4 Attachment(s)
Gigabyte X58 3 LEVEL LLC EXAMINED!!!
Intro: To provide longevity of the processor Intel has incorporated two special mechanisms that have been around in their processors for a nice lengthy period, VDrop and VDroop. Vdrop is when the processor’s voltage is dropped from bios setting or stock setting to a voltage a few notches down, vdroop on the other hand is when the processor voltage drops below idle vcore under conditions of load. For us overclockers this can be a peril. In a time long long ago in place not so far away we still overclocked, and this vdroop was a big issue for us, under load our processors would become unstable, the solution was a voltage modification of a variable resistor (pontimeter) tweaked to a certain resistance and soldered directly to the power control IC. I myself have done so many voltage modifications that attacking such a task is no problem, but for many Overclockers this problem is one without a viable solution. The solution used to void the warranty of the motherboard or worse make the motherboard void. Motherboard manufacturers in the time of Q9xx processors came up with a solution called Load Line Calibration. In effect it made it possible to control the droop in voltage and in some cases provide the processor more voltage than when at idle, successfully reversing the effect of vdroop. This technology doesn’t come without its consequences; LLC (Load Line Calibration) is not a perfect fix. The problem in large with LLC is that when coming from heavy load situation to idle the vcore will spike above set voltage, the thing vdrop and vdroop compensate for. It is indeed not healthy for the processor, but does give us an advantage. Not only is voltage spiking an issue but so is temperature increase because of the raise in voltage to the processor. For those of us that know how much difference .05v can make in a processor temperature, it’s a give and take situation.
The Gigabyte X58A series of motherboard have 3 levels of Load level Calibration control. The first level is standard manufacturer (Intel) operation, no load line calibration. In standard mode the motherboard does not try to eliminate vdrop nor vdroop. Level 1 is slight vdroop control as stated in the motherboard manual as well as in the bios, and finally we have Level 2 which is moderate vdroop control. You are probably wondering why I am writing this, the reason is because no one has tested and wrote about the three levels of LLC on Gigabyte motherboards. All other motherboards to my knowledge just have one level of LLC, on and off. Here in forth I am going to test the voltage through a windows program, OCCT, a popular stress testing program that also has a data recorder. I will then do the following:
1. OCCT makes it own graph, but its graph has its own problems as does the software, first of all software is very inaccurate for reading analog voltages, drop outs of voltage are common and persist in software that does not actually occur in real life. Also the spikes related to LLC happen in a fraction of a second, and OCCT can only record every second, so those spikes do not show up.
2. I will formulate my own graphs, simple scatter plots with simple lines. One graph for each run, 1 hour per run, 3601 data points.
3. I will go through every set of data and pick out the outliers, not the spike, but the drop outs, anything under .1 volt difference gets tossed, but I will post them so you know exactly what voltage readings are omitted.
4. I have researched and measured for myself the voltage on the motherboard with the digital multi meter (DMM), my finding are like many others, that the voltage readout in windows is within <5% of actual voltage on the gigabyte x58 boards. Other boards i have tested do not have this accuracy, most being Asus. So for this test I will use software to find trends among data points.
5. Minimum, Maximum, Average, and Median voltages will be reported for each run.
6. A graph of all runs will be put together, as well as a conclusion.
7. Bios voltage set is 1.35v for all runs, QPI/VTT voltage is set to 1.30, VRMs are consistently cooled. CPU frequency is at 4011mhz (21x191) to emulate overclocking, voltage is set at 1.35 (Intel’s stated safe maximum) to show overclocked voltage. Actual voltage needed for 4ghz may vary, I bumped it up to 1.35 to show maximum voltage moderate overclockers may use.
8. Voltages are rounded to 3rd – 4th significant figure.
9. The test gives 10 minutes idle time, then starts test, then goes back to idle 4 minutes (56 minutes in) before stop of test.
TEST RUN Components:
X58A-UD5 Rev 2.0 BIOS: GOOC 2010
I7 930 D0 Batch: 3002A608 21x191=4.011ghz-1.35v set in bios.
Corsair XMS3 2gbx2 1600mhz 9,9,9,24.
GT 220
Antec 500 watt Earthwatts, for motherboard and cpu only, I have a secondary PSU for everything else. Voltages on 12, 15v, and 3.3v rails have been measured and all were in 3% fluctuation from idle to load.
Custom Water loop DangerDen TDX (lapped)
Crucial C300 64gb/WD 150 Velociraptor
Same programs are run in the background for every run, nothing but OCCT running and computer was not accessed during the duration of each test.
Run 1 NO LLC:
Max: 1.312
Min:1.28
Average: 1.283
Median:1.28
Vdrop:0.038
Vdroop:0.032
Total Vdrop+VDroop:0.07
Omitted data points (sec/voltage): (4-5/.160),(65-68/.496),(1867-1871/.560),(3445-3446/.160),(3493/.160)
General trend as follows: voltage idle is 1.312v volts, under load it drops to 1.280v, during load the voltage spikes to 1.296 many times, the spikes are as follows in seconds: 6,94-99,129,192-196,418-421,442-450,475-478,676-680,684,705-708,745-746,764-769,958-959,985-986,1007-1009,1022-1025,1256-1261,1301-1303,1354-1358,1556-1571,1582-1585,1653-1656,1919-1922,1951-1959,2321-2323,2588-2594,2706-2708,2880,2934-2938,3175,3202-3203,3226-3230. 1.280v load continues. Idle voltage reappears at 3363 seconds, approximately 56 minutes.
http://www.xtremesystems.org/forums/...1&d=1287277877
Run 2, LLC LEVEL 1:
Omitted data points (sec/voltage):(28/0.16),(3447,.176)
Max: 1.328
Min:1.312
Average:1.314
Median:1.312
Vdrop:0.022
Vdroop:0.016
Total Vdrop+VDroop:0.038
General trend as follows: Idle voltage of 1.328v , under load drops to 1.312, voltage spikes to 1.328 (exact idle voltage) many times as follows in seconds: 126-131, 200-203,416-421,473-475,990-994,1053-1057,1279-1280,2267,2286,2509,2548-2551,2845-2849,3195-3196,3271-3274. 1.312 load voltage ends at 3363 seconds and 1.328 continues on till the end.
http://www.xtremesystems.org/forums/...1&d=1287277877
Run 3, LLC LEVEL 2:
Max: 1.344
Min:1.328
Average: 1.343
Median: 1.344
Vdrop:0.022
Vdroop:+0.016
Total Vdrop+VDroop:0.006
Omitted data points (sec/voltage):(3476,1.112)
General Trend as follows: 1.328v start idle to 1.344v load at 64sec-3363sec, then 1.328 commences again.
http://www.xtremesystems.org/forums/...1&d=1287277877
ALL TOGETHER:
http://www.xtremesystems.org/forums/...&d=12872778771
Conclusion: As expected the average voltage increases from no LLC to Full LLC in a semi linear fashion. Looking at the graphical analysis we can deduce that the LLC settings do indeed vary from one another. For starters the standard level shows a (1.35-1.312) 0.038v vdrop from bios setting to actual voltage given. Vdroop of 0.032v slightly less than vdrop, compared to what it used to be, this voltage drop isn’t that bad, but for current systems the drop and droops are significant, in total under load voltage is 0.07 less than set in bios, but only about half that that is droop. Moving on to LLC Level 1 vdrop is (1.35-1.328) 0.022, less than standard vdrop, but still significant, under load there is another .016v droop, this is almost half that of standard level. LLC level 1 comes to a combined 0.038 voltage drop. This is slightly more than half the total drop without LLC, a very significant improvement. Moving on to the most aggressive level of LLC control Level 2, we have a starting 1.328 idle voltage (1.35-1.328) 0.022, vdrop. If you recall this is the same as the vdrop of LLC Level 1, but vdroop is an opposite story, there is virtually no voltage droop, instead we have voltage rise to (1.344-1.328) (+0.016), close to bios set amount. Not only does this vrise occur, it occurs without drop outs nor fluctuations. Another thing to note is that vdroop of Level 1 is Level 2’s vrise. This is a very powerful mode, Level 2 LLC is the LLC available on other motherboards, vdrop is not totally eliminated but reduced, and vdroop is reversed. Although LEVEL 1 seems to be much like standard no LLC, it is a much better alternative than LEVEL 2, for 2 reasons. First it offers 50% less vdroop which is what LLC should be for, has no vrise, and shouldn’t have the dangerous voltage spikes LEVEL 2 most defiantly has. Even though vdrop is present, it’s not a big issue, because we know how much vdrop is you can set voltage accordingly. Standard is like usual Intel vcore fluctuation, standard and safe. The spikes we see in Standard and Level1 does not follow a pattern other than it occurs throughout load, this may be due to the program, but it does not occur in level 2, so it might also be a function of normal voltage operation, either way it shouldn’t be a factor in causing overclocking voltage instability. In general you should look at the results and pick a level that fits your overclock, but know that level 2 is dangerous, level 1 is slightly less dangerous, and standard is safe. But overclocking in itself is also a dangerous risk.
