Lower your load temps: minimize your vcore

graysky · Post by **graysky** » Tue Jul 17, 2007 1:20 pm

Sounds pretty simple, but you'd be surprised how many people don't know about how much vcore can affect your load temps. Temp and frequency (FSB) have a linear relationship whereas temp and voltage have an exponential relationship. Conclusion: minimize voltage.

Here is the Intel document that helps explain it, see page 31:

An increase in processor operating frequency not only increases system performance, but also increases the processor power dissipation. The relationship between frequency and power is generalized in the following equation: P = CFV^2 (where P = power, C = capacitance, V = voltage, F = frequency). From this equation, it is evident that power increases linearly with frequency and with the square of voltage.

Same thing holds true for speed in a car: energy = 0.5mv^2 where m is mass and v is velocity. This is the basis of the old expression, "speed kills." You generate way more energy driving 75 MPH than you do driving 55 MPH since energy and velocity have an exponential relationship.

Anyway, to test how low you can go, simply manually set your vcore for something low. I started @ 1.2375 for my Q6600 running @ 9x333. If you can boot into windows load up a couple instances of orthos. If you have a quad make sure you set the CPU affinity such that one of the orthos gets cores 0 and 1 and the other gets cores 2 and 3. Let em run for a while. If the vcore is too low, one or both will give an error message. Orthos checks e when for rounding errors that can occur when the system isn't stable due to vcore, or temp, etc. Using a vcore of 1.2375v for my system gave an error pretty quickly:

If you don't get an error after say, 30 min, lower the vcore in the BIOS and repeat until you do get an error, then start working your way up until you can run them with no errors for a good 6-8 hours. In a nutshell, that's it.

Here is a detailed analysis of two difference vcore settings and the temps they produce on a Q6600 @ 9x266=2.4 GHz as well as @ 9x333=3.0 GHz. The two voltages I used were 1.112 V and 1.232 V (both of these are the load voltage, the actual BIOS settings were 1.1375V and 1.2625V respectively). 2x orthos ran for 30 minutes and the temperatures were averaged over the last 10 minutes of those runs (well after they stabilized). Room temps was 75-76 Â°F. Notice that the difference in voltage is ONLY 0.120 V or 120 mV, but this seemingly small difference brought the load temps up by an average of 6-7 Â°C per core!

Code: Select all

Run1 (9x266 @ 1.112 V), Average temps (Â°C): 51,52,50,50
Run2 (9x266 @ 1.232 V), Average temps (Â°C): 57,58,57,57
Differences (Â°C): +6, +6, +7, +7

Now if I add a faster FSB, they increased further:

Code: Select all

Run3 (9x333 @ 1.232 V), Average temps (Â°C): 61,61,60,60
Differences from lowest voltage (Â°C): +10, +9, +10, +10
Differences from same voltage (Â°C): +4, +3, +3, +3

Blacktales · Post by **Blacktales** » Tue Jul 17, 2007 2:53 pm

It is funny this post should appear just as i was writing another about the same matter, that is causing trouble on my computer.

I recently changed my Athlon 64 3200+ single core to a X2 3800+ on my s939 Asus A8V deluxe rev.1 motherboard. I knew my processor was supported by the bios and knew the cpu settings in the bios were set to automatic so i didn't bother looking at new settings for the new CPU.

But after reading recently this CPU is supposed to be run between 1.3 and 1.35v and realizing my vcore was hovering around 1.4, i tried to set the settings in the bios manually, that is keep everything standard but the vcore was lowered to 1.35. The system booted but after a few minutes into XP, the computer shuts down. I boot back up, restore the automatic or standard settings in the BIOS and then update the bios to the latest beta to try again with 1.35v. It kept up longer but after no more than 5mn the computer shuts down again. Since then, i restore everything to standard and had no problem whatsoever since.

Further informations :
- 4*512 RAM PC3200 Kingston ValueRAM
- Ninja + D12SL12 (though during the time the computer ran at 1.35 no difference in temps was seen)
- Antec TruePower II 430

Now the questions :
- Why can't the computer keep up at 1.35v ?
- If i have to keep this voltage, should i as well try to overclock a bit the CPU ? Correct me if i'm wrong but slightly increasing clock speed at the same voltage should not result in a significant increase in idle/load temps ?

Thanks IA

woodsman · Post by **woodsman** » Tue Jul 17, 2007 2:54 pm

No argument with the thrust or intent of the post, but something happened with your copy-and-paste of the original text. You posted P = CVF^2, but the voltage element (V) should be squared, not frequency (F).

A side note: the original paper ignores system inductance (L), to which power is linearly related but inversely. I suppose the original Intel author can be left off the hook for using the term "generalized," however.

More specifically, the association presented in the paper is not an equation. An equation contains all elements of a system, and must be mathematically correct such that the units balance on both sides of the equals sign. A mathematical equation allows moving any system element to either side of the equals sign pursuant to mathematical practices and the equation remains true.

A system identity, however, does not need to mathematically balance units, but only needs to include the essential elements to describe the system relationships.

For example, in the original paper pi is not included. Pi is unitless, but is an essential element of the mathematical relationships when frequency is involved. By excluding pi (and inductance) from the discussion, the "equation" actually is a system identity, the purpose of which is only to demonstrate system relationships.

Another side note: with many system identities, system elements cannot be moved to either side of the equals sign. For example, consider basic chemistry "equations." Moving any element from one side to the other produces an entirely different system. This effect is also true when using system identities to describe non-mathematical relationships. One way to avoid problems with system identities is to use an arrow rather than an equals sign. The arrow represents a one-way relationship, whereas an equals sign represents a two-way relationship.

Okay, enough pontificating for one day!

graysky · Post by **graysky** » Tue Jul 17, 2007 3:54 pm

@woodsman: good catch w/ regards to the error (I just corrected it).

cmthomson · Post by **cmthomson** » Tue Jul 17, 2007 5:15 pm

graysky wrote:whereas temp and voltage have an exponential relationship

Another nitpick: the power consumption (not the temperature) varies with the square of the voltage, which is polynomial, not exponential.

But the main point is valid: reducing CPU voltage has more of an effect than nearly anything else on how hot the CPU gets.

cmthomson · Post by **cmthomson** » Tue Jul 17, 2007 5:20 pm

Blacktales wrote:- Why can't the computer keep up at 1.35v ?

If you've set the voltage in the BIOS to 1.35, the actual voltage at the CPU is lower, due to what's called Vdroop. Some motherboards automatically adjust the VRM (voltage regulator module, a 12V to 1.xV DC-DC converter) based on the current the CPU draws, and others don't. Some do it well, and others don't. It is not at all uncommon to have a 0.2V difference between the BIOS setting and the actual voltage under load.

Programs such as SpeedFan and Everest provide sampled voltage values (for most motherboards). Compare these to your BIOS settings: I'll bet they are lower when the CPU is loaded.

Blacktales · Post by **Blacktales** » Tue Jul 17, 2007 5:50 pm

Thanks cmthomson for your answer.

I didn't know about the vdrop. So if i understand correctly, when i set the bios to 1.35, the voltage provided to the CPU (1.35) is enough to boot windows but when the CPU is under load, the vdrop causes it to receive less voltage (less than 1.30) and thus reboots ? I would have expected a good (for the time) motherboard to automatically regulate such things to ensure the required power goes to the CPU. As for the voltage being lower under load that is totally correct after testing as checked with speedfan.

So is there anything i can try when lowering the vcore except testing at 1.375 ? Would the CPU be fed 1.375 when idling and hopefully underload still be higher than the required 1.3 and keep on going ? Or is there something i didn't understand ?

Or is there any other setting in the BIOS i could play with to reduce the cpu temps ?

continuum · Post by **continuum** » Tue Jul 17, 2007 6:32 pm

Set it to 1.375v, during idle it should automatically lower the voltage.

Vdroop is fixable through other methods, but it's a lot more work. :p

cmthomson · Post by **cmthomson** » Wed Jul 18, 2007 5:40 pm

Some motherboards can be modified to reduce the Vdroop, typically with what is known as a "pencil mod" where you rub a resistor on the board with a soft-lead pencil to reduce its resistance and improve the tracking of the control circuit, but that is recommended only for true hobbyist/enthusiasts.

If you haven't disabled EIST, the BIOS will reduce the clock rate to 6x and the voltage to 1.05 when the system is idle. So as mentioned above, you should just boost your BIOS CPU voltage enough to keep your voltage under load high enough for stability.

RichA · Post by **RichA** » Wed Jul 18, 2007 6:51 pm

I don't understand why vdrop is designed to happen. I know that regulation isn't 100% accurate, but it can be incredibly accurate with minimal work, and I don't recall processors in years gone by suffering from such problems.

I dare say the answer lies in the datasheets, but I believe it is just a design 'feature' to keep the load on the processor, motherboard, and PSU, a little lower by widening the tolerances when under load. Reducing the supplied voltage means the PSU is worked less hard, the voltage regulators on the motherboard are worked less hard, the processor has less heat to dissipate, which will lower overall system temperatures.

Of course, it also means that regulation doesn't need to be so perfect, or critical. Nor does the quality of the PSU, nor anything else... so the end-product can be made more cheaply.

So most of the time, it's all good. Under consistent load, everything is worked less hard, which is a good thing. Joe Public has more money left in his pocket, less systems die from stress, etc. The only people who lose out are those of us who actually 'tweak' such things - where we look to squeeze as much performance from a system as we can - whether it be outright and overall power(overclocked), or those of us seeking to run as cool as possible whilst maintaining a consistent and reliable voltage(underclocked).

For the 1% of people who fall outside the scope of 'Joe Public', it causes problems. For the other 99%, it likely reduces problems, and overall component cost.

--Rich

continuum · Post by **continuum** » Thu Jul 19, 2007 9:20 am

RichA wrote:I don't understand why vdrop is designed to happen. I know that regulation isn't 100% accurate, but it can be incredibly accurate with minimal work, and I don't recall processors in years gone by suffering from such problems.

It's been happening since the Pentium III days at least probably longer. Processor power demands were much smaller then.

jojo4u · Post by **jojo4u** » Thu Jul 19, 2007 1:13 pm

Derating is a feature of intel processors and in fact quite clever. AMD systems shouldn't have this features. Please report if you AMD system droops.
Some english link: http://www.thetechrepository.com/showthread.php?t=126
Some points from a german article:
- silizium needs less voltage the higher the temperatures
- the higher the load, the higher the current, the higher the temperature, the lower the resistance -> even more current -> lowering vcore as countermeasure
- more stability during fast changes of current or less capacitors for same stability

Lower your load temps: minimize your vcore

Lower your load temps: minimize your vcore

Re: Lower your load temps: minimize your vcore