awesome dude.

Click to expand...

No! As noticed on the 1st post, CPUz, Coretemp, HWM do not show voltages correctly as others have reported on Z68 boards and this board without exception.
I believe that it actually shows QPI/VTT voltage instead on CPUz.

Currently, Gigabyte's Touch Bios does show the correct voltage was set in the bios.

My air cooling has reached its limitation for hot a Summer's day at 4900MHz stable. Room ambient 85F
Not that I would use this clock as 24/7 but just for my curiousity about Multi LLC, Vcore and energy consumption relationship.

4900MHz
1.430 VCORE set in bios.
Level 7 LLC
1.310 QPI/VTT
1.840 PLL
1.115 Graphic Core
1.520 DRAM

I've gone through various tests from light to heavy threaded apps.
Idling voltage fluctuates between 1.44~1.45v

1) ConvertXtoDVD = 1.52v Load = 167 watts
Setting: 4 cores, Encoding SP (short projects), target DVD5 - 89 minutes file finished encoding under 10 minutes. A real world testing.
It looks like more voltage being shifted automatically to the heaviest thread while the other 3 cores were running cooler at slower speed.



2) wPrime 1024M = 1.49v Load = 230 watts



3) Cinebench R11.5 = 1.48v Load = 237 watts
(LOAD screenshot taken while it was at about 99.999999% finishing)



4) Prime95 sFFTs = 1.46v Load = 270 watts




5) LinX64 = 1.46v = 272 watts




Alright. What's going on here? The pattern shows more voltage being applied for lighter threaded apps.
Prime95 and LinX64 are the heaviest bench which used most wattage with 1.46v underload.
Cinebench R11.5 and wPrime 1024M demanded less energy but underload voltage surged to 1.48~1.49v.
Lastly, ConvertXtoDVD consumed the least energy but highest voltage of 1.52v was needed on demand for light thread.

Right now I don't know what to tell you. Can't nail the exact voltage that actually being used by the system.
It's telling me that VCORE set in bios does not mean much while Multi LLC was enabled.
Or it could be that Gigabyte's Touch Bios reported voltages were incorrectly as well.
Hopefully someone with more expertise could explains better.

====================

Energy consumption differences between 200MHz and 400Mhz...base on my specifications.
(Do notice that there's no high power discreet GPU being used)
85 watts difference based on Prime95 at 4.5GHz vs 4.9GHz --- or --- 56 watts difference for 4.7GHz vs 4.9GHz
68 watts difference based on wPrime at 4.5GHz vs 4.9GHz --- or --- 45 watts difference for 4.7GHz vs 4.9GHz

Great analysis!

Energy consumption is based on Watts which is Voltage X Current. P95 may drop the voltage a bit, but I'm sure the current is much higher which leads to a higher power consumption.

Very cool and interesting thread here SSM, I will have to play around with this when I get a chance as well!

Well, I've been thinking about the voltage being raised much higher for light threaded application.
That might raise a concern for bencher like me if inappropriate level of LLC was used for 2D benches. Which might led to RIP CPU due to the unknown hidden voltage being surged.

If anyone cares to know the results and how it effected with various level of LLC applied for light threaded app.
Setting: 4700MHz // 1.390vcore // 1.290 QPI/VTT

Level 3 LLC - SuperPI 32M LOAD = 1.38v
Level 4 LLC - SuperPI 32M LOAD = 1.40v (this level looks good but surely crashed with HEAVY threaded as it drops much further)
Level 5 LLC - SuperPI 32M LOAD = 1.44v
Level 6 LLC - SuperPI 32M LOAD = 1.46v
Level 7 LLC - SuperPI 32M LOAD = 1.51v

An example of Level 10 LLC - SuperPI 32M LOAD = 1.52v


The difference between Level 7 vs Level 10 isn't much.
Be cautious beyond level 6 LLC.

wow, that is a pretty serious increase with LCC from 0 to 10

I haven't tested a socket 1155 Gigabyte board or even a z68 chipset board... just a few other brands of p67 motherboards but I was wondering if you might be able to run some of your o.c. with less vccio (vtt/qpi)? I would think with DRAM running at 1600Mhz speed that your not stressing the IMC that much.

Since you are running with LLC at level 5 (somwhere ~ xx%?) wouldn't Prime small fft and and LinX place more stress/load (and hence more droop) on the cpu then Cinebench R11.5 and WPrime 1024M? What does your combo report for voltages when comparing benchs running with LLC at ~ 100% does it keep the load voltages more consistent?

Since you are looking for both energy savings and overclocking performance for a daily use system have tried doing some testing while using DVID vcore settings instead of fixed vcore settings?

Since you are looking for both energy savings and overclocking performance for a daily use system have tried doing some testing while using DVID vcore settings instead of fixed vcore settings?

Click to expand...

Tried that didn't work. It kept on failing during Windows loading. Overclock maybe too high for it to work?
Setting: NORMAL vcore // DVID: +.100~+.400

Since you are running with LLC at level 5 (somwhere ~ xx%?) wouldn't Prime small fft and and LinX place more stress/load (and hence more droop) on the cpu then Cinebench R11.5 and WPrime 1024M? What does your combo report for voltages when comparing benchs running with LLC at ~ 100% does it keep the load voltages more consistent?

Click to expand...

That's correct. Prime & LinX are the the highest stress/load caused most droop compared with others (see results of LOAD @ 4900MHz).
The underload voltages are very consistent as tested at 4.7 and 4.9GHz. Light threaded app always load higher vs lower for heavily threaded.
I'm not sure if I've answered to your question correctly.

I haven't tested a socket 1155 Gigabyte board or even a z68 chipset board... just a few other brands of p67 motherboards but I was wondering if you might be able to run some of your o.c. with less vccio (vtt/qpi)? I would think with DRAM running at 1600Mhz speed that your not stressing the IMC that much.

Click to expand...

Yes, lowered QPI/VTT work fine. I'm having a little issue and was experimenting with various QPI/VTT trying find the cause of a problem.
Survived/finished 55 minutes real world testing of transcoding from a 9.64GB .mp4 file to 3GB .mkv = LOAD 195 watts
4700MHz / 1.390 vcore / 1.150 QPI/VTT

SamSaveMax said:

Tried that didn't work. It kept on failing during Windows loading. Overclock maybe too high for it to work?
Setting: NORMAL vcore // DVID: +.100~+.400

Click to expand...

I'm surprised that it didn't work... I've been able to run all my SB setups with offset vcore for daily overclocks (crunching/100% load). I would think that your chip could do ~ 43x/44x at default DVID or with maybe just a "+" bump or two. I wonder what the VID is for your 2500K sample?

PolRoger said:

I'm surprised that it didn't work... I've been able to run all my SB setups with offset vcore for daily overclocks (crunching/100% load). I would think that your chip could do ~ 43x/44x at default DVID or with maybe just a "+" bump or two. I wonder what the VID is for your 2500K sample?

Click to expand...

Thanks for the tips.
I had to reset and start it from 43x, 45x and then 46x....it now works great.

My apology for stating Coretemp does not report voltage correctly.
I do not know why I've forgotten that it has always been reporting VID voltages.

================
================

Previously, have been known my stable LOAD voltages for 45x and 47x; it makes sense to keep load voltage inbetween for 4600MHz stability.

Setting:
NORMAL Vcore
+.060 DVID
1.150 QPI/VTT
1.815 PLL
1.510 DRAM

It now idles at very low voltage until duty calls.

4600MHz __ LOAD 1.34v - Prime95 sFFTs = 199 watts


4600MHz __ IDLE 0.82v - 65 watts

Hmm, for some reason when I use the offset voltage option (with no LLC) my voltage is much higher at idle than at load (as in 1.384v at idle/ 1.328v at load! ) Not sure why I'm seeing totally different behavior. I'm currently running these settings:

+0.050 offset (Vcore at 1.215, not really sure if that's stock, since I never ran it stock )
LLC at regular (0%)
VCCIO at 1.15v
PLL at 1.75v (read that it often gives greater stability and lower temps if you lower the PLL voltage, and I'm completely Prime stable)
VDRAM at 1.6125v

And all this is at a 45x multiplier. Any help for how I could lower that idle voltage would be awesome!

Left is at 100% Prime95 Blend load, right is at idle.

Do you have Speedstep and all the C-states turned on in the BIOS?

jason4207 said:

Do you have Speedstep and all the C-states turned on in the BIOS?

Click to expand...

Yes, EIST is enabled, and all the C-states are set to Auto (which is the stock config in my BIOS, should I see if I can enable them instead of auto?)

seth.i.simmons said:

Yes, EIST is enabled, and all the C-states are set to Auto (which is the stock config in my BIOS, should I see if I can enable them instead of auto?)

Click to expand...

You've Asus board. I don't know if it would make any difference but I've ENABLED all 4:
C1E, C3/C6, EIST and Intel Thermal Monitor.
Enabled instead of left them on AUTO....sure cut down some power comsumption the last time I checked.

===================
===================

OK, I got to have my 4700MHz back and did it.

One thing I noticed is that light threaded applications are no longer require high boost of voltage as seen previously under load.

Cinebench - 1.39v LOAD
wPrime 1024M - 1.39v LOAD
ConvertXtoDVD - 1.36v LOAD
SuperPI 32M - 1.33v LOAD


Prime95 sFFTs - 1.38v LOAD


Prime95 sFFTs - 0.86v IDLE <=====


I rest my case.
Thanks everyone!